External Memory Interface Handbook Volume 3: Reference

External Memory Interface
Handbook Volume 3: Reference
Material
EMI_RM
2017.05.08
Last updated for Intel® Quartus® Prime Design Suite: 17.0
Subscribe
Send Feedback
Contents
Contents
1 Functional Description—UniPHY..................................................................................... 13
1.1
1.2
1.3
1.4
1.5
I/O Pads.............................................................................................................. 14
Reset and Clock Generation....................................................................................14
Dedicated Clock Networks...................................................................................... 15
Address and Command Datapath............................................................................ 16
Write Datapath..................................................................................................... 17
1.5.1 Leveling Circuitry...................................................................................... 17
1.6 Read Datapath..................................................................................................... 19
1.7 Sequencer........................................................................................................... 20
1.7.1 Nios II-Based Sequencer............................................................................21
1.7.2 RTL-based Sequencer................................................................................ 26
1.8 Shadow Registers................................................................................................. 27
1.8.1 Shadow Registers Operation....................................................................... 29
1.9 UniPHY Interfaces................................................................................................. 29
1.9.1 The DLL and PLL Sharing Interface.............................................................. 30
1.9.2 The OCT Sharing Interface......................................................................... 32
1.10 UniPHY Signals................................................................................................... 33
1.11 PHY-to-Controller Interfaces................................................................................. 37
1.12 Using a Custom Controller.................................................................................... 42
1.13 AFI 3.0 Specification............................................................................................43
1.13.1 Bus Width and AFI Ratio...........................................................................43
1.13.2 AFI Parameters....................................................................................... 44
1.13.3 AFI Signals.............................................................................................45
1.14 Register Maps.....................................................................................................49
1.14.1 UniPHY Register Map............................................................................... 50
1.14.2 Controller Register Map............................................................................52
1.15 Ping Pong PHY.................................................................................................... 52
1.15.1 Ping Pong PHY Feature Description............................................................ 52
1.15.2 Ping Pong PHY Architecture.......................................................................54
1.15.3 Ping Pong PHY Operation..........................................................................55
1.16 Efficiency Monitor and Protocol Checker................................................................. 55
1.16.1 Efficiency Monitor....................................................................................56
1.16.2 Protocol Checker..................................................................................... 56
1.16.3 Read Latency Counter..............................................................................56
1.16.4 Using the Efficiency Monitor and Protocol Checker........................................56
1.16.5 Avalon CSR Slave and JTAG Memory Map................................................... 57
1.17 UniPHY Calibration Stages.................................................................................... 58
1.17.1 Calibration Overview................................................................................58
1.17.2 Calibration Stages................................................................................... 59
1.17.3 Memory Initialization............................................................................... 60
1.17.4 Stage 1: Read Calibration Part One—DQS Enable Calibration and DQ/DQS
Centering................................................................................................60
1.17.5 Stage 2: Write Calibration Part One........................................................... 65
1.17.6 Stage 3: Write Calibration Part Two—DQ/DQS Centering...............................66
1.17.7 Stage 4: Read Calibration Part Two—Read Latency Minimization.................... 66
1.17.8 Calibration Signals...................................................................................67
1.17.9 Calibration Time......................................................................................67
External Memory Interface Handbook Volume 3: Reference Material
2
Contents
1.18 Document Revision History................................................................................... 67
2 Functional Description—Intel Stratix® 10 EMIF IP......................................................... 70
2.1
2.2
2.3
2.4
Stratix 10 Supported Memory Protocols................................................................... 70
Stratix 10 EMIF IP Support for 3DS/TSV DDR4 Devices.............................................. 71
Migrating to Stratix 10 from Previous Device Families................................................ 71
Stratix 10 EMIF Architecture: Introduction................................................................71
2.4.1 Stratix 10 EMIF Architecture: I/O Subsystem................................................ 72
2.4.2 Stratix 10 EMIF Architecture: I/O Column.....................................................73
2.4.3 Stratix 10 EMIF Architecture: I/O SSM......................................................... 74
2.4.4 Stratix 10 EMIF Architecture: I/O Bank........................................................ 74
2.4.5 Stratix 10 EMIF Architecture: I/O Lane.........................................................75
2.4.6 Stratix 10 EMIF Architecture: Input DQS Clock Tree....................................... 77
2.4.7 Stratix 10 EMIF Architecture: PHY Clock Tree................................................ 78
2.4.8 Stratix 10 EMIF Architecture: PLL Reference Clock Networks........................... 78
2.4.9 Stratix 10 EMIF Architecture: Clock Phase Alignment..................................... 79
2.5 Hardware Resource Sharing Among Multiple Stratix 10 EMIFs..................................... 80
2.5.1 I/O SSM Sharing.......................................................................................80
2.5.2 I/O Bank Sharing...................................................................................... 81
2.5.3 PLL Reference Clock Sharing.......................................................................82
2.5.4 Core Clock Network Sharing....................................................................... 83
2.6 Stratix 10 EMIF IP Component................................................................................ 84
2.6.1 Instantiating Your Stratix 10 EMIF IP in a Qsys Project................................... 84
2.6.2 File Sets.................................................................................................. 87
2.6.3 Customized readme.txt File........................................................................ 87
2.6.4 Clock Domains..........................................................................................88
2.6.5 ECC in Stratix 10 EMIF IP...........................................................................88
2.7 Examples of External Memory Interface Implementations for DDR4............................. 90
2.8 Stratix 10 EMIF Sequencer..................................................................................... 95
2.8.1 Stratix 10 EMIF DQS Tracking..................................................................... 96
2.9 Stratix 10 EMIF Calibration.....................................................................................96
2.9.1 Stratix 10 Calibration Stages ..................................................................... 97
2.9.2 Stratix 10 Calibration Stages Descriptions.................................................... 97
2.9.3 Stratix 10 Calibration Algorithms.................................................................98
2.9.4 Stratix 10 Calibration Flowchart................................................................ 101
2.10 Stratix 10 EMIF and SmartVID............................................................................ 102
2.11 Stratix 10 Hard Memory Controller Rate Conversion Feature....................................102
2.12 Differences Between User-Requested Reset in Stratix 10 versus Arria 10.................. 103
2.12.1 Method for Initiating a User-requested Reset.............................................104
2.13 Compiling Stratix 10 EMIF IP with the Quartus Prime Software................................ 105
2.13.1 Instantiating the Stratix 10 EMIF IP......................................................... 105
2.13.2 Setting I/O Assignments in Stratix 10 EMIF IP........................................... 106
2.14 Debugging Stratix 10 EMIF IP............................................................................. 106
2.14.1 External Memory Interface Debug Toolkit..................................................106
2.14.2 On-Chip Debug for Stratix 10.................................................................. 107
2.14.3 Configuring Your EMIF IP for Use with the Debug Toolkit............................. 107
2.14.4 Stratix 10 EMIF Debugging Examples....................................................... 108
2.15 Stratix 10 EMIF for Hard Processor Subsystem...................................................... 110
2.15.1 Restrictions on I/O Bank Usage for Stratix 10 EMIF IP with HPS................... 111
2.16 Stratix 10 EMIF Ping Pong PHY............................................................................ 114
2.16.1 Stratix 10 Ping Pong PHY Feature Description............................................ 114
External Memory Interface Handbook Volume 3: Reference Material
3
Contents
2.17
2.18
2.19
2.20
2.21
2.16.2 Stratix 10 Ping Pong PHY Architecture...................................................... 115
2.16.3 Stratix 10 Ping Pong PHY Limitations........................................................ 117
2.16.4 Stratix 10 Ping Pong PHY Calibration........................................................ 119
2.16.5 Using the Ping Pong PHY.........................................................................120
2.16.6 Ping Pong PHY Simulation Example Design................................................120
AFI 4.0 Specification.......................................................................................... 120
2.17.1 Bus Width and AFI Ratio.........................................................................121
2.17.2 AFI Parameters..................................................................................... 121
2.17.3 AFI Signals........................................................................................... 123
2.17.4 AFI 4.0 Timing Diagrams........................................................................ 128
Stratix 10 Resource Utilization.............................................................................142
2.18.1 QDR-IV Resource Utilization in Stratix 10 Devices...................................... 142
Stratix 10 EMIF Latency..................................................................................... 142
Integrating a Custom Controller with the Hard PHY................................................ 143
Document Revision History................................................................................. 143
3 Functional Description—Intel Arria 10 EMIF IP............................................................ 144
3.1
3.2
3.3
3.4
Supported Memory Protocols................................................................................ 145
Key Differences Compared to UniPHY IP and Previous Device Families........................ 145
Migrating from Previous Device Families................................................................. 146
Arria 10 EMIF Architecture: Introduction................................................................ 146
3.4.1 Arria 10 EMIF Architecture: I/O Subsystem.................................................147
3.4.2 Arria 10 EMIF Architecture: I/O Column..................................................... 148
3.4.3 Arria 10 EMIF Architecture: I/O AUX.......................................................... 149
3.4.4 Arria 10 EMIF Architecture: I/O Bank......................................................... 149
3.4.5 Arria 10 EMIF Architecture: I/O Lane......................................................... 153
3.4.6 Arria 10 EMIF Architecture: Input DQS Clock Tree........................................155
3.4.7 Arria 10 EMIF Architecture: PHY Clock Tree................................................. 156
3.4.8 Arria 10 EMIF Architecture: PLL Reference Clock Networks............................156
3.4.9 Arria 10 EMIF Architecture: Clock Phase Alignment...................................... 157
3.5 Hardware Resource Sharing Among Multiple EMIFs.................................................. 158
3.5.1 I/O Aux Sharing...................................................................................... 158
3.5.2 I/O Bank Sharing.................................................................................... 159
3.5.3 PLL Reference Clock Sharing..................................................................... 160
3.5.4 Core Clock Network Sharing..................................................................... 161
3.6 Arria 10 EMIF IP Component.................................................................................162
3.6.1 Instantiating Your Arria 10 EMIF IP in a Qsys Project....................................162
3.6.2 File Sets.................................................................................................165
3.6.3 Customized readme.txt File...................................................................... 165
3.6.4 Clock Domains........................................................................................166
3.6.5 ECC in Arria 10 EMIF IP............................................................................166
3.7 Examples of External Memory Interface Implementations for DDR4........................... 168
3.8 Arria 10 EMIF Sequencer......................................................................................173
3.8.1 Arria 10 EMIF DQS Tracking......................................................................174
3.9 Arria 10 EMIF Calibration......................................................................................174
3.9.1 Calibration Stages .................................................................................. 174
3.9.2 Calibration Stages Descriptions................................................................. 175
3.9.3 Calibration Algorithms..............................................................................176
3.9.4 Calibration Flowchart............................................................................... 179
3.9.5 Periodic OCT Recalibration........................................................................ 180
3.10 Back-to-Back User-Controlled Refresh Usage in Arria 10......................................... 182
External Memory Interface Handbook Volume 3: Reference Material
4
Contents
3.11
3.12
3.13
3.14
3.15
3.16
3.17
3.18
3.19
3.20
3.21
3.22
3.23
Arria 10 EMIF and SmartVID............................................................................... 183
Hard Memory Controller Rate Conversion Feature.................................................. 184
Back-to-Back User-Controlled Refresh for Hard Memory Controller........................... 184
Compiling Arria 10 EMIF IP with the Quartus Prime Software................................... 185
3.14.1 Instantiating the Arria 10 EMIF IP............................................................ 185
3.14.2 Setting I/O Assignments in Arria 10 EMIF IP..............................................185
Debugging Arria 10 EMIF IP................................................................................ 186
3.15.1 External Memory Interface Debug Toolkit..................................................186
3.15.2 On-Chip Debug for Arria 10.................................................................... 186
3.15.3 Configuring Your EMIF IP for Use with the Debug Toolkit............................. 186
3.15.4 Arria 10 EMIF Debugging Examples..........................................................187
Arria 10 EMIF for Hard Processor Subsystem.........................................................189
3.16.1 Restrictions on I/O Bank Usage for Arria 10 EMIF IP with HPS......................190
3.16.2 Using the EMIF Debug Toolkit with Arria 10 HPS Interfaces..........................192
Arria 10 EMIF Ping Pong PHY...............................................................................192
3.17.1 Ping Pong PHY Feature Description...........................................................193
3.17.2 Ping Pong PHY Architecture..................................................................... 194
3.17.3 Ping Pong PHY Limitations...................................................................... 196
3.17.4 Ping Pong PHY Calibration.......................................................................197
3.17.5 Using the Ping Pong PHY.........................................................................198
3.17.6 Ping Pong PHY Simulation Example Design................................................198
AFI 4.0 Specification.......................................................................................... 199
3.18.1 Bus Width and AFI Ratio.........................................................................199
3.18.2 AFI Parameters..................................................................................... 200
3.18.3 AFI Signals........................................................................................... 201
3.18.4 AFI 4.0 Timing Diagrams........................................................................ 206
Resource Utilization........................................................................................... 220
3.19.1 QDR-IV Resource Utilization in Arria 10 Devices......................................... 220
Arria 10 EMIF Latency........................................................................................ 220
Arria 10 EMIF Calibration Times...........................................................................221
Integrating a Custom Controller with the Hard PHY................................................ 222
Memory Mapped Register (MMR) Tables................................................................222
3.23.1 ctrlcfg0: Controller Configuration............................................................. 224
3.23.2 ctrlcfg1: Controller Configuration............................................................. 225
3.23.3 ctrlcfg2: Controller Configuration............................................................. 226
3.23.4 ctrlcfg3: Controller Configuration............................................................. 227
3.23.5 ctrlcfg4: Controller Configuration............................................................. 228
3.23.6 ctrlcfg5: Controller Configuration............................................................. 229
3.23.7 ctrlcfg6: Controller Configuration............................................................. 230
3.23.8 ctrlcfg7: Controller Configuration............................................................. 230
3.23.9 ctrlcfg8: Controller Configuration............................................................. 230
3.23.10 ctrlcfg9: Controller Configuration........................................................... 230
3.23.11 dramtiming0: Timing Parameters........................................................... 231
3.23.12 dramodt0: On-Die Termination Parameters..............................................231
3.23.13 dramodt1: On-Die Termination Parameters..............................................231
3.23.14 sbcfg0: Sideband Configuration............................................................. 232
3.23.15 sbcfg1: Sideband Configuration............................................................. 232
3.23.16 sbcfg2: Sideband Configuration............................................................. 232
3.23.17 sbcfg3: Sideband Configuration............................................................. 233
3.23.18 sbcfg4: Sideband Configuration............................................................. 233
3.23.19 sbcfg5: Sideband Configuration............................................................. 233
External Memory Interface Handbook Volume 3: Reference Material
5
Contents
3.23.20 sbcfg6: Sideband Configuration............................................................. 234
3.23.21 sbcfg7: Sideband Configuration............................................................. 234
3.23.22 sbcfg8: Sideband Configuration............................................................. 234
3.23.23 sbcfg9: Sideband Configuration............................................................. 234
3.23.24 caltiming0: Command/Address/Latency Parameters................................. 235
3.23.25 caltiming1: Command/Address/Latency Parameters................................. 235
3.23.26 caltiming2: Command/Address/Latency Parameters................................. 235
3.23.27 caltiming3: Command/Address/Latency Parameters................................. 236
3.23.28 caltiming4: Command/Address/Latency Parameters................................. 236
3.23.29 caltiming5: Command/Address/Latency Parameters................................. 236
3.23.30 caltiming6: Command/Address/Latency Parameters................................. 237
3.23.31 caltiming7: Command/Address/Latency Parameters................................. 237
3.23.32 caltiming8: Command/Address/Latency Parameters................................. 237
3.23.33 caltiming9: Command/Address/Latency Parameters................................. 238
3.23.34 caltiming10: Command/Address/Latency Parameters................................238
3.23.35 dramaddrw: Row/Column/Bank Address Width Configuration.....................238
3.23.36 sideband0: Sideband............................................................................239
3.23.37 sideband1: Sideband............................................................................239
3.23.38 sideband2: Sideband............................................................................239
3.23.39 sideband3: Sideband............................................................................239
3.23.40 sideband4: Sideband............................................................................239
3.23.41 sideband5: Sideband............................................................................240
3.23.42 sideband6: Sideband............................................................................240
3.23.43 sideband7: Sideband............................................................................240
3.23.44 sideband8: Sideband............................................................................240
3.23.45 sideband9: Sideband............................................................................240
3.23.46 sideband10: Sideband.......................................................................... 240
3.23.47 sideband11: Sideband.......................................................................... 241
3.23.48 sideband12: Sideband.......................................................................... 241
3.23.49 sideband13: Sideband.......................................................................... 241
3.23.50 sideband14: Sideband.......................................................................... 241
3.23.51 sideband15: Sideband.......................................................................... 242
3.23.52 dramsts: Calibration Status...................................................................242
3.23.53 ecc1: ECC General Configuration............................................................242
3.23.54 ecc2: Width Configuration.....................................................................243
3.23.55 ecc3: ECC Error and Interrupt Configuration............................................243
3.23.56 ecc4: Status and Error Information........................................................ 244
3.23.57 ecc5: Address of Most Recent SBE/DBE.................................................. 244
3.23.58 ecc6: Address of Most Recent Correct Command Dropped......................... 245
3.24 Document Revision History................................................................................. 245
4 Functional Description—Intel MAX® 10 EMIF IP........................................................... 247
4.1
4.2
4.3
4.4
4.5
MAX 10 EMIF Overview........................................................................................ 247
External Memory Protocol Support.........................................................................248
MAX 10 Memory Controller................................................................................... 248
MAX 10 Low Power Feature.................................................................................. 248
MAX 10 Memory PHY........................................................................................... 249
4.5.1 Supported Topologies...............................................................................249
4.5.2 Read Datapath........................................................................................249
4.5.3 Write Datapath....................................................................................... 250
4.5.4 Address and Command Datapath...............................................................252
External Memory Interface Handbook Volume 3: Reference Material
6
Contents
4.5.5 Sequencer..............................................................................................252
4.6 Calibration......................................................................................................... 255
4.6.1 Read Calibration......................................................................................255
4.6.2 Write Calibration..................................................................................... 255
4.7 Sequencer Debug Information.............................................................................. 256
4.8 Register Maps.....................................................................................................257
4.9 Document Revision History................................................................................... 257
6 Functional Description—Hard Memory Interface.......................................................... 258
6.1 Multi-Port Front End (MPFE)..................................................................................259
6.2 Multi-port Scheduling...........................................................................................260
6.2.1 Port Scheduling.......................................................................................260
6.2.2 DRAM Burst Scheduling............................................................................260
6.2.3 DRAM Power Saving Modes.......................................................................261
6.3 MPFE Signal Descriptions..................................................................................... 261
6.4 Hard Memory Controller....................................................................................... 264
6.4.1 Clocking.................................................................................................265
6.4.2 Reset.....................................................................................................265
6.4.3 DRAM Interface.......................................................................................265
6.4.4 ECC.......................................................................................................266
6.4.5 Bonding of Memory Controllers................................................................. 266
6.5 Hard PHY........................................................................................................... 268
6.5.1 Interconnections..................................................................................... 268
6.5.2 Clock Domains........................................................................................268
6.5.3 Hard Sequencer...................................................................................... 268
6.5.4 MPFE Setup Guidelines.............................................................................269
6.5.5 Soft Memory Interface to Hard Memory Interface Migration Guidelines........... 270
6.5.6 Bonding Interface Guidelines.................................................................... 271
6.6 Document Revision History................................................................................... 271
7 Functional Description—HPS Memory Controller.......................................................... 273
7.1
7.2
7.3
7.4
Features of the SDRAM Controller Subsystem......................................................... 273
SDRAM Controller Subsystem Block Diagram.......................................................... 274
SDRAM Controller Memory Options........................................................................ 275
SDRAM Controller Subsystem Interfaces................................................................ 276
7.4.1 MPU Subsystem Interface.........................................................................276
7.4.2 L3 Interconnect Interface......................................................................... 276
7.4.3 CSR Interface......................................................................................... 276
7.4.4 FPGA-to-HPS SDRAM Interface..................................................................276
7.5 Memory Controller Architecture.............................................................................277
7.5.1 Multi-Port Front End................................................................................ 278
7.5.2 Single-Port Controller.............................................................................. 279
7.6 Functional Description of the SDRAM Controller Subsystem.......................................281
7.6.1 MPFE Operation Ordering......................................................................... 281
7.6.2 MPFE Multi-Port Arbitration....................................................................... 281
7.6.3 MPFE SDRAM Burst Scheduling..................................................................285
7.6.4 Single-Port Controller Operation................................................................ 285
7.7 SDRAM Power Management.................................................................................. 295
7.8 DDR PHY............................................................................................................296
7.9 Clocks............................................................................................................... 296
7.10 Resets............................................................................................................. 296
External Memory Interface Handbook Volume 3: Reference Material
7
Contents
7.10.1 Taking the SDRAM Controller Subsystem Out of Reset ............................... 297
7.11 Port Mappings................................................................................................... 297
7.12 Initialization..................................................................................................... 298
7.12.1 FPGA-to-SDRAM Protocol Details..............................................................298
7.13 SDRAM Controller Subsystem Programming Model................................................. 302
7.13.1 HPS Memory Interface Architecture..........................................................302
7.13.2 HPS Memory Interface Configuration........................................................ 302
7.13.3 HPS Memory Interface Simulation............................................................303
7.13.4 Generating a Preloader Image for HPS with EMIF....................................... 304
7.14 Debugging HPS SDRAM in the Preloader............................................................... 305
7.14.1 Enabling UART or Semihosting Printout.....................................................305
7.14.2 Enabling Simple Memory Test..................................................................306
7.14.3 Enabling the Debug Report..................................................................... 307
7.14.4 Writing a Predefined Data Pattern to SDRAM in the Preloader...................... 310
7.15 SDRAM Controller Address Map and Register Definitions......................................... 311
7.15.1 SDRAM Controller Address Map............................................................... 311
7.16 Document Revision History................................................................................. 341
8 Functional Description—HPC II Controller....................................................................343
8.1 HPC II Memory Interface Architecture.................................................................... 343
8.2 HPC II Memory Controller Architecture................................................................... 344
8.2.1 Backpressure Support..............................................................................346
8.2.2 Command Generator............................................................................... 347
8.2.3 Timing Bank Pool.................................................................................... 347
8.2.4 Arbiter...................................................................................................347
8.2.5 Rank Timer............................................................................................ 348
8.2.6 Read Data Buffer and Write Data Buffer..................................................... 348
8.2.7 ECC Block.............................................................................................. 348
8.2.8 AFI and CSR Interfaces............................................................................ 348
8.3 HPC II Controller Features.................................................................................... 348
8.3.1 Data Reordering......................................................................................348
8.3.2 Pre-emptive Bank Management.................................................................349
8.3.3 Quasi-1T and Quasi-2T............................................................................ 349
8.3.4 User Autoprecharge Commands................................................................ 349
8.3.5 Address and Command Decoding Logic...................................................... 349
8.3.6 Low-Power Logic..................................................................................... 350
8.3.7 ODT Generation Logic.............................................................................. 350
8.3.8 Burst Merging......................................................................................... 353
8.3.9 ECC.......................................................................................................353
8.4 External Interfaces.............................................................................................. 356
8.4.1 Clock and Reset Interface.........................................................................356
8.4.2 Avalon-ST Data Slave Interface................................................................. 356
8.4.3 AXI Data Slave Interface.......................................................................... 356
8.4.4 Controller-PHY Interface...........................................................................362
8.4.5 Memory Side-Band Signals....................................................................... 362
8.4.6 Controller External Interfaces................................................................... 363
8.5 Top-Level Signals Description................................................................................364
8.5.1 Clock and Reset Signals........................................................................... 364
8.5.2 Local Interface Signals............................................................................. 365
8.5.3 Controller Interface Signals...................................................................... 369
8.5.4 CSR Interface Signals.............................................................................. 371
External Memory Interface Handbook Volume 3: Reference Material
8
Contents
8.5.5 Soft Controller Register Map..................................................................... 371
8.5.6 Hard Controller Register Map.................................................................... 375
8.6 Sequence of Operations....................................................................................... 379
8.7 Document Revision History................................................................................... 380
9 Functional Description—QDR II Controller................................................................... 382
9.1 Block Description................................................................................................ 382
9.1.1 Avalon-MM Slave Read and Write Interfaces................................................382
9.1.2 Command Issuing FSM.............................................................................383
9.1.3 AFI........................................................................................................383
9.2 Avalon-MM and Memory Data Width...................................................................... 383
9.3 Signal Descriptions..............................................................................................384
9.4 Document Revision History................................................................................... 385
10 Functional Description—QDR-IV Controller................................................................ 386
10.1 Block Description...............................................................................................386
10.1.1 Avalon-MM Slave Read and Write Interfaces.............................................. 386
10.1.2 AFI...................................................................................................... 387
10.2 Avalon-MM and Memory Data Width.....................................................................388
10.3 Signal Descriptions............................................................................................ 388
10.4 Document Revision History................................................................................. 389
11 Functional Description—RLDRAM II Controller........................................................... 390
11.1 Block Description...............................................................................................390
11.1.1 Avalon-MM Slave Interface..................................................................... 390
11.1.2 Write Data FIFO Buffer........................................................................... 391
11.1.3 Command Issuing FSM...........................................................................391
11.1.4 Refresh Timer....................................................................................... 391
11.1.5 Timer Module........................................................................................391
11.1.6 AFI...................................................................................................... 391
11.2 User-Controlled Features.................................................................................... 392
11.2.1 Error Detection Parity.............................................................................392
11.2.2 User-Controlled Refresh......................................................................... 392
11.3 Avalon-MM and Memory Data Width.....................................................................392
11.4 Signal Descriptions............................................................................................ 393
11.5 Document Revision History................................................................................. 394
12 Functional Description—RLDRAM 3 PHY-Only IP........................................................ 395
12.1
12.2
12.3
12.4
12.5
Block Description...............................................................................................395
Features...........................................................................................................395
RLDRAM 3 AFI Protocol...................................................................................... 396
RLDRAM 3 Controller with Arria 10 EMIF Interfaces................................................ 397
Document Revision History................................................................................. 398
13 Functional Description—Example Designs.................................................................. 399
13.1 Arria 10 EMIF IP Example Designs Quick Start Guide..............................................399
13.1.1 Typical Example Design Workflow............................................................ 399
13.1.2 Example Designs Interface Tab................................................................400
13.1.3 Development Kit Preset Workflow............................................................ 402
13.1.4 Compiling and Simulating the Design....................................................... 403
13.1.5 Compiling and Testing the Design in Hardware.......................................... 404
13.2 Testing the EMIF Interface Using the Traffic Generator 2.0...................................... 405
External Memory Interface Handbook Volume 3: Reference Material
9
Contents
13.2.1 Configurable Traffic Generator 2.0 Configuration Options.............................406
13.2.2 Performing Your Own Tests Using Traffic Generator 2.0............................... 411
13.2.3 Signal Splitter Component ..................................................................... 413
13.3 UniPHY-Based Example Designs...........................................................................413
13.3.1 Synthesis Example Design...................................................................... 413
13.3.2 Simulation Example Design.....................................................................415
13.3.3 Traffic Generator and BIST Engine........................................................... 416
13.3.4 Creating and Connecting the UniPHY Memory Interface and the Traffic
Generator in Qsys.................................................................................. 420
13.4 Document Revision History................................................................................. 422
14 Introduction to UniPHY IP......................................................................................... 424
14.1
14.2
14.3
14.4
14.5
14.6
14.7
Release Information...........................................................................................424
Device Support Levels........................................................................................425
Device Family and Protocol Support..................................................................... 425
UniPHY-Based External Memory Interface Features................................................ 426
System Requirements........................................................................................ 427
Intel FPGA IP Core Verification............................................................................ 427
Resource Utilization........................................................................................... 427
14.7.1 DDR2, DDR3, and LPDDR2 Resource Utilization in Arria V Devices................ 428
14.7.2 DDR2 and DDR3 Resource Utilization in Arria II GZ Devices.........................429
14.7.3 DDR2 and DDR3 Resource Utilization in Stratix III Devices..........................430
14.7.4 DDR2 and DDR3 Resource Utilization in Stratix IV Devices.......................... 431
14.7.5 DDR2 and DDR3 Resource Utilization in Arria V GZ and Stratix V Devices...... 433
14.7.6 QDR II and QDR II+ Resource Utilization in Arria V Devices.........................434
14.7.7 QDR II and QDR II+ Resource Utilization in Arria II GX Devices................... 435
14.7.8 QDR II and QDR II+ Resource Utilization in Arria II GZ, Arria V GZ,
Stratix III, Stratix IV, and Stratix V Devices............................................... 435
14.7.9 RLDRAM II Resource Utilization in Arria V Devices...................................... 435
14.7.10 RLDRAM II Resource Utilization in Arria II GZ, Arria V GZ, Stratix III,
Stratix IV, and Stratix V Devices...............................................................436
14.8 Document Revision History................................................................................. 436
15 Latency for UniPHY IP................................................................................................438
15.1
15.2
15.3
15.4
15.5
15.6
15.7
15.8
DDR2 SDRAM LATENCY...................................................................................... 438
DDR3 SDRAM LATENCY...................................................................................... 439
LPDDR2 SDRAM LATENCY................................................................................... 439
QDR II and QDR II+ SRAM Latency......................................................................440
RLDRAM II Latency............................................................................................440
RLDRAM 3 Latency............................................................................................ 441
Variable Controller Latency................................................................................. 441
Document Revision History................................................................................. 441
16 Timing Diagrams for UniPHY IP................................................................................. 443
16.1
16.2
16.3
16.4
16.5
16.6
16.7
DDR2 Timing Diagrams...................................................................................... 443
DDR3 Timing Diagrams...................................................................................... 448
QDR II and QDR II+ Timing Diagrams..................................................................456
RLDRAM II Timing Diagrams............................................................................... 460
LPDDR2 Timing Diagrams................................................................................... 467
RLDRAM 3 Timing Diagrams................................................................................474
Document Revision History................................................................................. 478
External Memory Interface Handbook Volume 3: Reference Material
10
Contents
17 External Memory Interface Debug Toolkit.................................................................. 480
17.1 User Interface...................................................................................................480
17.1.1 Communication..................................................................................... 480
17.1.2 Calibration and Report Generation........................................................... 481
17.2 Setup and Use.................................................................................................. 481
17.2.1 General Workflow.................................................................................. 482
17.2.2 Linking the Project to a Device................................................................ 482
17.2.3 Establishing Communication to Connections.............................................. 483
17.2.4 Selecting an Active Interface...................................................................483
17.2.5 Reports................................................................................................ 484
17.3 Operational Considerations................................................................................. 485
17.4 Troubleshooting................................................................................................ 487
17.5 Debug Report for Arria V and Cyclone V SoC Devices............................................. 487
17.5.1 Enabling the Debug Report for Arria V and Cyclone V SoC Devices............... 487
17.5.2 Determining the Failing Calibration Stage for a Cyclone V or Arria V HPS
SDRAM Controller...................................................................................487
17.6 On-Chip Debug Port for UniPHY-based EMIF IP...................................................... 488
17.6.1 Access Protocol..................................................................................... 489
17.6.2 Command Codes Reference.................................................................... 490
17.6.3 Header Files......................................................................................... 491
17.6.4 Generating IP With the Debug Port.......................................................... 491
17.6.5 Example C Code for Accessing Debug Data............................................... 492
17.7 On-Chip Debug Port for Arria 10 EMIF IP.............................................................. 493
17.7.1 Access Protocol..................................................................................... 494
17.7.2 EMIF On-Chip Debug Port....................................................................... 495
17.7.3 On-Die Termination Calibration ............................................................... 496
17.7.4 Eye Diagram ........................................................................................ 496
17.8 Driver Margining for Arria 10 EMIF IP................................................................... 496
17.8.1 Determining Margin............................................................................... 497
17.9 Read Setting and Apply Setting Commands for Arria 10 EMIF IP.............................. 497
17.9.1 Reading or Applying Calibration Settings...................................................497
17.10 Traffic Generator 2.0........................................................................................ 498
17.10.1 Configuring the Traffic Generator 2.0...................................................... 498
17.10.2 Running the Traffic Generator 2.0.......................................................... 500
17.10.3 Understanding the Custom Traffic Generator User Interface....................... 500
17.10.4 Applying the Traffic Generator 2.0.......................................................... 505
17.11 The Traffic Generator 2.0 Report........................................................................ 508
17.12 Example Tcl Script for Running the EMIF Debug Toolkit......................................... 509
17.13 Calibration Adjustment Delay Step Sizes for Arria 10 Devices................................ 509
17.13.1 Addressing..........................................................................................509
17.13.2 Output and Strobe Enable Minimum and Maximum Phase Settings............. 512
17.14 Using the EMIF Debug Toolkit with Arria 10 HPS Interfaces....................................513
17.15 Document Revision History............................................................................... 513
18 Upgrading to UniPHY-based Controllers from ALTMEMPHY-based Controllers............515
18.1
18.2
18.3
18.4
18.5
18.6
Generating Equivalent Design............................................................................. 516
Replacing the ALTMEMPHY Datapath with UniPHY Datapath..................................... 516
Resolving Port Name Differences......................................................................... 517
Creating OCT Signals......................................................................................... 518
Running Pin Assignments Script.......................................................................... 518
Removing Obsolete Files.....................................................................................519
External Memory Interface Handbook Volume 3: Reference Material
11
Contents
18.7 Simulating your Design...................................................................................... 519
18.8 Document Revision History................................................................................. 520
External Memory Interface Handbook Volume 3: Reference Material
12
1 Functional Description—UniPHY
1 Functional Description—UniPHY
UniPHY is the physical layer of the external memory interface.
The major functional units of the UniPHY layer include the following:
•
Reset and clock generation
•
Address and command datapath
•
Write datapath
•
Read datapath
•
Sequencer
The following figure shows the PHY block diagram.
Figure 1.
PHY Block Diagram
PHY - Memory
Domain
PHY - AFI
Domain
FPGA
UniPHY
Sequencer
Address
and
Command
Datapath
External
Memory
Device
I/O Pads
Write
Datapath
MUX
Memory
Controller
Read
Datapath
Reset
Generation
Related Links
•
UniPHY Interfaces on page 29
The following figure shows the major blocks of the UniPHY and how it
interfaces with the external memory device and the controller.
•
UniPHY Signals on page 33
The following tables list the UniPHY signals.
Intel Corporation. All rights reserved. Intel, the Intel logo, Altera, Arria, Cyclone, Enpirion, MAX, Nios, Quartus
and Stratix words and logos are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other
countries. Intel warrants performance of its FPGA and semiconductor products to current specifications in
accordance with Intel's standard warranty, but reserves the right to make changes to any products and services
at any time without notice. Intel assumes no responsibility or liability arising out of the application or use of any
information, product, or service described herein except as expressly agreed to in writing by Intel. Intel
customers are advised to obtain the latest version of device specifications before relying on any published
information and before placing orders for products or services.
*Other names and brands may be claimed as the property of others.
ISO
9001:2008
Registered
1 Functional Description—UniPHY
•
PHY-to-Controller Interfaces on page 37
Various modules connect to UniPHY through specific ports.
•
Using a Custom Controller on page 42
By default, the UniPHY-based external memory interface IP cores are delivered
with both the PHY and the memory controller integrated, as depicted in the
following figure.
•
AFI 4.0 Specification on page 120
The Altera PHY interface (AFI) 4.0 defines communication between the
controller and physical layer (PHY) in the external memory interface IP.
•
Register Maps on page 49
The following table lists the overall register mapping for the DDR2, DDR3, and
LPDDR2 SDRAM Controllers with UniPHY.
•
Ping Pong PHY on page 52
Ping Pong PHY is an implementation of UniPHY that allows two memory
interfaces to share address and command buses through time multiplexing.
•
Efficiency Monitor on page 56
The Efficiency Monitor reports read and write throughput on the controller
input, by counting command transfers and wait times, and making that
information available to the External Memory Interface Toolkit via an Avalon
slave port.
•
Calibration Stages on page 59
The calibration process begins when the PHY reset signal deasserts and the PLL
and DLL lock.
1.1 I/O Pads
The I/O pads contain all the I/O instantiations.
1.2 Reset and Clock Generation
At a high level, clocks in the PHY can be classified into two domains: the PHY-memory
domain and the PHY-AFI domain.
External Memory Interface Handbook Volume 3: Reference Material
14
1 Functional Description—UniPHY
The PHY-memory domain interfaces with the external memory device and always
operate at full-rate. The PHY-AFI domain interfaces with the memory controller and
can be a full-rate, half-rate, or quarter-rate clock, based on the controller in use.
The number of clock domains in a memory interface can vary depending on its
configuration; for example:
•
At the PHY-memory boundary, separate clocks may exist to generate the memory
clock signal, the output strobe, and to output write data, as well as address and
command signals. These clocks include pll_dq_write_clk, pll_write_clk,
pll_mem_clk, and pll_addr_cmd_clk. These clocks are phase-shifted as
required to achieve the desired timing relationships between memory clock,
address and command signals, output data, and output strobe.
•
For quarter-rate interfaces, additional clock domains such as pll_hr_clock are
required to convert signals between half-rate and quarter-rate.
•
For high-performance memory interfaces using Arria V, Cyclone V, or Stratix V
devices, additional clocks may be required to handle transfers between the device
core and the I/O periphery for timing closure. For core-to-periphery transfers, the
latch clock is pll_c2p_write_clock; for periphery-to-core transfers, it is
pll_p2c_read_clock. These clocks are automatically phase-adjusted for timing
closure during IP generation, but can be further adjusted in the parameter editor.
If the phases of these clocks are zero, the Fitter may remove these clocks during
optimization.Also, high-performance interfaces using a Nios II-based sequencer
require two additional clocks, pll_avl_clock for the Nios II processor, and
pll_config_clock for clocking the I/O scan chains during calibration.
For a complete list of clocks in your memory interface, compile your design and run
the Report Clocks command in the TimeQuest Timing Analyzer.
1.3 Dedicated Clock Networks
The UniPHY layer employs three types of dedicated clock networks:
•
Global clock network
•
Dual-regional clock network
•
PHY clock network (applicable to Arria V, Cyclone V, and Stratix V devices, and
later)
The PHY clock network is a dedicated high-speed, low-skew, balanced clock tree
designed for high-performance external memory interface. For device families that
support the PHY clock network, UniPHY always uses the PHY clock network for all
clocks at the PHY-memory boundary.
For families that do not support the PHY clock network, UniPHY uses either dualregional or global clock networks for clocks at the PHY-memory boundary. During
generation, the system selects dual-regional or global clocks automatically, depending
on whether a given interface spans more than one quadrant. UniPHY does not mix the
usage of dual-regional and global clock networks for clocks at the PHY-memory
boundary; this ensures that timing characteristics of the various output paths are as
similar as possible.
The <variation_name>_pin_assignments.tcl script creates the appropriate
clock network type assignment. The use of the PHY clock network is specified directly
in the RTL code, and does not require an assignment.
External Memory Interface Handbook Volume 3: Reference Material
15
1 Functional Description—UniPHY
The UniPHY uses an active-low, asychronous assert and synchronous de-assert reset
scheme. The global reset signal resets the PLL in the PHY and the rest of the system is
held in reset until after the PLL is locked.
1.4 Address and Command Datapath
The memory controller controls the read and write addresses and commands to meet
the memory specifications.
The PHY is indifferent to address or command—that is, it performs no decoding or
other operations—and the circuitry is the same for both. In full-rate and half-rate
interfaces, address and command is full rate, while in quarter-rate interfaces, address
and command is half rate.
Address and command signals are generated in the Altera PHY interface (AFI) clock
domain and sent to the memory device in the address and command clock domain.
The double-data rate input/output (DDIO) stage converts the half-rate signals into
full-rate signals, when the AFI clock runs at half-rate. For quarter-rate interfaces,
additional DDIO stages exist to convert the address and command signals in the
quarter-rate AFI clock domain to half-rate.
The address and command clock is offset with respect to the memory clock to balance
the nominal setup and hold margins at the memory device (center-alignment). In the
example below, this offset is 270 degrees. The Fitter can further optimize margins
based on the actual delays and clock skews. In half-rate and quarter-rate designs, the
full-rate cycle shifter blocks can perform a shift measured in full-rate cycles to
implement the correct write latency; without this logic, the controller would only be
able to implement even write latencies as it operates at half the speed. The full-rate
cycle shifter is clocked by either the AFI clock or the address and command clock,
depending on the PHY configuration, to maximize timing margins on the path from the
AFI clock to the address and command clock.
Figure 2.
Address and Command Datapath (Half-rate example shown)
Full-Rate
Cycle Shifter
Core
DDIO
clk
afi_clk
add_cmd_clk
270 Degrees
afi_clk
Address/Command
mem_clk
Center-aligned at
the memory device
H0/L0
add_cmd_clk
L0
mem_clk
External Memory Interface Handbook Volume 3: Reference Material
16
H0
1 Functional Description—UniPHY
1.5 Write Datapath
The write datapath passes write data from the memory controller to the I/O. The write
data valid signal from the memory controller generates the output enable signal to
control the output buffer. For memory protocols with a bidirectional data bus, it also
generates the dynamic termination control signal, which selects between series
(output mode) and parallel (input mode) termination.
The figure below illustrates a simplified write datapath of a typical half-rate interface.
The full-rate DQS write clock is sent to a DDIO_OUT cell. The output of DDIO_OUT
feeds an output buffer which creates a pair of pseudo differential clocks that connects
to the memory. In full-rate mode, only the SDR-DDR portion of the path is used; in
half-rate mode, the HDR-SDR circuitry is also required. The use of DDIO_OUT in both
the output strobe and output data generation path ensures that their timing
characteristics are as similar as possible. The
<variation_name>_pin_assignments.tcl script automatically specifies the logic
option that associates all data pins to the output strobe pin. The Fitter treats the pins
as a DQS/DQ pin group.
Figure 3.
Write Datapath
vcc
Output strobe
Output strobe (n)
ALTIOBUF
gnd
DQS write clock (full-rate)
DDIO_OUT
DDIO_OUT
0
DDIO_OUT
0
Output data [0]
DDIO_OUT
1
wdata[0]
wdata[1]
wdata[2]
wdata[3]
wdata[4n-1:0]
DDIO_OUT
2n-2
DDIO_OUT
n-1
Output data [n-1]
SDR DDR
DDIO_OUT
2n-1
HDR SDR
wdata[4n-4]
wdata[4n-3]
wdata[4n-2]
wdata[4n-1]
Half-rate clock
DQ write clock (full-rate, -90 degrees
from DQS write clock)
1.5.1 Leveling Circuitry
Leveling circuitry is dedicated I/O circuitry to provide calibration support for fly-by
address and command networks. For DDR3, leveling is always invoked, whether the
interface targets a DIMM or a single component. For DDR3 implementations at higher
frequencies, a fly-by topology is recommended for optimal performance. For DDR2,
leveling circuitry is invoked automatically for frequencies above 240 MHz; no leveling
is used for frequencies below 240 MHz.
External Memory Interface Handbook Volume 3: Reference Material
17
1 Functional Description—UniPHY
For DDR2 at frequencies below 240 MHz, you should use a tree-style layout. For
frequencies above 240 MHz, you can choose either a leveled or balanced-T or Y
topology, as the leveled PHY calibrates to either implementation. Regardless of
protocol, for devices without a levelling block—such as Arria II GZ, Arria V, and
Cyclone V—a balanced-T PCB topology for address/command/clock must be used
because fly-by topology is not supported.
For details about leveling delay chains, consult the memory interfaces hardware
section of the device handbook for your FPGA.
The following figure shows the write datapath for a leveling interface. The full-rate PLL
output clock phy_write_clk goes to a leveling delay chain block which generates all
other periphery clocks that are needed. The data signals that generate DQ and DQS
signals pass to an output phase alignment block. The output phase alignment block
feeds an output buffer which creates a pair of pseudo differential clocks that connect
to the memory. In full-rate designs, only the SDR-DDR portion of the path is used; in
half-rate mode, the HDR-SDR circuitry is also required. The use of DDIO_OUT in both
the output strobe and output data generation paths ensures that their timing
characteristics are as similar as possible. The
<variation_name>_pin_assignments.tcl script automatically specifies the logic
option that associates all data pins to the output strobe pin. The Quartus Prime Fitter
treats the pins as a DQS/DQ pin group.
External Memory Interface Handbook Volume 3: Reference Material
18
1 Functional Description—UniPHY
Figure 4.
Write Datapath for a Leveling Interface
Output Phase Alignment
SDR - DDR
DDIO_OUT_
DQS_0
DQS
DQSn
HDR - SDR
ALTIOBUF
DDIO_OUT_
DQS_1
afi_data_valid
0
0
DDIO_OUT
DQ[0]
DDIO_OUT
0
DDIO_OUT
0
DDIO_OUT
1
wdata[0]
wdata[1]
wdata[2]
wdata[3]
wdata[4n-1:0]
DDIO_OUT
2n-2
DQ[n-1]
DDIO_OUT
n-1
DDIO_OUT
2n-1
wdata[4n-4]
wdata[4n-3]
wdata[4n-2]
wdata[4n-1]
0_Phase
DQ clock
DQS clock
phy_write_clk
Leveling Delay Chain
1.6 Read Datapath
The read datapath passes read data from memory to the PHY. The following figure
shows the blocks and flow in the read datapath.
For all protocols, the DQS logic block delays the strobe by 90 degrees to center-align
the rising strobe edge within the data window. For DDR2, DDR3, and LPDDR2
protocols, the logic block also performs strobe gating, holding the DQS enable signal
high for the entire period that data is received. One DQS logic block exists for each
data group.
External Memory Interface Handbook Volume 3: Reference Material
19
1 Functional Description—UniPHY
One VFIFO buffer exists for each data group. For DDR2, DDR3, and LPDDR2 protocols,
the VFIFO buffer generates the DQS enable signal, which is delayed (by an amount
determined during calibration) to align with the incoming DQS signal. For QDR and
RLDRAM protocols, the output of the VFIFO buffer serves as the write enable signal for
the Read FIFO buffer, signaling when to begin capturing data.
DDIO_IN receives data from memory at double-data rate and passes data on to the
Read FIFO buffer at single-data rate.
The Read FIFO buffer temporarily holds data read from memory; one Read FIFO buffer
exists for each data group. For half-rate interfaces, the Read FIFO buffer converts the
full-rate, single data-rate input to a half-rate, single data-rate output which is then
passed to the PHY core logic. In the case of a quarter-rate interface, soft logic in the
PHY performs an additional conversion from half-rate single data rate to quarter-rate
single data rate.
One LFIFO buffer exists for each memory interface; the LFIFO buffer generates the
read enable signal for all Read FIFO blocks in an interface. The read enable signal is
asserted when the Read FIFO blocks have buffered sufficient data from the memory to
be read. The timing of the read enable signal is determined during calibration.
Figure 5.
Read Datapath
LFIFO
(one per interface)
Delayed
DQS/clock
Strobe
Memory
Data bus
DQS Logic
Block
DQS enable
(DDRx)
DDIO_IN
(Read
Capture)
Read
enable
Write
Half-rate
clk
clk
Read
FIFO
Data
Data
in
out
Write
enable
PHY
(QDR & RLDRAM)
VFIFO
(one per group)
Full-rate
double data rate
Full-rate
single data rate
Half-rate
single data rate
(or quarter-rate
single data rate)
1.7 Sequencer
Depending on the combination of protocol and IP architecture in your external
memory interface, you may have either an RTL-based sequencer or a Nios® II-based
sequencer.
External Memory Interface Handbook Volume 3: Reference Material
20
1 Functional Description—UniPHY
RTL-based sequencer implementations and Nios II-based sequencer implementations
can have different pin requirements. You may not be able to migrate from an RTLbased sequencer to a Nios II-based sequencer and maintain the same pinout.
For information on sequencer support for different protocol-architecture combinations,
refer to Introduction to Intel® FPGA Memory Solutions in Volume 1 of this handbook.
For information on pin planning, refer to Planning Pin and FPGA Resources in Volume 2
of this handbook.
Related Links
•
Protocol Support Matrix
•
Planning Pin and FPGA Resources
1.7.1 Nios II-Based Sequencer
The DDR2, DDR3, and LPDDR2 controllers with UniPHY employ a Nios II-based
sequencer that is parameterizable and is dynamically generated at run time. The
Nios II-based sequencer is also available with the QDR II and RLDRAM II controllers.
1.7.1.1 NIOS II-based Sequencer Function
The sequencer enables high-frequency memory interface operation by calibrating the
interface to compensate for variations in setup and hold requirements caused by
transmission delays.
UniPHY converts the double-data rate interface of high-speed memory devices to a
full-rate or half-rate interface for use within an FPGA. To compensate for slight
variations in data transmission to and from the memory device, double-data rate is
usually center-aligned with its strobe signal; nonetheless, at high speeds, slight
variations in delay can result in setup or hold time violations. The sequencer
implements a calibration algorithm to determine the combination of delay and phase
settings necessary to maintain center-alignment of data and clock signals, even in the
presence of significant delay variations. Programmable delay chains in the FPGA I/Os
then implement the calculated delays to ensure that data remains centered.
Calibration also applies settings to the FIFO buffers within the PHY to minimize latency
and ensures that the read valid signal is generated at the appropriate time.
When calibration is completed, the sequencer returns control to the memory
controller.
For more information about calibration, refer to UniPHY Calibration Stages, in this
chapter.
Related Links
UniPHY Calibration Stages on page 58
The DDR2, DDR3, and LPDDR2 SDRAM, QDR II and QDR II+ SRAM, and RLDRAM
II Controllers with UniPHY, and the RLDRAM 3 PHY-only IP, go through several
stages of calibration. Calibration information is useful in debugging calibration
failures.
External Memory Interface Handbook Volume 3: Reference Material
21
1 Functional Description—UniPHY
1.7.1.2 Nios II-based Sequencer Architecture
The sequencer is composed of a Nios II processor and a series of hardware-based
component managers, connected together by an Avalon bus. The Nios II processor
performs the high-level algorithmic operations of calibration, while the component
managers handle the lower-level timing, memory protocol, and bit-manipulation
operations.
The high-level calibration algorithms are specified in C code, which is compiled into
Nios II code that resides in the FPGA RAM blocks. The debug interface provides a
mechanism for interacting with the various managers and for tracking the progress of
the calibration algorithm, and can be useful for debugging problems that arise within
the PHY. The various managers are specified in RTL and implement operations that
would be slow or inefficient if implemented in software.
Figure 6.
NIOS II-based Sequencer Block Diagram
DQS enable
Samples
To Debug
Module
Tracking
Manager
Debug
Interface
RAM
Nios II
Processor
Avalon-MM Interface
SCC
Manager
RW
Manager
PHY
Manager
I/O
Scan Chain
AFI
Interface
PHY
Parameters
Data
Manager
The C code that defines the calibration routines is available for your reference in the
\<name>_s0_software subdirectory. Intel recommends that you do not modify this
C code.
External Memory Interface Handbook Volume 3: Reference Material
22
1 Functional Description—UniPHY
1.7.1.3 Nios II-based Sequencer SCC Manager
The scan chain control (SCC) manager allows the sequencer to set various delays and
phases on the I/Os that make up the memory interface. The latest Intel device
families provide dynamic delay chains on input, output, and output enable paths which
can be reconfigured at runtime. The SCC manager provides the calibration routines
access to these chains to add delay on incoming and outgoing signals. A master on
the Avalon-MM interface may require the maximum allowed delay setting on input and
output paths, and may set a particular delay value in this range to apply to the paths.
The SCC manager implements the Avalon-MM interface and the storage mechanism
for all input, output, and phase settings. It contains circuitry that configures a DQ- or
DQS-configuration block. The Nios II processor may set delay, phases, or register
settings; the sequencer scans the settings serially to the appropriate DQ or DQS
configuration block.
1.7.1.4 NIOS II-based Sequencer RW Manager
The read write (RW) manager encapsulates the protocol to read and write to the
memory device through the Altera PHY Interface (AFI). It provides a buffer that stores
the data to be sent to and read from memory, and provides the following commands:
•
Write configuration—configures the memory for use. Sets up burst lengths, read
and write latencies, and other device specific parameters.
•
Refresh—initiates a refresh operation at the DRAM. The command does not exist
on SRAM devices. The sequencer also provides a register that determines whether
the RW manager automatically generates refresh signals.
•
Enable or disable multi-purpose register (MPR)—for memory devices with a special
register that contains calibration specific patterns that you can read, this
command enables or disables access to the register.
•
Activate row—for memory devices that have both rows and columns, this
command activates a specific row. Subsequent reads and writes operate on this
specific row.
•
Precharge—closes a row before you can access a new row.
•
Write or read burst—writes or reads a burst length of data.
•
Write guaranteed—writes with a special mode where the memory holds address
and data lines constant. Intel guarantees this type of write to work in the presence
of skew, but constrains to write the same data across the entire burst length.
•
Write and read back-to-back—performs back-to-back writes or reads to adjacent
banks. Most memory devices have strict timing constraints on subsequent
accesses to the same bank, thus back-to-back writes and reads have to reference
different banks.
•
Protocol-specific initialization—a protocol-specific command required by the
initialization sequence.
1.7.1.5 NIOS II-based Sequencer PHY Manager
The PHY Manager provides access to the PHY for calibration, and passes relevant
calibration results to the PHY. For example, the PHY Manager sets the VFIFO and
LFIFO buffer parameters resulting from calibration, signals the PHY when the memory
initialization sequence finishes, and reports the pass/fail status of calibration.
External Memory Interface Handbook Volume 3: Reference Material
23
1 Functional Description—UniPHY
1.7.1.6 NIOS II-based Sequencer Data Manager
The Data Manager stores parameterization-specific data in RAM, for the software to
query.
1.7.1.7 NIOS II-based Sequencer Tracking Manager
The Tracking Manager detects the effects of voltage and temperature variations that
can occur on the memory device over time resulting in reduced margins, and adjusts
the DQS enable delay as necessary to maintain adequate operating margins.
The Tracking Manager briefly assumes control of the AFI interface after each memory
refresh cycle, issuing a read routine to the RW Manager, and then sampling the DQS
tracking. Ideally, the falling edge of the DQS enable signal would align to the last
rising edge of the raw DQS signal from the memory device. The Tracking Manager
determines whether the DQS enable signal is leading or trailing the raw DQS signal.
Each time a refresh occurs, the Tracking Manager takes a sample of the raw DQS
signal; any adjustments of the DQS enable signal occur only after sufficient samples of
raw DQS have been taken. When the Tracking Manager determines that the DQS
enable signal is either leading or lagging the raw DQS signal, it adjusts the DQS
enable appropriately.
The following figure shows the Tracking manager signals.
Figure 7.
Tracking Manager Signals
When the Refresh Completes, the
Controller Asserts the Signal & Waits
for the Tracking Manager’s Response
After afi_seq_busy Goes Low, the
Controller Deasserts the Signal &
Continues with Normal Operation
afi_clk
afi_ctl_refresh_done
afi_seq_busy
The Tracking Manager Responds by
Driving afi_seq_busy High & Can Begin
Taking Over the AFI Interface
External Memory Interface Handbook Volume 3: Reference Material
24
When the Tracking Manager Is Done
with DQS Tracking, It Asserts the
afi_seq_busy Signal
1 Functional Description—UniPHY
Some notes on Tracking Manager operation:
•
The time taken by the Tracking Manager is arbitrary; if the period taken exceeds
the refresh period, the Tracking Manager handles memory refresh.
•
afi_seq_busy should go high fewer than 10 clock cycles after
afi_ctl_refresh_done or afi_ctl_long_idle is asserted.
•
afi_refresh_done should deassert fewer than 10 clock cycles after
afi_seq_busy deasserts.
•
afi_ctl_long_idle causes the Tracking Manager to execute an algorithm
different than periodic refresh; use afi_ctl_long_idle when a long session
has elapsed without a periodic refresh.
•
Table 1.
The Tracking Manager is instantiated into the sequencer system when DQS
Tracking is turned on.
Configurations Supporting DQS Tracking
Device Family
Protocol
Memory Clock Frequency
Arria V (GX/GT/SX/ST) , Cyclone V
LPDDR2 (single rank)
All frequencies.
Arria V (GX/GT/SX/ST)
DDR3 (single rank)
450 MHz or higher for speed grade 5,
or higher than 534 MHz.
Arria V GZ, Stratix V (E/GS/GT/GX)
•
750 MHz or higher.
If you do not want to use DQS tracking, you can disable it (at your own risk), by
opening the Verilog file <variant_name>_if0_c0.v in an editor, and changing
the value of the USE_DQS_TRACKING parameter from 1 to 0.
1.7.1.8 NIOS II-based Sequencer Processor
The Nios II processor manages the calibration algorithm; the Nios II processor is
unavailable after calibration is completed.
The same calibration algorithm supports all device families, with some differences. The
following sections describe the calibration algorithm for DDR3 SDRAM on Stratix III
devices. Calibration algorithms for other protocols and families are a subset and
significant differences are pointed out when necessary. As the algorithm is fully
contained in the software of the sequencer (in the C code) enabling and disabling
specific steps involves turning flags on and off.
Calibration consists of the following stages:
•
Initialize memory.
•
Calibrate read datapath.
•
Calibrate write datapath.
•
Run diagnostics.
1.7.1.9 NIOS II-based Sequencer Calibration and Diagnostics
Calibration must initialize all memory devices before they can operate properly. The
sequencer performs this memory initialization stage when it takes control of the PHY
at startup.
External Memory Interface Handbook Volume 3: Reference Material
25
1 Functional Description—UniPHY
Calibrating the read datapath comprises the following steps:
•
Calibrate DQS enable cycle and phase.
•
Perform read per-bit deskew to center the strobe signal within data valid window.
•
Reduce LFIFO latency.
Calibrating the write datapath involves the following steps:
•
Center align DQS with respect to DQ.
•
Align DQS with mem_clk.
The sequencer estimates the read and write margins under noisy conditions, by
sweeping input and output DQ and DQS delays to determine the size of the data valid
windows on the input and output sides. The sequencer stores this diagnostic
information in the local memory and you can access it through the debugging
interface.
When the diagnostic test finishes, control of the PHY interface passes back to the
controller and the sequencer issues a pass or fail signal.
Related Links
External Memory Interface Debug Toolkit on page 480
The EMIF Toolkit lets you run your own traffic patterns, diagnose and debug
calibration problems, and produce margining reports for your external memory
interface.
1.7.2 RTL-based Sequencer
The RTL-based sequencer is available for QDR II and RLDRAM II interfaces, on
supported device families other than Arria V. The RTL sequencer is a state machine
that processes the calibration algorithm.
The sequencer assumes control of the interface at reset (whether at initial startup or
when the IP is reset) and maintains control throughout the calibration process. The
sequencer relinquishes control to the memory controller only after successful
calibration. The following tables list the major states in the RTL-based sequencer.
Table 2.
Sequencer States
RTL-based Sequencer State
Description
RESET
Remain in this state until reset is released.
LOAD_INIT
Load any initialization values for simulation purposes.
STABLE
Wait until the memory device is stable.
WRITE_ZERO
Issue write command to address 0.
WAIT_WRITE_ZERO
Write all 0xAs to address 0.
WRITE_ONE
Issue write command to address 1.
WAIT_WRITE_ONE
Write all 0x5s to address 1.
Valid Calibration States
V_READ_ZERO
Issue read command to address 0 (expected data is all 0xAs).
continued...
External Memory Interface Handbook Volume 3: Reference Material
26
1 Functional Description—UniPHY
RTL-based Sequencer State
Description
V_READ_NOP
This state represents the minimum number of cycles required between 2 back-toback read commands. The number of NOP states depends on the burst length.
V_READ_ONE
Issue read command to address 1 (expected data is all 0x5s).
V_WAIT_READ
Wait for read valid signal.
V_COMPARE_READ_ZERO_READ_ONE
Parameterizable number of cycles to wait before making the read data
comparisons.
V_CHECK_READ_FAIL
When a read fails, the write pointer (in the AFI clock domain) of the valid FIFO
buffer is incremented. The read pointer of the valid FIFO buffer is in the DQS clock
domain. The gap between the read and write pointers is effectively the latency
between the time when the PHY receives the read command and the time valid
data is returned to the PHY.
V_ADD_FULL_RATE
Advance the read valid FIFO buffer write pointer by an extra full rate cycle.
V_ADD_HALF_RATE
Advance the read valid FIFO buffer write pointer by an extra half rate cycle. In
full-rate designs, equivalent to V_ADD_FULL_RATE.
V_READ_FIFO_RESET
Reset the read and write pointers of the read data synchronization FIFO buffer.
V_CALIB_DONE
Valid calibration is successful.
Latency Calibration States
L_READ_ONE
Issue read command to address 1 (expected data is all 0x5s).
L_WAIT_READ
Wait for read valid signal from read datapath. Initial read latency is set to a
predefined maximum value.
L_COMPARE_READ_ONE
Check returned read data against expected data. If data is correct, go to
L_REDUCE_LATENCY; otherwise go to L_ADD_MARGIN.
L_REDUCE_LATENCY
Reduce the latency counter by 1.
L_READ_FLUSH
Read from address 0, to flush the contents of the read data resynchronization
FIFO buffer.
L_WAIT_READ_FLUSH
Wait until the whole FIFO buffer is flushed, then go back to L_READ and try again.
L_ADD_MARGIN
Increment latency counter by 3 (1 cycle to get the correct data, 2 more cycles of
margin for run time variations). If latency counter value is smaller than predefined
ideal condition minimum, then go to CALIB_FAIL.
CALIB_DONE
Calibration is successful.
CALIB_FAIL
Calibration is not successful.
1.8 Shadow Registers
Shadow registers are a hardware feature of Arria V GZ and Stratix V devices that
enables high-speed multi-rank calibration for DDR3 quarter-rate and half-rate memory
interfaces, up to 800MHz for dual-rank interfaces and 667MHz for quad-rank
interfaces.
Prior to the introduction of shadow registers, the data valid window of a multi-rank
interface was calibrated to the overlapping portion of the data valid windows of the
individual ranks. The resulting data valid window for the interface would be smaller
than the individual data valid windows, limiting overall performance.
External Memory Interface Handbook Volume 3: Reference Material
27
1 Functional Description—UniPHY
Figure 8.
Calibration of Overlapping Data Valid Windows, without Shadow Registers
Rank 0 Window
Rank 1 Window
Actual Window
Shadow registers allow the sequencer to calibrate each rank separately and fully, and
then to save the calibrated settings for each rank in its own set of shadow registers,
which are part of the IP scan chains. During a rank-to-rank switch, the rank-specific
set of calibration settings is restored just-in-time to optimize the data valid window for
each rank.
The following figure illustrates how the use of rank-specific calibration settings results
in a data valid window appropriate for the current rank.
Figure 9.
Rank-specific Calibration Settings, with Shadow Registers
Rank 0 Window
Rank 1 Window
Actual window when accessing Rank 0
Actual window when accessing Rank 1
The shadow registers and their associated rank-switching circuitry are part of the
device I/O periphery hardware.
External Memory Interface Handbook Volume 3: Reference Material
28
1 Functional Description—UniPHY
1.8.1 Shadow Registers Operation
The sequencer calibrates each rank individually and stores the resulting configuration
in shadow registers, which are part of the IP scan chains. UniPHY then selects the
appropriate configuration for the rank in use, switching between configurations as
necessary. Calibration results for deskew delay chains are stored in the shadow
registers. For DQS enable/disable, delay chain configurations come directly from the
FPGA core.
Signals
The afi_wrank signal indicates the rank to which the controller is writing, so that the
PHY can switch to the appropriate setting. Signal timing is identical to afi_dqs_burst;
that is, afi_wrank must be asserted at the same time as afi_dqs_burst, and must be of
the same duration.
The afi_rrank signal indicates the rank from which the controller is reading, so that the
PHY can switch to the appropriate setting. This signal must be asserted at the same
time as afi_rdata_en when issuing a read command, and once asserted, must remain
unchanged until the controller issues a new read command to another rank.
1.9 UniPHY Interfaces
The following figure shows the major blocks of the UniPHY and how it interfaces with
the external memory device and the controller.
Note:
Instantiating the delay-locked loop (DLL) and the phase-locked loop (PLL) on the same
level as the UniPHY eases DLL and PLL sharing.
Figure 10.
UniPHY Interfaces with the Controller and the External Memory
UniPHY Top-Level File
Memory Interface
RUP and RDN
AFI
Reset Interface
UniPHY
OCT
DLL
PLL
DLL Sharing PLL Sharing
Interface
Interface
The following interfaces are on the UniPHY top-level file:
•
AFI
•
Memory interface
•
DLL sharing interface
•
PLL sharing interface
•
OCT interface
External Memory Interface Handbook Volume 3: Reference Material
29
1 Functional Description—UniPHY
AFI
The UniPHY datapath uses the Altera PHY interface (AFI). The AFI is in a simple
connection between the PHY and controller. The AFI is based on the DDR PHY interface
(DFI) specification, with some calibration-related signals not used and some additional
Intel-specific sideband signals added.
For more information about the AFI, refer to AFI 4.0 Specification, in this chapter.
The Memory Interface
For information on the memory interface, refer to UniPHY Signals, in this chapter.
Related Links
•
AFI 4.0 Specification on page 120
The Altera PHY interface (AFI) 4.0 defines communication between the
controller and physical layer (PHY) in the external memory interface IP.
•
UniPHY Signals on page 33
The following tables list the UniPHY signals.
1.9.1 The DLL and PLL Sharing Interface
You can generate the UniPHY memory interface and configure it to share its PLL, DLL,
or both interfaces.
By default, a UniPHY memory interface variant contains a PLL and DLL; the PLL
produces a variety of required clock signals derived from the reference clock, and the
DLL produces a delay codeword. In this case the PLL sharing mode is "No sharing". A
UniPHY variant can be configured as a PLL Master and/or DLL Master, in which case
the corresponding interfaces are exported to the UniPHY top-level and can be
connected to an identically configured UniPHY variant PLL Slave and/or DLL Slave. The
UniPHY slave variant is instantiated without a PLL and/or DLL, which saves device
resources.
Note:
For Arria II GX, Arria II GZ, Stratix III, and Stratix IV devices, the PLL and DLL must
both be shared at the same time—their sharing modes must match. This restriction
does not apply to Arria V, Arria V GZ, Cyclone V, or Stratix V devices.
Note:
For devices with hard memory interface components onboard, you cannot share PLL or
DLL resources between soft and hard interfaces.
1.9.1.1 Sharing PLLs or DLLs
To share PLLs or DLLs, follow these steps:
1. To create a PLL or DLL master, create a UniPHY memory interface IP core. To make
the PLL and/or DLL interface appear at the top-level in the core, on the PHY
Settings tab in the parameter editor, set the PLL Sharing Mode and/or DLL
Sharing Mode to Master.
2. To create a PLL or DLL slave, create a second UniPHY memory interface IP core. To
make the PLL and/or DLL interface appear at the top-level in the core, on the PHY
Settings tab set the PLL Sharing Mode and/or DLL Sharing Mode to Slave.
3.
Connect the PLL and/or DLL sharing interfaces by following the appropriate step,
below:
External Memory Interface Handbook Volume 3: Reference Material
30
1 Functional Description—UniPHY
•
For cores generated with IP Catalog : connect the PLL and/or DLL
interface ports between the master and slave cores in your wrapper RTL.
When using PLL sharing, connect the afi_clk, afi_half_clk, and
afi_reset_export_n outputs from the UniPHY PLL master to the afi_clk,
afi_half_clk, and afi_reset_in inputs on the UniPHY PLL slave.
•
For cores generated with Qsys , connect the PLL and/or DLL interface in
the Qsys GUI. When using PLL sharing, connect the afi_clk,
afi_half_clk, and afi_reset_export_n outputs from the UniPHY PLL
master to the afi_clk, afi_half_clk, and afi_reset_in inputs on the
UniPHY PLL slave.
Qsys supports only one-to-one conduit connections in the patch panel. To share a
PLL from a UniPHY PLL master with multiple slaves, you should replicate the
number of PLL sharing conduit interfaces in the Qsys patch panel by choosing
Number of PLL sharing interfaces in the parameter editor.
Note:
You may connect a slave UniPHY instance to the clocks from a user-defined PLL
instead of from a UniPHY master. The general procedure for doing so is as follows:
1. Make a template, by generating your IP with PLL Sharing Mode set to No
Sharing, and then compiling the example project to determine the frequency and
phases of the clock outputs from the PLL.
2.
Generate an external PLL using the IP Catalog flow, with the equivalent output
clocks.
3.
Generate your IP with PLL Sharing Mode set to Slave, and connect the external
PLL to the PLL sharing interface.
You must be very careful when connecting clock signals to the slave. Connecting to
clocks with frequency or phase different than what the core expects may result in
hardware failure.
Note:
The signal dll_pll_locked is an internal signal from the PLL to the DLL which
ensures that the DLL remains in reset mode until the PLL becomes locked. This signal
is not available for use by customer logic.
1.9.1.2 About PLL Simulation
PLL frequencies may differ between the synthesis and simulation file sets. In either
case the achieved PLL frequencies and phases are calculated and reported in real time
in the parameter editor.
For the simulation file set, clocks are specified in the RTL, not in units of frequency but
by the period in picoseconds, thus avoiding clock drift due to picosecond rounding
error.
For the synthesis file set, there are two mechanisms by which clock frequencies are
specified in the RTL, based on the target device family:
•
For Arria V, Arria V GZ, Cyclone V, and Stratix V, clock frequencies are specified in
MHz.
•
For Arria II GX, Arria II GZ, Stratix III, and Stratix IV, clock frequencies are
specified by integer multipliers and divisors. For these families, the real simulation
model—as opposed to the default abstract simulation model—also uses clock
frequencies specified by integer ratios.
External Memory Interface Handbook Volume 3: Reference Material
31
1 Functional Description—UniPHY
1.9.2 The OCT Sharing Interface
By default, the UniPHY IP generates the required OCT control block at the top-level
RTL file for the PHY.
If you want, you can instantiate this block elsewhere in your code and feed the
required termination control signals into the IP core by turning off Master for OCT
Control Block on the PHY Settings tab. If you turn off Master for OCT Control
Block, you must instantiate the OCT control block or use another UniPHY instance as
a master, and ensure that the parallel and series termination control bus signals are
connected to the PHY.
The following figures show the PHY architecture with and without Master for OCT
Control Block.
Figure 11.
PHY Architecture with Master for OCT Control Block
UniPHY Top-Level File
Memory Interface
UniPHY
OCT
RUP and RDN
AFI
Reset Interface
DLL
PLL
OCT
Sharing DLL Sharing PLL Sharing
Interface Interface Interface
Figure 12.
PHY Architecture without Master for OCT Control Block
UniPHY Top-Level File
Memory Interface
RUP and RDN
AFI
OCT
UniPHY
Series and Parallel
Termination Control
Buses
Reset Interface
DLL
DLL Sharing
Interface
PLL
PLL Sharing
Interface
1.9.2.1 Modifying the Pin Assignment Script for QDR II and RLDRAM II
If you generate a QDR II or RLDRAM II slave IP core, you must modify the pin
assignment script to allow the fitter to correctly resolve the OCT termination block
name in the OCT master core.
External Memory Interface Handbook Volume 3: Reference Material
32
1 Functional Description—UniPHY
To modify the pin assignment script for QDR II or RLDRAM II slaves, follow these
steps:
1.
In a text editor, open your system's Tcl pin assignments script file, as follows:
2.
•
For systems generated with the IP Catalog: Open the <IP core name>/
<slave core name>_p0_pin_assignments.tcl file.
•
For systems generated with Qsys: Open the <HDL Path>/<submodules>/
<slave core name>_p0_pin_assignments.tcl file.
Search for the following line:
set ::master_corename "_MASTER_CORE_"
3. Replace _MASTER_CORE_ with the instance name of the UniPHY master to which
the slave is connected. The instance name is determined from the pin assignments
file name, as follows:
•
For systems generated with Qsys, the instance name is the <master core
name> component of the pins assignments file name: <HDL path>/
<submodules>/<master core name>_p0_pin_assignments.tcl.
•
For systems generated with the IP Catalog, the instance name is the <master
core name> component of the pins assignments file name: <IP core
name>/<master core name>_p0_pin_assignments.tcl .
1.10 UniPHY Signals
The following tables list the UniPHY signals.
Table 3.
Clock and Reset Signals
Name
Direction
Width
Description
Input
1
PLL reference clock input.
Input
1
Active low global reset for PLL and all logic in the
PHY, which causes a complete reset of the whole
system. Minimum recommended pulse width is
100ns.
Input
1
Holding soft_reset_n low holds the PHY in a
reset state. However it does not reset the PLL,
which keeps running. It also holds the
afi_reset_n output low.
pll_ref_clk
global_reset_n
soft_reset_n
Table 4.
DDR2 and DDR3 SDRAM Interface Signals
Name
Direction
Width
Output
mem_ck, mem_ck_n
Memory clock.
MEM_CK_WIDTH
Output
mem_cke
Description
Clock enable.
MEM_CLK_EN_WIDTH
continued...
External Memory Interface Handbook Volume 3: Reference Material
33
1 Functional Description—UniPHY
Name
Direction
Width
Output
mem_cs_n
Description
Chip select..
MEM_CHIP_SELECT_WIDTH
Output
mem_cas_n
Column address strobe.
MEM_CONTROL_WIDTH
Output
mem_ras_n
Row address strobe.
MEM_CONTROL_WIDTH
Output
mem_we_n
Write enable.
MEM_CONTROL_WIDTH
Output
mem_a
Address.
MEM_ADDRESS_WIDTH
Output
mem_ba
Bank address.
MEM_BANK_ADDRESS_WIDTH
Bidirectional
mem_dqs, mem_dqs_n
Data strobe.
MEM_DQS_WIDTH
Bidirectional
mem_dq
Data.
MEM_DQ_WIDTH
Output
mem_dm
Data mask.
MEM_DM_WIDTH
Output
mem_odt
On-die termination.
MEM_ODT_WIDTH
Output
mem_reset_n (DDR3 only)
Reset
1
Output
mem_ac_parity (DDR3 only, RDIMM/
LRDIMM only)
MEM_CONTROL_WIDTH
Input
mem_err_out_n (DDR3 only, RDIMM/
LRDIMM only)
Table 5.
Address/command parity bit.
(Even parity, per the RDIMM spec,
JESD82-29A.)
Address/command parity error.
MEM_CONTROL_WIDTH
UniPHY Parameters
Parameter Name
AFI_RATIO
Description
AFI_RATIO is 1 in full-rate designs.
AFI_RATIO is 2 for half-rate designs.
AFI_RATIO is 4 for quarter-rate designs.
The number of DQS pins in the interface.
MEM_IF_DQS_WIDTH
The address width of the specified memory device.
MEM_ADDRESS_WIDTH
continued...
External Memory Interface Handbook Volume 3: Reference Material
34
1 Functional Description—UniPHY
Parameter Name
Description
The bank width of the specified memory device.
MEM_BANK_WIDTH
The chip select width of the specified memory device.
MEM_CHIP_SELECT_WIDTH
The control width of the specified memory device.
MEM_CONTROL_WIDTH
The DM width of the specified memory device.
MEM_DM_WIDTH
The DQ width of the specified memory device.
MEM_DQ_WIDTH
The READ DQS width of the specified memory device.
MEM_READ_DQS_WIDTH
The WRITE DQS width of the specified memory device.
MEM_WRITE_DQS_WIDTH
—
OCT_SERIES_TERM_CONTROL_WIDTH
—
OCT_PARALLEL_TERM_CONTROL_WIDTH
The AFI address width, derived from the corresponding memory interface width.
AFI_ADDRESS_WIDTH
The AFI bank width, derived from the corresponding memory interface width.
AFI_BANK_WIDTH
The AFI chip select width, derived from the corresponding memory interface width.
AFI_CHIP_SELECT_WIDTH
The AFI data mask width.
AFI_DATA_MASK_WIDTH
The AFI control width, derived from the corresponding memory interface width.
AFI_CONTROL_WIDTH
The AFI data width.
AFI_DATA_WIDTH
The AFI DQS width.
AFI_DQS_WIDTH
The DLL delay output control width.
DLL_DELAY_CTRL_WIDTH
A read datapath parameter for timing purposes.
NUM_SUBGROUP_PER_READ_DQS
continued...
External Memory Interface Handbook Volume 3: Reference Material
35
1 Functional Description—UniPHY
Parameter Name
Description
A read datapath parameter for timing purposes.
QVLD_EXTRA_FLOP_STAGES
A read datapath parameter; calibration fails when the timeout counter expires.
READ_VALID_TIMEOUT_WIDTH
A read datapath parameter; the write address width for half-rate clocks.
READ_VALID_FIFO_WRITE_ADDR_WIDTH
A read datapath parameter; the read address width for full-rate clocks.
READ_VALID_FIFO_READ_ADDR_WIDTH
A latency calibration parameter; the maximum latency count width.
MAX_LATENCY_COUNT_WIDTH
A latency calibration parameter; the maximum read latency.
MAX_READ_LATENCY
—
READ_FIFO_READ_ADDR_WIDTH
—
READ_FIFO_WRITE_ADDR_WIDTH
A write datapath parameter; the maximum write latency count width.
MAX_WRITE_LATENCY_COUNT_WIDTH
An initailization sequence.
INIT_COUNT_WIDTH
A memory-specific initialization parameter.
MRSC_COUNT_WIDTH
A memory-specific initialization parameter.
INIT_NOP_COUNT_WIDTH
A memory-specific initialization parameter.
MRS_CONFIGURATION
A memory-specific initialization parameter.
MRS_BURST_LENGTH
A memory-specific initialization parameter.
MRS_ADDRESS_MODE
A memory-specific initialization parameter.
MRS_DLL_RESET
A memory-specific initialization parameter.
MRS_IMP_MATCHING
A memory-specific initialization parameter.
MRS_ODT_EN
continued...
External Memory Interface Handbook Volume 3: Reference Material
36
1 Functional Description—UniPHY
Parameter Name
Description
A memory-specific initialization parameter.
MRS_BURST_LENGTH
A memory-specific initialization parameter.
MEM_T_WL
A memory-specific initialization parameter.
MEM_T_RL
The burst count width for the sequencer.
SEQ_BURST_COUNT_WIDTH
The width of a counter that the sequencer uses.
VCALIB_COUNT_WIDTH
—
DOUBLE_MEM_DQ_WIDTH
—
HALF_AFI_DATA_WIDTH
The width of the calibration status register.
CALIB_REG_WIDTH
The number of AFI resets to generate.
NUM_AFI_RESET
Note:
For information about the AFI signals, refer to AFI 4.0 Specification in this chapter.
Related Links
AFI 4.0 Specification on page 120
The Altera PHY interface (AFI) 4.0 defines communication between the controller
and physical layer (PHY) in the external memory interface IP.
1.11 PHY-to-Controller Interfaces
Various modules connect to UniPHY through specific ports.
The AFI standardizes and simplifies the interface between controller and PHY for all
Intel memory designs, thus allowing you to easily interchange your own controller
code with Intel's high-performance controllers. The AFI PHY interface includes an
administration block that configures the memory for calibration and performs
necessary accesses to mode registers that configure the memory as required.
For half-rate designs, the address and command signals in the UniPHY are asserted for
one mem_clk cycle (1T addressing), such that there are two input bits per address
and command pin in half-rate designs. If you require a more conservative 2T
addressing (where signals are asserted for two mem_clk cycles), drive both input bits
(of the address and command signal) identically in half-rate designs.
External Memory Interface Handbook Volume 3: Reference Material
37
1 Functional Description—UniPHY
The following figure shows the half-rate write operation.
Figure 13.
Half-Rate Write with Word-Aligned Data
afi_clk
afi_dqs_burst
afi_wdata_valid
afi_wdata
00
10
00
--
ba
11
00
11
00
dc
--
b
--
The following figure shows a full-rate write.
Figure 14.
Full-Rate Write
afi_clk
afi_dqs_burst
afi_wdata_valid
afi_wdata
--
a
For quarter-rate designs, the address and command signals in the UniPHY are
asserted for one mem_clk cycle (1T addressing), such that there are four input bits
per address and command pin in quarter-rate designs. If you require a more
conservative 2T addressing (where signals are asserted for two mem_clk cycles), drive
either the two lower input bits or the two upper input bits (of the address and
command signal) identically.
External Memory Interface Handbook Volume 3: Reference Material
38
1 Functional Description—UniPHY
After calibration is completed, the sequencer sends the write latency in number of
clock cycles to the controller.
The AFI has the following conventions:
•
With the AFI, high and low signals are combined in one signal, so for a single chip
select (afi_cs_n) interface, afi_cs_n[1:0] , location 0 appears on the
memory bus on one mem_clk cycle and location 1 on the next mem_clk cycle.
Note: This convention is maintained for all signals so for an 8 bit memory
interface, the write data (afi_wdata) signal is afi_wdata[31:0], where
the first data on the DQ pins is afi_wdata[7:0], then
afi_wdata[15:8], then afi_wdata[23:16], then afi_wdata[31:24].
•
Spaced reads and writes have the following definitions:
—
Spaced writes—write commands separated by a gap of one controller clock
(afi_clk) cycle.
—
Spaced reads—read commands separated by a gap of one controller clock
(afi_clk) cycle.
The following figures show writes and reads, where the IP core writes data to and
reads from the same address. In each example, afi_rdata and afi_wdata are
aligned with controller clock (afi_clk) cycles. All the data in the bit vector is valid at
once. These figures assume the following general points:
•
The burst length is four.
•
An 8-bit interface with one chip select.
•
The data for one controller clock (afi_clk) cycle represents data for two memory
clock (mem_clk) cycles (half-rate interface).
External Memory Interface Handbook Volume 3: Reference Material
39
1 Functional Description—UniPHY
Figure 15.
Word-Aligned Writes
Note 1
Note 2
Note 3
Note 4
afi_clk
3
afi_wlat
afi_ras_n
00
11
afi_cas_n
11
00
afi_we_n
11
00
afi_cs_n
11
afi_dqs_burst
afi_wdata_valid
01
11
00
01
10
00
11
11
10
11
11
00
11
afi_wdata
00000000
afi_addr
00000000
0020008
ACT
WR
03020100
07060504
0b0a0908
00
0f0e0d0c
Memory
Interface
mem_clk
command
(Note 5 )
mem_cs_n
mem_dqs
mem_dq
Notes to Figure:
1.
To show the even alignment of afi_cs_n, expand the signal (this convention
applies for all other signals).
2. The afi_dqs_burst must go high one memory clock cycle before
afi_wdata_valid. Compare with the word-unaligned case.
3.
The afi_wdata_valid is asserted afi_wlat + 1 controller clock (afi_clk) cycles
after chip select (afi_cs_n) is asserted. The afi_wlat indicates the required
write latency in the system. The value is determined during calibration and is
dependant upon the relative delays in the address and command path and the
write datapath in both the PHY and the external DDR SDRAM subsystem. The
controller must drive afi_cs_n and then wait afi_wlat (two in this example)
afi_clks before driving afi_wdata_valid.
4.
Observe the ordering of write data (afi_wdata). Compare this to data on the
mem_dq signal.
5. In all waveforms a command record is added that combines the memory pins
ras_n, cas_n and we_n into the current command that is issued. This command
is registered by the memory when chip select (mem_cs_n) is low. The important
commands in the presented waveforms are WR= write, ACT = activate.
External Memory Interface Handbook Volume 3: Reference Material
40
1 Functional Description—UniPHY
Figure 16.
Word-Aligned Reads
Note 1
Note 2
Note 3
Note 2
Note 4
afi_clk
afi_rlat
15
afi_ras_n
11
afi_cas_n
0
afi_we_n
00
afi_cs_n
11
afi_rdata_en
00
11
01
11
11
01
00
11
11
00
afi_rdata_valid
00
afi_rdata
FFFFFFFF
afi_ba
afi_addr
11
00
11
00
00
0000000
0020008
afi_dm
Memory
Interface
mem_clk
command
ACT
RD
mem_cs_n
mem_dqs
mem_dq
Notes to Figure:
1.
For AFI, afi_rdata_en is required to be asserted one memory clock cycle before
chip select (afi_cs_n) is asserted. In the half-rate afi_clk domain, this
requirement manifests as the controller driving 11 (as opposed to the 01) on
afi_rdata_en.
2.
AFI requires that afi_rdata_en is driven for the duration of the read. In this
example, it is driven to 11 for two half-rate afi_clks, which equates to driving
to 1, for the four memory clock cycles of this four-beat burst.
3. The afi_rdata_valid returns 15 (afi_rlat) controller clock (afi_clk) cycles
after afi_rdata_en is asserted. Returned is when the afi_rdata_valid signal
is observed at the output of a register within the controller. A controller can use
the afi_rlat value to determine when to register to returned data, but this is
unnecessary as the afi_rdata_valid is provided for the controller to use as an
enable when registering read data.
4. Observe the alignment of returned read data with respect to data on the bus.
External Memory Interface Handbook Volume 3: Reference Material
41
1 Functional Description—UniPHY
Related Links
•
AFI 4.0 Specification on page 120
The Altera PHY interface (AFI) 4.0 defines communication between the
controller and physical layer (PHY) in the external memory interface IP.
•
Timing Diagrams for UniPHY IP on page 443
The following topics contain timing diagrams for UniPHY-based external
memory interface IP for supported protocols.
1.12 Using a Custom Controller
By default, the UniPHY-based external memory interface IP cores are delivered with
both the PHY and the memory controller integrated, as depicted in the following
figure.
If you want to use your own custom controller with the UniPHY PHY, check the
Generate PHY only box on the PHY Settings tab of the parameter editor and
generate the IP. The resulting top-level IP consists of only the sequencer, UniPHY
datapath, and PLL/DLL — the shaded area in the figure below.
Figure 17.
Memory Controller with UniPHY
Controller with UniPHY
PHY Only
Sequencer
Managers
Controller
Front End
Avalon Controller
AFI
UniPHY
Datapath
Memory
Interface
Memory
Device
PLL/DLL
The AFI interface is exposed at the top-level of the generated IP core; you can
connect the AFI interface to your custom controller.
When you enable Generate PHY only, the generated example designs include the
memory controller appropriately instantiated to mediate read/write commands from
the traffic generator to the PHY-only IP.
For information on the AFI protocol, refer to the AFI 4.0 Specification, in this chapter.
For information on the example designs, refer to Chapter 9, Example Designs, in this
volume.
Related Links
•
AFI 4.0 Specification on page 120
External Memory Interface Handbook Volume 3: Reference Material
42
1 Functional Description—UniPHY
The Altera PHY interface (AFI) 4.0 defines communication between the
controller and physical layer (PHY) in the external memory interface IP.
•
Traffic Generator and BIST Engine on page 416
The traffic generator and built-in self test (BIST) engine for Avalon-MM
memory interfaces generates Avalon-MM traffic on an Avalon-MM master
interface.
1.13 AFI 3.0 Specification
The Altera PHY interface (AFI) 3.0 defines communication between the controller and
physical layer (PHY) in the external memory interface IP.
The AFI is a single-data-rate interface, meaning that data is transferred on the rising
edge of each clock cycle. Most memory interfaces, however, operate at double-datarate, transferring data on both the rising and falling edges of the clock signal. If the
AFI interface is to directly control a double-data-rate signal, two single-data-rate bits
must be transmitted on each clock cycle; the PHY then sends out one bit on the rising
edge of the clock and one bit on the falling edge.
The AFI convention is to send the low part of the data first and the high part second,
as shown in the following figure.
Figure 18.
Single Versus Double Data Rate Transfer
clock
Single-data-rate
Double-data-rate
A High , A Low
A Low
A High
B High , B Low
B Low
B High
1.13.1 Bus Width and AFI Ratio
In cases where the AFI clock frequency is one-half or one-quarter of the memory clock
frequency, the AFI data must be twice or four times as wide, respectively, as the
corresponding memory data. The ratio between AFI clock and memory clock
frequencies is referred to as the AFI ratio. (A half-rate AFI interface has an AFI ratio of
2, while a quarter-rate interface has an AFI ratio of 4.)
In general, the width of the AFI signal depends on the following three factors:
•
The size of the equivalent signal on the memory interface. For example, if
a[15:0] is a DDR3 address input and the AFI clock runs at the same speed as
the memory interface, the equivalent afi_addr bus will be 16-bits wide.
•
The data rate of the equivalent signal on the memory interface. For example, if
d[7:0] is a double-data-rate QDR II input data bus and the AFI clock runs at the
same speed as the memory interface, the equivalent afi_write_data bus will
be 16-bits wide.
•
The AFI ratio. For example, if cs_n is a single-bit DDR3 chip select input and the
AFI clock runs at half the speed of the memory interface, the equivalent
afi_cs_n bus will be 2-bits wide.
External Memory Interface Handbook Volume 3: Reference Material
43
1 Functional Description—UniPHY
The following formula summarizes the three factors described above:
AFI_width = memory_width * signal_rate * AFI_RATE_RATIO
Note:
The above formula is a general rule, but not all signals obey it. For definite signal-size
information, refer to the specific table.
1.13.2 AFI Parameters
The following tables list Altera PHY interface (AFI) parameters for AFI 4.0.
The parameters described in the following tables affect the width of AFI signal buses.
Parameters prefixed by MEM_IF_ refer to the signal size at the interface between the
PHY and memory device.
Table 6.
Ratio Parameters
Parameter Name
Description
AFI_RATE_RATIO
The ratio between the AFI clock frequency and the memory clock frequency.
For full-rate interfaces this value is 1, for half-rate interfaces the value is 2,
and for quarter-rate interfaces the value is 4.
DATA_RATE_RATIO
The number of data bits transmitted per clock cycle. For single-date rate
protocols this value is 1, and for double-data rate protocols this value is 2.
ADDR_RATE_RATIO
The number of address bits transmitted per clock cycle. For single-date rate
address protocols this value is 1, and for double-data rate address protocols
this value is 2.
Table 7.
Memory Interface Parameters
Parameter Name
Description
MEM_IF_ADDR_WIDTH
The width of the address bus on the memory device(s).
MEM_IF_BANKADDR_WIDTH
The width of the bank address bus on the interface to the memory device(s).
Typically, the log 2 of the number of banks.
MEM_IF_CS_WIDTH
The number of chip selects on the interface to the memory device(s).
MEM_IF_WRITE_DQS_WIDTH
The number of DQS (or write clock) signals on the write interface. For
example, the number of DQS groups.
MEM_IF_CLK_PAIR_COUNT
The number of CK/CK# pairs.
MEM_IF_DQ_WIDTH
The number of DQ signals on the interface to the memory device(s). For
single-ended interfaces such as QDR II, this value is the number of D or Q
signals.
MEM_IF_DM_WIDTH
The number of data mask pins on the interface to the memory device(s).
MEM_IF_READ_DQS_WIDTH
The number of DQS signals on the read interface. For example, the number of
DQS groups.
Table 8.
Derived AFI Parameters
Parameter Name
Derivation Equation
AFI_ADDR_WIDTH
MEM_IF_ADDR_WIDTH * AFI_RATE_RATIO * ADDR_RATE_RATIO
AFI_BANKADDR_WIDTH
MEM_IF_BANKADDR_WIDTH * AFI_RATE_RATIO * ADDR_RATE_RATIO
AFI_CONTROL_WIDTH
AFI_RATE_RATIO * ADDR_RATE_RATIO
continued...
External Memory Interface Handbook Volume 3: Reference Material
44
1 Functional Description—UniPHY
Parameter Name
Derivation Equation
AFI_CS_WIDTH
MEM_IF_CS_WIDTH * AFI_RATE_RATIO
AFI_DM_WIDTH
MEM_IF_DM_WIDTH * AFI_RATE_RATIO * DATA_RATE_RATIO
AFI_DQ_WIDTH
MEM_IF_DQ_WIDTH * AFI_RATE_RATIO * DATA_RATE_RATIO
AFI_WRITE_DQS_WIDTH
MEM_IF_WRITE_DQS_WIDTH * AFI_RATE_RATIO
AFI_LAT_WIDTH
6
AFI_RLAT_WIDTH
AFI_LAT_WIDTH
AFI_WLAT_WIDTH
AFI_LAT_WIDTH * MEM_IF_WRITE_DQS_WIDTH
AFI_CLK_PAIR_COUNT
MEM_IF_CLK_PAIR_COUNT
AFI_WRANK_WIDTH
Number of ranks * MEM_IF_WRITE_DQS_WIDTH *AFI_RATE_RATIO
AFI_RRANK_WIDTH
Number of ranks * MEM_IF_READ_DQS_WIDTH *AFI_RATE_RATIO
1.13.3 AFI Signals
The following tables list Altera PHY interface (AFI) signals grouped according to their
functions.
In each table, the Direction column denotes the direction of the signal relative to the
PHY. For example, a signal defined as an output passes out of the PHY to the
controller. The AFI specification does not include any bidirectional signals.
Not all signals are used for all protocols.
1.13.3.1 AFI Clock and Reset Signals
The AFI interface provides up to two clock signals and an asynchronous reset signal.
Table 9.
Clock and Reset Signals
Signal Name
Direction
Width
Description
afi_clk
Output
1
Clock with which all data exchanged on the AFI bus
is synchronized. In general, this clock is referred to
as full-rate, half-rate, or quarter-rate, depending on
the ratio between the frequency of this clock and
the frequency of the memory device clock.
afi_half_clk
Output
1
Clock signal that runs at half the speed of the
afi_clk. The controller uses this signal when the
half-rate bridge feature is in use. This signal is
optional.
afi_reset_n
Output
1
Asynchronous reset output signal. You must
synchronize this signal to the clock domain in which
you use it.
1.13.3.2 AFI Address and Command Signals
The address and command signals for AFI 3.0 encode read/write/configuration
commands to send to the memory device. The address and command signals are
single-data rate signals.
External Memory Interface Handbook Volume 3: Reference Material
45
1 Functional Description—UniPHY
Table 10.
Address and Command Signals
Signal Name
Direction
Width
Description
afi_ba
Input
AFI_BANKADDR_WIDTH
Bank address. (Not applicable for
LPDDR3.)
afi_cke
Input
AFI_CLK_EN_WIDTH
Clock enable.
afi_cs_n
Input
AFI_CS_WIDTH
Chip select signal. (The number of
chip selects may not match the
number of ranks; for example,
RDIMMs and LRDIMMs require a
minimum of 2 chip select signals
for both single-rank and dual-rank
configurations. Consult your
memory device data sheet for
information about chip select signal
width.) (Matches the number of
ranks for LPDDR3.)
afi_ras_n
Input
AFI_CONTROL_WIDTH
RAS# (for DDR2 and DDR3
memory devices.)
afi_we_n
Input
AFI_CONTROL_WIDTH
WE# (for DDR2, DDR3, and
RLDRAM II memory devices.)
afi_cas_n
Input
AFI_CONTROL_WIDTH
CAS# (for DDR2 and DDR3
memory devices.)
afi_ref_n
Input
AFI_CONTROL_WIDTH
REF# (for RLDRAM II memory
devices.)
afi_rst_n
Input
AFI_CONTROL_WIDTH
RESET# (for DDR3 and DDR4
memory devices.)
afi_odt
Input
AFI_CLK_EN_WIDTH
On-die termination signal for DDR2,
DDR3, and LPDDR3 memory
devices. (Do not confuse this
memory device signal with the
FPGA’s internal on-chip termination
signal.)
afi_mem_clk_disable
Input
AFI_CLK_PAIR_COUNT
When this signal is asserted,
mem_clk and mem_clk_n are
disabled. This signal is used in lowpower mode.
afi_wps_n
Output
AFI_CS_WIDTH
WPS (for QDR II/II+ memory
devices.)
afi_rps_n
Output
AFI_CS_WIDTH
RPS (for QDR II/II+ memory
devices.)
1.13.3.3 AFI Write Data Signals
Write Data Signals for AFI 3.0 control the data, data mask, and strobe signals passed
to the memory device during write operations.
Table 11.
Write Data Signals
Signal Name
afi_dqs_burst
Direction
Input
Width
AFI_WRITE_DQS_WIDTH
Description
Controls the enable on the strobe
(DQS) pins for DDR2, DDR3, and
LPDDR2 memory devices. When
this signal is asserted, mem_dqs
and mem_dqsn are driven.
continued...
External Memory Interface Handbook Volume 3: Reference Material
46
1 Functional Description—UniPHY
Signal Name
Direction
Width
Description
This signal must be asserted before
afi_ wdata_valid to implement the
write preamble, and must be driven
for the correct duration to generate
a correctly timed mem_dqs signal.
afi_wdata_valid
Input
AFI_WRITE_DQS_WIDTH
Write data valid signal. This signal
controls the output enable on the
data and data mask pins.
afi_wdata
Input
AFI_DQ_WIDTH
Write data signal to send to the
memory device at double-data
rate. This signal controls the PHY’s
mem_dq output.
afi_dm
Input
AFI_DM_WIDTH
Data mask. This signal controls the
PHY’s mem_dm signal for DDR2,
DDR3, LPDDR2 and RLDRAM II
memory devices.)
afi_bws_n
Input
AFI_DM_WIDTH
Data mask. This signal controls the
PHY’s mem_bws_n signal for
QDR II/II+ memory devices.
afi_wrank
Input
AFI_WRANK_WIDTH
Shadow register signal. Signal
indicating the rank to which the
controller is writing, so that the
PHY can switch to the appropriate
setting. Signal timing is identical to
afi_dqs_burst; that is, afi_ wrank
must be asserted at the same time
as afi_dqs_burst, and must be of
the same duration.
1.13.3.4 AFI Read Data Signals
Read Data Signals for AFI 3.0 control the data sent from the memory device during
read operations.
Table 12.
Read Data Signals
Signal Name
Direction
Width
Description
afi_rdata_en
Input
AFI_RATE_RATIO
Read data enable. Indicates that the
memory controller is currently performing
a read operation. This signal is held high
only for cycles of relevant data (read data
masking).If this signal is aligned to even
clock cycles, it is possible to use 1-bit
even in half-rate mode (i.e.,
AFI_RATE=2).
afi_rdata_en_full
Input
AFI_RATE_RATIO
Read data enable full. Indicates that the
memory controller is currently performing
a read operation. This signal is held high
for the entire read burst.If this signal is
aligned to even clock cycles, it is possible
to use 1-bit even in half-rate mode (i.e.,
AFI_RATE=2).
continued...
External Memory Interface Handbook Volume 3: Reference Material
47
1 Functional Description—UniPHY
Signal Name
Direction
Width
Description
afi_rdata
Output
AFI_DQ_WIDTH
Read data from the memory device. This
data is considered valid only when
afi_rdata_valid is asserted by the PHY.
afi_rdata_valid
Output
AFI_RATE_RATIO
Read data valid. When asserted, this
signal indicates that the afi_rdata bus is
valid. If this signal is aligned to even
clock cycles, it is possible to use 1-bit
even in half-rate mode (i.e.,
AFI_RATE=2).
afi_rrank
Input
AFI_RRANK_WIDTH
Shadow register signal. Signal indicating
the rank from which the controller is
reading, so that the PHY can switch to
the appropriate setting. Must be asserted
at the same time as afi_rdata_en when
issuing a read command, and once
asserted, must remain unchanged until
the controller issues a new read
command to another rank.
1.13.3.5 AFI Calibration Status Signals
The PHY instantiates a sequencer which calibrates the memory interface with the
memory device and some internal components such as read FIFOs and valid FIFOs.
The sequencer reports the results of the calibration process to the controller through
the Calibration Status Signals in the AFI interface.
Table 13.
Calibration Status Signals
Signal Name
Direction
Width
Description
afi_cal_success
Output
1
Asserted to indicate that calibration has
completed successfully.
afi_cal_fail
Output
1
Asserted to indicate that calibration has
failed.
afi_cal_req
Input
1
Effectively a synchronous reset for the
sequencer. When this signal is asserted,
the sequencer returns to the reset state;
when this signal is released, a new
calibration sequence begins.
afi_wlat
Output
AFI_WLAT_WIDTH
The required write latency in afi_clk
cycles, between address/command and
write data being issued at the PHY/
controller interface. The afi_wlat value
can be different for different groups; each
group’s write latency can range from 0 to
63. If write latency is the same for all
groups, only the lowest 6 bits are
required.
afi_rlat
Output
AFI_RLAT_WIDTH
The required read latency in afi_clk cycles
between address/command and read
data being returned to the PHY/controller
interface. Values can range from 0 to 63.
(1)
Note to Table:
1. The afi_rlat signal is not supported for PHY-only designs. Instead, you can sample the afi_rdata_valid signal to
determine when valid read data is available.
External Memory Interface Handbook Volume 3: Reference Material
48
1 Functional Description—UniPHY
1.13.3.6 AFI Tracking Management Signals
When tracking management is enabled, the sequencer can take control over the AFI
3.0 interface at given intervals, and issue commands to the memory device to track
the internal DQS Enable signal alignment to the DQS signal returning from the
memory device. The tracking management portion of the AFI 3.0 interface provides a
means for the sequencer and the controller to exchange handshake signals.
Table 14.
Tracking Management Signals
Signal Name
Direction
Width
Description
afi_ctl_refresh_done
Input
MEM_IF_CS_WIDTH
Handshaking signal from controller to
tracking manager, indicating that a
refresh has occurred and waiting for a
response.
afi_seq_busy
Output
MEM_IF_CS_WIDTH
Handshaking signal from sequencer to
controller, indicating when DQS tracking
is in progress.
afi_ctl_long_idle
Input
MEM_IF_CS_WIDTH
Handshaking signal from controller to
tracking manager, indicating that it has
exited low power state without a periodic
refresh, and waiting for response.
1.14 Register Maps
The following table lists the overall register mapping for the DDR2, DDR3, and LPDDR2
SDRAM Controllers with UniPHY.
Note:
Addresses shown in the table are 32-bit word addresses. If a byte-addressed master
such as a Nios II processor accesses the CSR, it is necessary to multiply the addresses
by four.
Table 15.
Register Map
Address
Description
UniPHY Register Map
0x001
Reserved.
0x004
UniPHY status register 0.
0x005
UniPHY status register 1.
0x006
UniPHY status register 2.
0x007
UniPHY memory initialization parameters register 0.
Controller Register Map
0x100
Reserved.
0x110
Controller status and configuration register.
0x120
Memory address size register 0.
0x121
Memory address size register 1.
0x122
Memory address size register 2.
0x123
Memory timing parameters register 0.
continued...
External Memory Interface Handbook Volume 3: Reference Material
49
1 Functional Description—UniPHY
Address
Description
0x124
Memory timing parameters register 1.
0x125
Memory timing parameters register 2.
0x126
Memory timing parameters register 3.
0x130
ECC control register.
0x131
ECC status register.
0x132
ECC error address register.
1.14.1 UniPHY Register Map
The UniPHY register map allows you to control the memory components’ mode
register settings. The following table lists the register map for UniPHY.
Note:
Addresses shown in the table are 32-bit word addresses. If a byte-addressed master
such as a Nios II processor accesses the CSR, it is necessary to multiply the addresses
by four.
Table 16.
UniPHY Register Map
Address
0x001
0x002
0x004
0x005
0x006
Bit
Name
Default
Access
Description
15:0
Reserved.
0
—
Reserved for future use.
31:16
Reserved.
0
—
Reserved for future use.
15:0
Reserved.
0
—
Reserved for future use.
31:16
Reserved.
0
—
Reserved for future use.
0
SOFT_RESET
—
Write only
23:1
Reserved.
0
—
24
AFI_CAL_SUCCESS
—
Read only
Reports the value of the UniPHY
afi_cal_success. Writing to this
bit has no effect.
25
AFI_CAL_FAIL
—
Read only
Reports the value of the UniPHY
afi_cal_fail. Writing to this bit
has no effect.
26
PLL_LOCKED
—
Read only
Reports the PLL lock status.
31:27
Reserved.
0
—
Reserved for future use.
7:0
Reserved.
0
—
Reserved for future use.
15:8
Reserved.
0
—
Reserved for future use.
23:16
Reserved.
0
—
Reserved for future use.
31:24
Reserved.
0
—
Reserved for future use.
7:0
INIT_FAILING_STAGE
—
Read only
Initiate a soft reset of the interface.
This bit is automatically deasserted
after reset.
Reserved for future use.
Initial failing error stage of
calibration. Only applicable if
AFI_CAL_FAIL=1.
0: None
1: Read Calibration - VFIFO
2: Write Calibration - Write Leveling
continued...
External Memory Interface Handbook Volume 3: Reference Material
50
1 Functional Description—UniPHY
Address
Bit
Name
Default
Access
Description
3: Read Calibration - LFIFO
Calibration
4: Write Calibration - Write Deskew
5: Unused
6: Refresh
7: Calibration Skipped
8: Calibration Aborted
9: Read Calibration - VFIFO After
Writes
15:8
INIT_FAILING_SUBSTA
GE
—
Read only
Initial failing error substage of
calibration. Only applicable if
AFI_CAL_FAIL=1.
If INIT_FAILING_STAGE = 1 or 9:
1: Read Calibration - Guaranteed
read failure
2: Read Calibration - No working
DQSen phase found
3: Read Calibration - Per-bit read
deskew failure
If INIT_FAILING_STAGE = 2:
1: Write Calibration - No first
working write leveling phase found
2: Write Calibration - No last working
write leveling phase found
3: Write Calibration - Write leveling
copy failure
If INIT_FAILING_STAGE = other,
substages do not apply.
23:16
INIT_FAILING_GROUP
—
Read only
Initial failing error group of
calibration. Only applicable if
AFI_CAL_FAIL=1.
Returns failing DQ pin instead of
failing group, if:
INIT_FAILING_STAGE=1 and
INIT_FAILING_SUBSTAGE=3.
Or
INIT_FAILING_STAGE=4 and
INIT_FAILING_SUBSTAGE=1.
31:24
Reserved.
0
—
0x007
31:0
DQS_DETECT
—
Read only
Identifies if DQS edges have been
identified for each of the groups.
Each bit corresponds to one DQS
group.
0x008(DD
R2)
1:0
RTT_NOM
—
Read only
Rtt (nominal) setting of the DDR2
Extended Mode Register used during
memory initialization.
31:2
Reserved.
0
—
2:0
RTT_NOM
—
Rtt (nominal) setting of the DDR3
MR1 mode register used during
memory initialization.
4:3
Reserved.
0
Reserved for future use.
6:5
ODS
—
Output driver impedence control
setting of the DDR3 MR1 mode
register used during memory
initialization.
0x008(DD
R3)
Reserved for future use.
Reserved for future use.
continued...
External Memory Interface Handbook Volume 3: Reference Material
51
1 Functional Description—UniPHY
Address
0x008(LP
DDR2)
Bit
Name
Default
Access
Description
8:7
Reserved.
0
Reserved for future use.
10:9
RTT_WR
—
Rtt (writes) setting of the DDR3 MR2
mode register used during memory
initialization.
31:11
Reserved.
0
Reserved for future use.
3:0
DS
Driver impedence control for MR3
during initialization.
31:4
Reserved.
Reserved for future use.
1.14.2 Controller Register Map
The controller register map allows you to control the memory controller settings.
Note:
Dynamic reconfiguration is not currently supported.
For information on the controller register map, refer to Controller Register Map, in the
Functional Description—HPC II chapter.
Related Links
•
Soft Controller Register Map on page 371
The soft controller register map allows you to control the soft memory
controller settings.
•
Hard Controller Register Map on page 375
The hard controller register map allows you to control the hard memory
controller settings.
1.15 Ping Pong PHY
Ping Pong PHY is an implementation of UniPHY that allows two memory interfaces to
share address and command buses through time multiplexing. Compared to having
two independent interfaces, Ping Pong PHY uses fewer pins and less logic, while
maintaining equivalent throughput.
The Ping Pong PHY supports only quarter-rate configurations of the DDR3 protocol on
Arria V GZ and Stratix V devices.
1.15.1 Ping Pong PHY Feature Description
In conventional UniPHY, the address and command buses of a DDR3 quarter-rate
interface use 2T time—meaning that they are issued for two full-rate clock cycles, as
illustrated below.
External Memory Interface Handbook Volume 3: Reference Material
52
1 Functional Description—UniPHY
Figure 19.
2T Command Timing
CK
CSn
Addr, ba
Extra Setup Time
2T Command Issued
Active Period
With the Ping Pong PHY, address and command signals from two independent
controllers are multiplexed onto shared buses by delaying one of the controller
outputs by one full-rate clock cycle. The result is 1T timing, with a new command
being issued on each full-rate clock cycle. The following figure shows address and
command timing for the Ping Pong PHY.
Figure 20.
1T Command Timing Use by Ping Pong PHY
CK
CSn[0]
CSn[1]
Addr, ba
Cmd
Dev1
Cmd
Dev0
External Memory Interface Handbook Volume 3: Reference Material
53
1 Functional Description—UniPHY
1.15.2 Ping Pong PHY Architecture
The following figure shows a top-level block diagram of the Ping Pong PHY.
Functionally, the IP looks like two independent memory interfaces. The two controller
blocks are referred to as right-hand side (RHS) and left-hand side (LHS), respectively.
A gasket block located between the controllers and the PHY merges the AFI signals.
The PHY is double data width and supports both memory devices. The sequencer is
the same as with regular UniPHY, and calibrates the entire double-width PHY.
Figure 21.
Ping Pong PHY Architecture
Ping Pong PHY
Driver 0
DQ, DQS, DM
Sequencer
x2n
CS, ODT, CKE
PHY
x2n
Ping Pong
Gasket
Driver 1
Dev 0
xn
CAS, RAS, WE, ADDR, BA
AFI
Multiplexer
LHS
Controller
xn
CAS, RAS, WE, ADDR, BA
RHS
Controller
CS, ODT, CKE
DQ, DQS, DM
Dev 1
xn
xn
1.15.2.1 Ping Pong Gasket
The gasket delays and remaps quarter-rate signals so that they are correctly timemultiplexed at the full-rate PHY output. The gasket also merges address and command
buses ahead of the PHY.
AFI interfaces at the input and output of the gasket provide compatibility with the PHY
and with memory controllers.
Figure 22.
Ping Pong PHY Gasket Architecture
2x AFI
AFI
Ping Pong Gasket
RHS
Controller
ADD/CMD, WDATA
RDATA
LHS
Controller
ADD/CMD, WDATA
RDATA
1T Delay
Reorder &
Merge
Reorder &
Split
ADD/CMD, WDATA
RDATA
To AFI Multiplexer & PHY
From AFI Multiplexer & PHY
The following table shows how the gasket processes key AFI signals.
External Memory Interface Handbook Volume 3: Reference Material
54
1 Functional Description—UniPHY
Table 17.
Key AFI Signals Processed by Ping Pong PHY Gasket
Signal
Direction(Width multiplier)
Description
Gasket Conversions
cas, ras, we, addr, ba
Controller (1x) to PHY (1x)
Address and command buses
shared between devices.
Delay RHS by 1T; merge.
cs, odt, cke
Controller (1x) to PHY (2x)
Chip select, on-die termination,
and clock enable, one per
device.
Delay RHS by 1T;
reorder, merge.
wdata, wdata_valid,
dqs_burst, dm
Controller (1x) to PHY (2x)
Write datapath signals, one per
device.
Delay RHS by 1T;
reorder, merge.
rdata_en_rd,
rdata_en_rd_full
Controller (1x) to PHY (2x)
Read datapath enable signals
indicating controller performing
a read operation, one per
device.
Delay RHS by 1T.
rdata_rdata_valid
PHY (2x) to Controller (1x)
Read data, one per device.
Reorder; split.
cal_fail, cal_success,
seq_busy, wlat, rlat
PHY (1x) to Controller (1x)
Calibration result, one per
device.
Pass through.
rst_n, mem_clk_disable,
ctl_refresh_done,
ctl_long_idle
Controller (1x) to PHY (1x)
Reset and DQS tracking signals,
one per PHY.
AND (&)
cal_req, init_req
Controller (1x) to PHY (1x)
Controller to sequencer
requests.
OR (|)
wrank, rrank
Controller (1x) to PHY (2x)
Shadow register support.
Delay RHS by 1T;
reorder; merge.
1.15.2.2 Ping Pong PHY Calibration
The sequencer treats the Ping Pong PHY as a regular interface of double the width. For
example, in the case of two x16 devices, the sequencer calibrates both devices
together as a x32 interface. The sequencer chip select signal fans out to both devices
so that they are treated as a single interface. The VFIFO calibration process is
unchanged. For LFIFO calibration, the LFIFO buffer is duplicated for each interface and
the worst-case read datapath delay of both interfaces is used.
1.15.3 Ping Pong PHY Operation
To use the Ping Pong PHY, proceed as described below.
1. Configure a single memory interface according to your requirements.
2. Select the Enable Ping Pong PHY option in the Advanced PHY Options section
of the PHY Settings tab in the DDR3 parameter editor.
The Quartus Prime software then replicates the interface, resulting in two memory
controllers and a shared PHY, with the gasket block inserted between the controllers
and PHY. The system makes the necessary modifications to top-level component
connections, as well as the PHY read and write datapaths, and the AFI mux, without
further input from you.
1.16 Efficiency Monitor and Protocol Checker
The Efficiency Monitor and Protocol Checker allows measurement of traffic efficiency
on the Avalon-MM bus between the controller and user logic, measures read latencies,
and checks the legality of Avalon commands passed from the master. The Efficiency
External Memory Interface Handbook Volume 3: Reference Material
55
1 Functional Description—UniPHY
Monitor and Protocol Checker is available with the DDR2, DDR3, and LPDDR2 SDRAM
controllers with UniPHY and the RLDRAM II Controller with UniPHY. The Efficiency
Monitor and Protocol Checker is is not available for QDR II and QDR II+ SRAM, or for
the MAX 10 device family, or for Arria V or Cyclone V designs using the Hard Memory
Controller.
1.16.1 Efficiency Monitor
The Efficiency Monitor reports read and write throughput on the controller input, by
counting command transfers and wait times, and making that information available to
the External Memory Interface Toolkit via an Avalon slave port. This information may
be useful to you when experimenting with advanced controller settings, such as
command look ahead depth and burst merging.
1.16.2 Protocol Checker
The Protocol Checker checks the legality of commands on the controller’s input
interface against the Intel Avalon interface specification, and sets a flag in a register
on an Avalon slave port if an illegal command is detected.
1.16.3 Read Latency Counter
The Read Latency Counter measures the minimum and maximum wait times for read
commands to be serviced on the Avalon bus. Each read command is time-stamped
and placed into a FIFO buffer upon arrival, and latency is determined by comparing
that timestamp to the current time when the first beat of the returned read data is
provided back to the master.
1.16.4 Using the Efficiency Monitor and Protocol Checker
To include the Efficiency Monitor and Protocol Checker when you generate your IP
core, proceed as described below.
1.
On the Diagnostics tab in the parameter editor, turn on Enable the Efficiency
Monitor and Protocol Checker on the Controller Avalon Interface.
2.
To see the results of the data compiled by the Efficiency Monitor and Protocol
Checker, use the External Memory Interface Toolkit.
For information on the External Memory Interface Toolkit, refer to External Memory
Interface Debug Toolkit, in section 2 of this volume. For information about the Avalon
interface, refer to Avalon Interface Specifications.
Related Links
•
Avalon Interface Specifications
•
External Memory Interface Debug Toolkit on page 480
The EMIF Toolkit lets you run your own traffic patterns, diagnose and debug
calibration problems, and produce margining reports for your external memory
interface.
External Memory Interface Handbook Volume 3: Reference Material
56
1 Functional Description—UniPHY
1.16.5 Avalon CSR Slave and JTAG Memory Map
The following table lists the memory map of registers inside the Efficiency Monitor and
Protocol Checker. This information is only of interest if you want to communicate
directly with the Efficiency Monitor and Protocol Checker without using the External
Memory Interface Toolkit. This CSR map is not part of the UniPHY CSR map.
Prior to reading the data in the CSR, you must issue a read command to address 0x01
to take a snapshot of the current data.
Table 18.
Avalon CSR Slave and JTAG Memory Map
Address
Bit
Name
Default
Access
Description
Used internally by EMIF Toolkit to
identify Efficiency Monitor type. This
address must be read prior to reading
the other CSR contents.
0x01
31:0
Reserved
0
Read Only
0x02
31:0
Reserved
0
—
0x08
0
Efficiency Monitor reset
—
Write only
7:1
Reserved
—
—
8
Protocol Checker reset
—
Write only
15:9
Reserved
—
—
16
Start/stop Efficiency
Monitor
—
Read/Write
23:17
Reserved
—
—
31:24
Efficiency Monitor status
—
Read Only
bit
bit
bit
bit
15:0
Efficiency Monitor
address width
—
Read Only
Address width of the Efficiency
Monitor.
31:16
Efficiency Monitor data
width
—
Read Only
Data Width of the Efficiency Monitor.
15:0
Efficiency Monitor byte
enable
—
Read Only
Byte enable width of the Efficiency
Monitor.
31:16
Efficiency Monitor burst
count width
—
Read Only
Burst count width of the Efficiency
Monitor.
0x14
31:0
Cycle counter
—
Read Only
Clock cycle counter for the Efficiency
Monitor. Lists the number of clock
cycles elapsed before the Efficiency
Monitor stopped.
0x18
31:0
Transfer counter
—
Read Only
Counts any read or write data transfer
cycle.
0x1C
31:0
Write counter
—
Read Only
Counts write requests, including those
during bursts.
0x20
31:0
Read counter
—
Read Only
Counts read requests.
0x24
31:0
Readtotal counter
—
Read Only
Counts read requests (total burst
requests).
0x10
0x11
Used internally by EMIF Toolkit to
identify Efficiency Monitor version.
Write a 0 to reset.
Reserved for future use.
Write a 0 to reset.
Reserved for future use.
Starting and stopping statistics
gathering.
Reserved for future use.
0:
1:
2:
3:
Efficiency Monitor stopped
Waiting for start of pattern
Running
Counter saturation
continued...
External Memory Interface Handbook Volume 3: Reference Material
57
1 Functional Description—UniPHY
Address
Bit
Name
Default
Access
Description
0x28
31:0
NTC waitrequest counter
—
Read Only
Counts Non Transfer Cycles (NTC) due
to slave wait request high.
0x2C
31:0
NTC noreaddatavalid
counter
—
Read Only
Counts Non Transfer Cycles (NTC) due
to slave not having read data.
0x30
31:0
NTC master write idle
counter
—
Read Only
Counts Non Transfer Cycles (NTC) due
to master not issuing command, or
pause in write burst.
0x34
31:0
NTC master idle counter
—
Read Only
Counts Non Transfer Cycles (NTC) due
to master not issuing command
anytime.
0x40
31:0
Read latency min
—
Read Only
The lowest read latency value.
0x44
31:0
Read latency max
—
Read Only
The highest read latency value.
0x48
31:0
Read latency total [31:0]
—
Read Only
The lower 32 bits of the total read
latency.
0x49
31:0
Read latency total
[63:32]
—
Read Only
The upper 32 bits of the total read
latency.
0x50
7:0
Illegal command
—
Read Only
Bits used to indicate which illegal
command has occurred. Each bit
represents a unique error.
31:8
Reserved
—
—
Reserved for future use.
1.17 UniPHY Calibration Stages
The DDR2, DDR3, and LPDDR2 SDRAM, QDR II and QDR II+ SRAM, and RLDRAM II
Controllers with UniPHY, and the RLDRAM 3 PHY-only IP, go through several stages of
calibration. Calibration information is useful in debugging calibration failures.
The section includes an overview of calibration, explanation of the calibration stages,
and a list of generated calibration signals. The information in this section applies only
to the Nios II-based sequencer used in the DDR2, DDR3, and LPDDR2 SDRAM
Controllers with UniPHY versions 10.0 and later, and, optionally, in the QDR II and
QDR II+ SRAM and RLDRAM II Controllers with UniPHY version 11.0 and later, and the
RLDRAM 3 PHY-only IP. The information in this section applies to the Arria II GZ,
Arria V, Arria V GZ, Cyclone V, Stratix III, Stratix IV, and Stratix V device families.
Note:
For QDR II and QDR II+ SRAM and RLDRAM II Controllers with UniPHY version 11.0
and later, you have the option to select either the RTL-based sequencer or the
Nios II-based sequencer. Generally, choose the RTL-based sequencer when area is the
major consideration, and choose the Nios II-based sequencer when performance is the
major consideration.
Note:
For RLDRAM 3, write leveling is not performed. The sequencer does not attempt to
optimize margin for the tCKDK timing requirement.
1.17.1 Calibration Overview
Calibration configures the memory interface (PHY and I/Os) so that data can pass
reliably to and from memory.
External Memory Interface Handbook Volume 3: Reference Material
58
1 Functional Description—UniPHY
The sequencer illustrated in the figure below calibrates the PHY and the I/Os. To
correctly transmit data between a memory device and the FPGA at high speed, the
data must be center-aligned with the data clock.
Calibration also determines the delay settings needed to center-align the various data
signals with respect to their clocks. I/O delay chains implement the required delays in
accordance with the computed alignments. The Nios II-based sequencer performs two
major tasks: FIFO buffer calibration and I/O calibration. FIFO buffer calibration adjusts
FIFO lengths and I/O calibration adjusts any delay chain and phase settings to centeralign data signals with respect to clock signals for both reads and writes. When the
calibration process completes, the sequencer shuts off and passes control to the
memory controller.
Sequencer in Memory Interface Logic
Memory Device
UniPHY
PHY
PLL
Controller
Sequencer
User Interface (Avalon-MM)
Figure 23.
1.17.2 Calibration Stages
The calibration process begins when the PHY reset signal deasserts and the PLL and
DLL lock.
The following stages of calibration take place:
1. Read calibration part one—DQS enable calibration (only for DDR2 and DDR3
SDRAM Controllers with UniPHY) and DQ/DQS centering
2. Write calibration part one—Leveling
3. Write calibration part two—DQ/DQS centering
4. Read calibration part two—Read latency minimization
Note:
For multirank calibration, the sequencer transmits every read and write command to
each rank in sequence. Each read and write test is successful only if all ranks pass the
test. The sequencer calibrates to the intersection of all ranks.
The calibration process assumes the following conditions; if either of these conditions
is not true, calibration likely fails in its early stages:
•
The address and command paths must be functional; calibration does not tune the
address and command paths. (The Quartus Prime software fully analyzes the
timing for the address and command paths, and the slack report is accurate,
assuming the correct board timing parameters.)
•
At least one bit per group must work before running per-bit-deskew calibration.
(This assumption requires that DQ-to-DQS skews be within the recommended
20 ps.)
External Memory Interface Handbook Volume 3: Reference Material
59
1 Functional Description—UniPHY
1.17.3 Memory Initialization
The memory is powered up according to protocol initialization specifications. All ranks
power up simultaneously. Once powered, the device is ready to receive mode register
load commands. This part of initialization occurs separately for each rank. The
sequencer issues mode register set commands on a per-chip-select basis and
initializes the memory to the user-specified settings.
1.17.4 Stage 1: Read Calibration Part One—DQS Enable Calibration and
DQ/DQS Centering
Read calibration occurs in two parts. Part one is DQS enable calibration with DQ/DQS
centering, which happens during stage 1 of the overall calibration process; part two is
read latency minimization, which happens during stage 4 of the overall calibration
process.
The objectives of DQS enable calibration and DQ/DQS centering are as follows:
•
To calculate when the read data is received after a read command is issued to
setup the Data Valid Prediction FIFO (VFIFO) cycle
•
To align the input data (DQ) with respect to the clock (DQS) to maximize the read
margins (DDR2 and DDR3 only)
DQS enable calibration and DQ/DQS centering consists of the following actions:
•
Guaranteed Write
•
DQS Enable Calibration
•
DQ/DQS Centering
The following figure illustrates the components in the read data path that the
sequencer calibrates in this stage. (The round knobs in the figure represent
configurable hardware over which the sequencer has control.)
External Memory Interface Handbook Volume 3: Reference Material
60
1 Functional Description—UniPHY
Figure 24.
Read Data Path Calibration Model
DQS
Capture
DQ
Read Data
LFIFO
DQS
Enable
DQS
Read Enable
Read
Calibration
VFIFO
mem_clk
afi_clk
1.17.4.1 Guaranteed Write
Because initially no communication can be reliably performed with the memory device,
the sequencer uses a guaranteed write mechanism to write data into the memory
device. (For the QDR II protocol, guaranteed write is not necessary, a simple write
mechanism is sufficient.)
The guaranteed write is a write command issued with all data pins, all address and
bank pins, and all command pins (except chip select) held constant. The sequencer
begins toggling DQS well before the expected latch time at memory and continues to
toggle DQS well after the expected latch time at memory. DQ-to-DQS relationship is
not a factor at this stage because DQ is held constant.
Figure 25.
Guaranteed Write of Zeros
Ex tended ea rlier
DQ[0]
Ac tual burst
Ex tended later
000 0 00 0 00 0
DQ S
External Memory Interface Handbook Volume 3: Reference Material
61
1 Functional Description—UniPHY
The guaranteed write consists of a series of back-to-back writes to alternating
columns and banks. For example, for DQ[0] for the DDR3 protocol, the guaranteed
write performs the following operations:
•
Writes a full burst of zeros to bank 0, column 0
•
Writes a full burst of zeros to bank 0, column 1
•
Writes a full burst of ones to bank 3, column 0
•
Writes a full burst of ones to bank 3, column 1
(Different protocols may use different combinations of banks and columns.)
The guaranteed write is followed by back-to-back read operations at alternating
banks, effectively producing a stream of zeros followed by a stream of ones, or vice
versa. The sequencer uses the zero-to-one and one-to-zero transitions in between the
two bursts to identify a correct read operation, as shown in the figure below.
Although the approach described above for pin DQ[0] would work by writing the same
pattern to all DQ pins, it is more effective and robust to write (and read) alternating
ones and zeros to alternating DQ bits. The value of the DQ bit is still constant across
the burst, and the back-to-back read mechanism works exactly as described above,
except that odd DQ bits have ones instead of zeros, or vice versa.
The guaranteed write does not ensure a correct DQS-to-memory clock alignment at
the memory device—DQS-to-memory clock alignment is performed later, in stage 2 of
the calibration process. However, the process of guaranteed write followed by read
calibration is repeated several times for different DQS-to-memory clock alignments, to
ensure at least one correct alignment is found.
Figure 26.
Back to Back Reads on Pin DQ[0]
B an k 0
B an k 3
C o lu m n 0
000 0 00 0 0
111 1 11 1 1
C o lu m n 1
111 1 11 1 1
000 0 00 0 0
B an k 3
B an k 0
1.17.4.2 DQS Enable Calibration
DQS enable calibration ensures reliable capture of the DQ signal without glitches on
the DQS line. At this point LFIFO is set to its maximum value to guarantee a reliable
read from read capture registers to the core. Read latency is minimized later.
Note:
The full DQS enable calibration is applicable only for DDR2 and DDR3 protocols;
QDR II and RLDRAM protocols use only the VFIFO-based cycle-level calibration,
described below.
Note:
Delay and phase values used in this section are examples, for illustrative purposes.
Your exact values may vary depending on device and configuration.
DQS enable calibration controls the timing of the enable signal using 3 independent
controls: a cycle-based control (the VFIFO), a phase control, and a delay control. The
VFIFO selects the cycle by shifting the controller-generated read data enable signal,
External Memory Interface Handbook Volume 3: Reference Material
62
1 Functional Description—UniPHY
rdata_en, by a number of full-rate clock cycles. The phase is controlled using the
DLL, while the delays are adjusted using a sequence of individual delay taps. The
resolution of the phase and delay controls varies with family and configuration, but is
approximately 45° for the phase, and between 10 and 50 picoseconds for the delays.
The sequencer finds the two edges of the DQS enable window by searching the space
of cycles, phases, and delays (an exhaustive search can usually be avoided by initially
assuming the window is at least one phase wide). During the search, to test the
current settings, the sequencer issues back-to-back reads from column 0 of bank 0
and bank 3, and column 1 of bank 0 and bank 3, as shown in the preceding figure.
Two full bursts are read and compared with the reference data for each phase and
delay setting.
Once the sequencer identifies the two edges of the window, it center-aligns the falling
edge of the DQS enable signal within the window. At this point, per-bit deskew has not
yet been performed, therefore not all bits are expected to pass the read test; however,
for read calibration to succeed, at least one bit per group must pass the read test.
The following figure shows the DQS and DQS enable signal relationship. The goal of
DQS enable calibration is to find settings that satisfy the following conditions:
•
The DQS enable signal rises before the first rising edge of DQS.
•
The DQS enable signal is at one after the second-last falling edge of DQS.
•
The DQS enable signal falls before the last falling edge of DQS.
The ideal position for the falling edge of the DQS enable signal is centered between
the second-last and last falling edges of DQS.
Figure 27.
DQS and DQS Enable Signal Relationships
Row 1
Row 2
DQS +90
dqs_enable (inside I/O) VFIFO Latency
Row 3
dqs_enable aligned
Row 4
dqs_enable (inside I/O)
Row 5
dqs_enable aligned
nable
DQS E
VFIFO Latency 1
Search for first
working setting
nable
DQS E
Search for last
working setting
External Memory Interface Handbook Volume 3: Reference Material
63
1 Functional Description—UniPHY
The following points describe each row of the above figure:
•
Row 1 shows the DQS signal shifted by 90° to center-align it to the DQ data.
•
Row 2 shows the raw DQS enable signal from the VFIFO.
•
Row 3 shows the effect of sweeping DQS enable phases. The first two settings
(shown in red) fail to properly gate the DQS signal because the enable signal turns
off before the second-last falling edge of DQS. The next six settings (shown in
green) gate the DQS signal successfully, with the DQS signal covering DQS from
the first rising edge to the second-last falling edge.
•
Row 4 shows the raw DQS enable signal from the VFIFO, increased by one clock
cycle relative to Row 2.
•
Row 5 shows the effect of sweeping DQS enable, beginning from the initial DQS
enable of Row 4. The first setting (shown in green) successfully gates DQS, with
the signal covering DQS from the first rising edge to the second-last falling edge.
The second signal (shown in red), does not gate DQS successfully because the
enable signal extends past the last falling edge of DQS. Any further adjustment
would show the same failure.
1.17.4.3 Centering DQ/DQS
The centering DQ/DQS stage attempts to align DQ and DQS signals on reads within a
group. Each DQ signal within a DQS group might be skewed and consequently arrive
at the FPGA at a different time. At this point, the sequencer sweeps each DQ signal in
a DQ group to align them, by adjusting DQ input delay chains (D1).
The following figure illustrates a four DQ/DQS group per-bit-deskew and centering.
Figure 28.
Per-bit Deskew
DQ0
DQ1
DQ2
DQ3
DQ4
Original DQ relationship
in a DQ group
DQ signals aligned to left DQS centered with respect to
by tuning D1 delay chains DQ by adjusting DQ signals
To align and center DQ and DQS, the sequencer finds the right edge of DQ signals with
respect to DQS by sweeping DQ signals within a DQ group to the right until a failure
occurs. In the above figure, DQ0 and DQ3 fail after six taps to the right; DQ1 and DQ2
fail after 5 taps to the right. To align the DQ signals, DQ0 and DQ3 are shifted to the
right by 1 tap.
External Memory Interface Handbook Volume 3: Reference Material
64
1 Functional Description—UniPHY
To find the center of DVW, the DQS signal is shifted to the right until a failure occurs.
In the above figure, a failure occurs after 3 taps, meaning that there are 5 taps to the
right edge and 3 taps to the left edge. To center-align DQ and DQS, the sequencer
shifts the aligned DQ signal by 1 more tap to the right.
Note:
The sequencer does not adjust DQS directly; instead, the sequencer center-aligns
DQS with respect to DQ by delaying the DQ signals.
1.17.5 Stage 2: Write Calibration Part One
The objectives of the write calibration stage are to align DQS to the memory clock at
each memory device, and to compensate for address, command, and memory clock
skew at each memory device. This stage is important because the address, command,
and clock signals for each memory component arrive at different times.
Note:
This stage applies only to DDR2, DDR3, LPDDR2, and RLDRAM II protocols; it does not
apply to the QDR II and QDR II+ protocols.
Memory clock signals and DQ/DM and DQS signals have specific relationships
mandated by the memory device. The PHY must ensure that these relationships are
met by skewing DQ/DM and DQS signals. The relationships between DQ/DM and DQS
and memory clock signals must meet the tDQSS, tDSS, and tDSH timing constraints.
The sequencer calibrates the write data path using a variety of random burst patterns
to compensate for the jitter on the output data path. Simple write patterns are
insufficient to ensure a reliable write operation because they might cause imprecise
DQS-to-CK alignments, depending on the actual capture circuitry on a memory device.
The write patterns in the write leveling stage have a burst length of 8, and are
generated by a linear feedback shift register in the form of a pseudo-random binary
sequence.
The write data path architecture is the same for DQ, DM, and DQS pins. The following
figure illustrates the write data path for a DQ signal. The phase coming out of the
Output Phase Alignment block can be set to different values to center-align DQS with
respect to DQ, and it is the same for data, OE, and OCT of a given output.
Figure 29.
Write Data Path
OC T
write_ck
2
DDIO
T9
?
T10
?
DDIO
T9
?
T10
?
DDIO
T9
?
T10
?
?
Output
Phase Alignment
OE
2
data
2
OC T
External Memory Interface Handbook Volume 3: Reference Material
65
1 Functional Description—UniPHY
In write leveling, the sequencer performs write operations with different delay and
phase settings, followed by a read. The sequencer can implement any phase shift
between 0° and 720° (depending on device and configuration). The sequencer uses
the Output Phase Alignment for coarse delays and T9 and T10 for fine delays; T9 has
15 taps of 50 ps each, and T10 has 7 taps of 50 ps each.
The DQS signal phase is held at +90° with respect to DQ signal phase (Stratix IV
example).
Note:
Coarse delays are called phases, and fine delays are called delays; phases are
process, voltage, and temperature (PVT) compensated, delays are not (depending on
family).
For 28 nm devices:
Note:
•
I/O delay chains are not PVT compensated.
•
DQS input delay chain is PVT compensated.
•
Leveling delay chains are PVT compensated (does not apply to Arria V or Cyclone
V devices).
•
T11 delay chain for postamble gating has PVT and nonPVT compensated modes,
but the PVT compensated mode is not used.
Delay and phase values used in this section are examples, for illustrative purposes.
Your exact values may vary depending on device and configuration.
The sequencer writes and reads back several burst-length-8 patterns. Because the
sequencer has not performed per-bit deskew on the write data path, not all bits are
expected to pass the write test. However, for write calibration to succeed, at least one
bit per group must pass the write test. The test begins by shifting the DQ/DQS phase
until the first write operation completes successfully. The DQ/DQS signals are then
delayed to the left by D5 and D6 to find the left edge for that working phase. Then
DQ/DQS phase continues the shift to find the last working phase. For the last working
phase, DQ/DQS is delayed in 50 ps steps to find the right edge of the last working
phase.
The sequencer sweeps through all possible phase and delay settings for each DQ
group where the data read back is correct, to define a window within which the PHY
can reliably perform write operations. The sequencer picks the closest value to the
center of that window as the phase/delay setting for the write data path.
1.17.6 Stage 3: Write Calibration Part Two—DQ/DQS Centering
The process of DQ/DQS centering in write calibration is similar to that performed in
read calibration, except that write calibration is performed on the output path, using
D5 and D6 delay chains.
1.17.7 Stage 4: Read Calibration Part Two—Read Latency Minimization
At this stage of calibration the sequencer adjusts LFIFO latency to determine the
minimum read latency that guarantees correct reads.
External Memory Interface Handbook Volume 3: Reference Material
66
1 Functional Description—UniPHY
Read Latency Tuning
In general, DQ signals from different DQ groups may arrive at the FPGA in a staggered
fashion. In a DIMM or multiple memory device system, the DQ/DQS signals from the
first memory device arrive sooner, while the DQ/DQS signals from the last memory
device arrive the latest at the FPGA.
LFIFO transfers data from the capture registers in IOE to the core and aligns read data
to the AFI clock. Up to this point in the calibration process, the read latency has been
a maximum value set initially by LFIFO; now, the sequencer progressively lowers the
read latency until the data can no longer be transferred reliably. The sequencer then
increases the latency by one cycle to return to a working value and adds an additional
cycle of margin to assure reliable reads.
1.17.8 Calibration Signals
The following table lists signals produced by the calibration process.
Table 19.
Calibration Signals
Signal
Description
afi_cal_fail
Asserts high if calibration fails.
afi_cal_success
Asserts high if calibration is successful.
1.17.9 Calibration Time
The time needed for calibration varies, depending on many factors including the
interface width, the number of ranks, frequency, board layout, and difficulty of
calibration. In general, designs using the Nios II-based sequencer will take longer to
calibrate than designs using the RTL-based sequencer.
The following table lists approximate typical and maximum calibration times for
various protocols.
Table 20.
Approximate Calibration Times
Protocol
Typical Calibration Time
Maximum Calibration Time
DDR2, DDR3, LPDDR2, RLDRAM 3
50-250 ms
Can take several minutes if the interface is difficult to
calibrate, or if calibration initially fails and exhausts
multiple retries.
QDR II/II+, RLDRAM II (with
Nios II-based sequencer)
50-100 ms
Can take several minutes if the interface is difficult to
calibrate, or if calibration initially fails and exhausts
multiple retries.
QDR II/II+, RLDRAM II (with RTLbased sequencer)
<5 ms
<5 ms
1.18 Document Revision History
Date
Version
Changes
May 2017
2017.05.08
Rebranded as Intel.
October 2016
2016.10.31
•
Changed Note 3 to the Word-Aligned Writes figure, in the PHY-toController Interfaces topic.
continued...
External Memory Interface Handbook Volume 3: Reference Material
67
1 Functional Description—UniPHY
Date
Version
Changes
May 2016
2016.05.02
•
Added statement that Efficiency Monitor and Protocol checker is not
available for QDR II and QDR II+ SRAM, or for the MAX 10 device family,
or for Arria V or Cyclone V designs using the Hard Memory Controller, to
Efficiency Monitor and Protocol Checker topic.
November 2015
2015.11.02
•
•
Replaced the AFI 4.0 Specification with the AFI 3.0 Specification.
Replaced instances of Quartus II with Quartus Prime.
May 2015
2015.05.04
Maintenance release.
December 2014
2014.12.15
•
•
•
August 2014
2014.08.15
•
•
December 2013
2013.12.16
•
•
•
•
•
•
•
•
•
November 2012
3.1
•
•
•
•
Added several parameters to the AFI Specification section:
— MEM_IF_BANKGROUP_WIDTH
— MEM_IF_C_WIDTH
— MEM_IF_CKE_WIDTH
— MEM_IF_ODT_WIDTH
— AFI_BANKGROUP_WIDTH
— AFI_C_WIDTH
— AFI_CKE_WIDTH
— AFI_ODT_WIDTH
Added several signals to the AFI Specification section:
— afi_addr
— afi_bg
— afi_c_n
— afi_rw_n
— afi_act_n
— afi_par
— afi_alert_n
— afi_ainv
— afi_dinv
Changed the Width information for several signals in the AFI Specification
section:
— afi_dqs_burst
— afi_wdata_valid
— afi_stl_refresh_done
— afi_seq_busy
— afi_ctl_long_idle
Added note about 32-bit word addresses to Register Maps and UniPHY
Register Map tables.
Changed register map information for address 0x004, bit 26, in UniPHY
Register Map table.
Removed references to HardCopy.
Removed DLL Offset Control Block.
Removed references to SOPC Builder.
Increased minimum recommended pulse width for global_reset_n signal
to 100ns.
Corrected terminology inconsistency.
Added information explaining PVT compensation.
Added quarter-rate information to PHY-to-Controller Interfaces section.
Expanded descriptions of INIT_FAILING_STAGE and
INIT_FAILING_SUBSTAGE in UniPHY Register map.
Added footnote about afi_rlat signal to Calibration Status Signals table.
Moved Controller Register Map to Functional Description—HPC II
Controller chapter.
Updated Sequencer States information in Table 1–2.
Enhanced Using a Custom Controller information.
Enhanced Tracking Manager information.
continued...
External Memory Interface Handbook Volume 3: Reference Material
68
1 Functional Description—UniPHY
Date
Version
Changes
•
•
•
•
Added
Added
Added
Added
Ping Pong PHY information.
RLDRAM 3 support.
LRDIMM support.
Arria V GZ support.
Shadow Registers section.
LPDDR2 support.
new AFI signals.
Calibration Time section.
Feedback icon.
June 2012
3.0
•
•
•
•
•
Added
Added
Added
Added
Added
November 2011
2.1
•
Consolidated UniPHY information from 11.0 DDR2 and DDR3 SDRAM
Controller with UniPHY User Guide, QDR II and QDR II+ SRAM Controller
with UniPHY User Guide, and RLDRAM II Controller with UniPHY IP User
Guide.
Revised Reset and Clock Generation and Dedicated Clock Networks
sections.
Revised Figure 1–3 and Figure 1–5.
Added Tracking Manager to Sequencer section.
Revised Interfaces section for DLL, PLL, and OCT sharing interfaces.
Revised Using a Custom Controller section.
Added UniPHY Calibration Stages section; reordered stages 3 and 4,
removed stage 5.
•
•
•
•
•
•
External Memory Interface Handbook Volume 3: Reference Material
69
2 Functional Description—Intel Stratix® 10 EMIF IP
2 Functional Description—Intel Stratix® 10 EMIF IP
Intel Stratix® 10 devices can interface with external memory devices clocking at
frequencies of up to 1.3 GHz. The external memory interface IP component for Stratix
10 devices provides a single parameter editor for creating external memory interfaces,
regardless of memory protocol. Unlike earlier EMIF solutions which used protocolspecific parameter editors to create memory interfaces via a complex RTL generation
method, the Stratix 10 EMIF solution captures the protocol-specific hardened EMIF
logic of the Stratix 10 device together with more generic soft logic.
The Stratix 10 EMIF solution is designed with the following implementations in mind:
Hard Memory Controller and Hard PHY
This implementation provides a complete external memory interface, based on the
hard memory controller and hard PHY that are part of the Stratix 10 silicon. An
Avalon-MM interface is available for integration with user logic.
Soft Memory Controller and Hard PHY
This implementation provides a complete external memory interface, using an Intelprovided soft-logic-based memory controller and the hard PHY that is part of the
Stratix 10 silicon. An Avalon-MM interface is available for integration with user logic.
Custom Memory Controller and Hard PHY (PHY only)
This implementation provides access to the Altera PHY interface (AFI), to allow use of
a custom or third-party memory controller with the hard PHY that is part of the Stratix
10 silicon. Because only the PHY component is provided by Intel, this configuration is
also known as PHY only.
2.1 Stratix 10 Supported Memory Protocols
The following table lists the external memory protocols supported by Stratix 10
devices.
Table 21.
Supported Memory Protocols
Protocol
Hard Controller and Hard
PHY
Soft Controller and Hard
PHY
PHY Only
DDR4
Yes
—
Yes
DDR3
Yes
—
Yes
LPDDR3
Yes
—
Yes
RLDRAM 3
—
Third party
Yes
QDR II/II+/II+ Xtreme
—
Yes
—
QDR-IV
—
Yes
—
Intel Corporation. All rights reserved. Intel, the Intel logo, Altera, Arria, Cyclone, Enpirion, MAX, Nios, Quartus
and Stratix words and logos are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other
countries. Intel warrants performance of its FPGA and semiconductor products to current specifications in
accordance with Intel's standard warranty, but reserves the right to make changes to any products and services
at any time without notice. Intel assumes no responsibility or liability arising out of the application or use of any
information, product, or service described herein except as expressly agreed to in writing by Intel. Intel
customers are advised to obtain the latest version of device specifications before relying on any published
information and before placing orders for products or services.
*Other names and brands may be claimed as the property of others.
ISO
9001:2008
Registered
2 Functional Description—Intel Stratix® 10 EMIF IP
Memory protocols not listed above are not supported by the Stratix 10 EMIF IP;
however, you can implement a custom memory interface for these protocols using the
PHYLite Megafunction.
2.2 Stratix 10 EMIF IP Support for 3DS/TSV DDR4 Devices
The Stratix 10 EMIF IP supports high-density DDR4 memory devices with threedimensional stacked-die (3DS) memory technology in RDIMM format. Both 2H and 4H
3DS devices are supported.
To configure a DDR4 RDIMM memory interface to use 3DS devices, do one of the
following:
•
Select one of the 3DS-enabled IP presets available in the parameter editor.
•
1. Specify a non-zero value for the Chip ID Width parameter on the Memory
tab in the parameter editor.
2.
Enter the 3DS-specific specific timing parameters (tRDD_dir, tFAW_dir,
tRFC_dir) from the memory vendor's data sheet appropriately.
2.3 Migrating to Stratix 10 from Previous Device Families
There is no automatic migration mechanism for external memory interface IP
generated for previous device families.
To migrate an existing EMIF IP from an earlier device family to Stratix 10, you must
reparameterize and regenerate the IP targeting Stratix 10, using either the IP Catalog
or Qsys. If you attempt to recompile an existing IP generated for a previous device
family, you will encounter errors in the Quartus Prime software.
UniPHY-based IP continues to be supported for previous device families.
2.4 Stratix 10 EMIF Architecture: Introduction
The Stratix 10 EMIF architecture contains many new hardware features designed to
meet the high-speed requirements of emerging memory protocols, while consuming
the smallest amount of core logic area and power.
The following are key hardware features of the Stratix 10 EMIF architecture:
Hard Sequencer
The sequencer employs a hard Nios II processor, and can perform memory calibration
for a wide range of protocols. You can share the sequencer among multiple memory
interfaces of the same or different protocols.
Note:
You cannot use the hard Nios II processor for any user applications after calibration is
complete.
Hard PHY
The hard PHY in Stratix 10 devices can interface with external memories running at
speeds of up to 1.3 GHz. The PHY circuitry is hardened in the silicon, which simplifies
the challenges of achieving timing closure and minimal power consumption.
External Memory Interface Handbook Volume 3: Reference Material
71
2 Functional Description—Intel Stratix® 10 EMIF IP
Hard Memory Controller
The hard memory controller reduces latency and minimizes core logic consumption in
the external memory interface. The hard memory controller supports the DDR3,
DDR4, and LPDDR3 memory protocols.
PHY-Only Mode
Protocols that use a hard controller (DDR4, DDR3, LPDDR3, and RLDRAM 3), provide a
PHY-only option, which generates only the PHY and sequencer, but not the controller.
This PHY-only mode provides a mechanism by which to integrate your own custom
soft controller.
High-Speed PHY Clock Tree
Dedicated high speed PHY clock networks clock the I/O buffers in Stratix 10 EMIF IP.
The PHY clock trees exhibit low jitter and low duty cycle distortion, maximizing the
data valid window.
Automatic Clock Phase Alignment
Automatic clock phase alignment circuitry dynamically adjusts the clock phase of core
clock networks to match the clock phase of the PHY clock networks. The clock phase
alignment circuitry minimizes clock skew that can complicate timing closure in
transfers between the FPGA core and the periphery.
Resource Sharing
The Stratix 10 architecture simplifies resource sharing between memory interfaces.
Resources such as the OCT calibration block, PLL reference clock pin, and core clock
can be shared. The hard Nios processor in the I/O subsystem manager (I/O SSM)
must be shared across all interfaces in a column.
2.4.1 Stratix 10 EMIF Architecture: I/O Subsystem
The I/O subsystem consists of three columns inside the core of Stratix 10 devices.
Each column can be thought of as loosely analogous to an I/O bank.
Figure 30.
Stratix 10 I/O Subsystem
Core Fabric
I/O Column
Transceivers
(if applicable)
External Memory Interface Handbook Volume 3: Reference Material
72
2 Functional Description—Intel Stratix® 10 EMIF IP
The I/O subsystem provides the following features:
•
General-purpose I/O registers and I/O buffers
•
On-chip termination control (OCT)
•
I/O PLLs for external memory interfaces and user logic
•
Low-voltage differential signaling (LVDS)
•
External memory interface components, as follows:
—
Hard memory controller
—
Hard PHY
—
Hard Nios processor and calibration logic
—
DLL
2.4.2 Stratix 10 EMIF Architecture: I/O Column
Stratix 10 devices have two I/O columns, which contain the hardware related to
external memory interfaces.
Each I/O column contains the following major parts:
A hardened Nios processor with dedicated memory. This Nios block is referred to
as the I/O SSM.
•
Up to 13 I/O banks. Each I/O bank contains the hardware necessary for an
external memory interface.
I/O Column
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
Individual
I/O Banks
Transceiver
Block
2L
3H
2K
3G
2J
3F
2I
3E
2H
3D
2G
3C
2F
3B
2A
3A
I/O
Column
I/O
Column
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
I/O Center
Transceiver Block
Transceiver Block
Figure 31.
•
I/O PLL
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
I/O Lane
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
I/O Lane
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
I/O DLL
I/O CLK
OCT
VR
Hard Memory Controller
and
PHY Sequencer
I/O Lane
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
I/O Lane
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
Bank
Control
External Memory Interface Handbook Volume 3: Reference Material
73
2 Functional Description—Intel Stratix® 10 EMIF IP
2.4.3 Stratix 10 EMIF Architecture: I/O SSM
Each column includes one I/O subsystem manager (I/O SSM), which contains a
hardened Nios II processor with dedicated memory. The I/O SSM is responsible for
calibration of all the EMIFs in the column.
The I/O SSM includes dedicated memory which stores both the calibration algorithm
and calibration run-time data. The hardened Nios II processor and the dedicated
memory can be used only by an external memory interface, and cannot be employed
for any other use. The I/O SSM can interface with soft logic, such as the debug toolkit,
via an Avalon-MM bus.
The I/O SSM is clocked by an on-die oscillator, and therefore does not consume a PLL.
2.4.4 Stratix 10 EMIF Architecture: I/O Bank
A single I/O bank contains all the hardware needed to build an external memory
interface. Each I/O column contains up to 13 I/O banks; the exact number of banks
depends on device size and pin package. You can make a wider interface by
connecting multiple banks together.
Each I/O bank resides in an I/O column, and contains the following components:
Figure 32.
•
Hard memory controller
•
Sequencer components
•
PLL and PHY clock trees
•
DLL
•
Input DQS clock trees
•
48 pins, organized into four I/O lanes of 12 pins each
I/O Bank Architecture in Stratix 10 Devices
to / from bank above
I/O Bank
Output Path
Input Path
I/O Lane 3
S equencer
Output Path
Input Path
I/O Lane 2
Output Path
Input Path
I/O Lane 1
PLL
C lock Phase
Alignment
Output Path
Input Path
I/O Lane 0
Memory
C ontroller
to / from
FPGA core
to / from bank below
External Memory Interface Handbook Volume 3: Reference Material
74
2 Functional Description—Intel Stratix® 10 EMIF IP
I/O Bank Usage
The pins in an I/O bank can serve as address and command pins, data pins, or clock
and strobe pins for an external memory interface. You can implement a narrow
interface, such as a DDR3 or DDR4 x8 interface, with only a single I/O bank. A wider
interface, such as x72 or x144, can be implemented by configuring multiple adjacent
banks in a multi-bank interface. Any pins in a bank which are not used by the external
memory interface remain available for use as general purpose I/O pins (of the same
voltage standard).
Every I/O bank includes a hard memory controller which you can configure for DDR3
or DDR4. In a multi-bank interface, only the controller of one bank is active;
controllers in the remaining banks are turned off to conserve power.
To use a multi-bank Stratix 10 EMIF interface, you must observe the following rules:
•
Designate one bank as the address and command bank.
•
The address and command bank must contain all the address and command pins.
•
The locations of individual address and command pins within the address and
command bank must adhere to the pin map defined in the pin table— regardless
of whether you use the hard memory controller or not.
•
If you do use the hard memory controller, the address and command bank
contains the active hard controller.
All the I/O banks in a column are capable of functioning as the address and command
bank. However, for minimal latency, you should select the center-most bank of the
interface as the address and command bank.
2.4.5 Stratix 10 EMIF Architecture: I/O Lane
An I/O bank contains 48 I/O pins, organized into four I/O lanes of 12 pins each.
Each I/O lane can implement one x8/x9 read capture group (DQS group), with two
pins functioning as the read capture clock/strobe pair (DQS/DQS#), and up to 10 pins
functioning as data pins (DQ and DM pins). To implement x18 and x36 groups, you
can use multiple lanes within the same bank.
It is also possible to implement a pair of x4 groups in a lane. In this case, four pins
function as clock/strobe pair, and 8 pins function as data pins. DM is not available for
x4 groups. There must be an even number of x4 groups for each interface.
For x4 groups, DQS0 and DQS1 must be placed in the same I/O lane as a pair.
Similarly, DQS2 and DQS3 must be paired. In general, DQS(x) and DQS(x+1) must be
paired in the same I/O lane.
Table 22.
Lanes Used Per Group
Group Size
Number of Lanes Used
Maximum Number of Data Pins per
Group
x8 / x9
1
10
x18
2
22
x36
4
46
pair of x4
1
4 per group, 8 per lane
External Memory Interface Handbook Volume 3: Reference Material
75
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 33.
x4 Group
Memory
Controller
Sequencer
PLL
Clock Phase
Alignment
Figure 34.
Output Path
I/O Lane 3
Input Path
X4 Groups 6 and 7
Output Path
I/O Lane 2
Input Path
X4 Groups 4 and 5
Output Path
I/O Lane 1
Input Path
X4 Groups 2 and 3
Output Path
I/O Lane 0
Input Path
X4 Groups 0 and 1
Output Path
I/O Lane 3
Input Path
X8 Group 3
Output Path
I/O Lane 2
Input Path
X8 Group 2
Output Path
I/O Lane 1
Input Path
X8 Group 1
Output Path
I/O Lane 0
Input Path
X8 Group 0
x8 Group
Memory
Controller
Sequencer
PLL
Clock Phase
Alignment
External Memory Interface Handbook Volume 3: Reference Material
76
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 35.
x18 Group
M emo ry
Controller
Sequ encer
PLL
Clock Phase
Alignment
Figure 36.
Output Path
Input Path
X18 Group 0
Output Path
Input Path
I/O Lane 3
I/O Lane 2
Output Path
Input Path
X18 Group 1
Output Path
Input Path
I/O Lane 1
Output Path
Input Path
I/O Lane 3
Output Path
Input Path
X36 Group 0
Output Path
Input Path
I/O Lane 2
Output Path
Input Path
I/O Lane 0
I/O Lane 0
x36 Group
M emo ry
Controller
Sequ encer
PLL
Clock Phase
Alignment
I/O Lane 1
2.4.6 Stratix 10 EMIF Architecture: Input DQS Clock Tree
The input DQS clock tree is a balanced clock network that distributes the read capture
clock and strobe from the external memory device to the read capture registers inside
the I/Os.
You can configure an input DQS clock tree in x4 mode, x8/x9 mode, x18 mode, or x36
mode.
External Memory Interface Handbook Volume 3: Reference Material
77
2 Functional Description—Intel Stratix® 10 EMIF IP
Within every bank, only certain physical pins at specific locations can drive the input
DQS clock trees. The pin locations that can drive the input DQS clock trees vary,
depending on the size of the group.
Table 23.
Pins Usable as Read Capture Clock / Strobe Pair
Group Size
Index of Lanes Spanned
by Clock Tree
In-Bank Index of Pins Usable as Read Capture Clock /
Strobe Pair
Positive Leg
Negative Leg
x4
0A
4
5
x4
0B
8
9
x4
1A
16
17
x4
1B
20
21
x4
2A
28
29
x4
2B
32
33
x4
3A
40
41
x4
3B
44
45
x8 / x9
0
4
5
x8 / x9
1
16
17
x8 / x9
2
28
29
x8 / x9
3
40
41
x18
0, 1
12
13
x18
2, 3
36
37
x36
0, 1, 2, 3
20
21
2.4.7 Stratix 10 EMIF Architecture: PHY Clock Tree
Dedicated high-speed clock networks drive I/Os in Stratix 10 EMIF. Each PHY clock
network spans only one bank.
The relatively short span of the PHY clock trees results in low jitter and low duty-cycle
distortion, maximizing the data valid window.
The PHY clock tree in Stratix 10 devices can run as fast as 1.3 GHz. All Stratix 10
external memory interfaces use the PHY clock trees.
2.4.8 Stratix 10 EMIF Architecture: PLL Reference Clock Networks
Each I/O bank includes a PLL that can drive the PHY clock trees of that bank, through
dedicated connections. In addition to supporting EMIF-specific functions, such PLLs
can also serve as general-purpose PLLs for user logic.
Stratix 10 external memory interfaces that span multiple banks use the PLL in each
bank. (Some previous device families relied on a single PLL with clock signals
broadcast to all I/Os by a clock network.) The Stratix 10 architecture allows for
relatively short PHY clock networks, reducing jitter and duty-cycle distortion.
External Memory Interface Handbook Volume 3: Reference Material
78
2 Functional Description—Intel Stratix® 10 EMIF IP
The following mechanisms ensure that the clock outputs of individual PLLs in a multibank interface remain in phase:
•
A single PLL reference clock source feeds all PLLs. The reference clock signal
reaches the PLLs by a balanced PLL reference clock tree. The Quartus Prime
software automatically configures the PLL reference clock tree so that it spans the
correct number of banks.
•
The EMIF IP sets the PLL M and N values appropriately to maintain synchronization
among the clock dividers across the PLLs. This requirement restricts the legal PLL
reference clock frequencies for a given memory interface frequency and clock rate.
The Stratix 10 EMIF IP parameter editor automatically calculates and displays the
set of legal PLL reference clock frequencies. If you plan to use an on-board
oscillator, you must ensure that its frequency matches the PLL reference clock
frequency that you select from the displayed list. The correct M and N values of
the PLLs are set automatically based on the PLL reference clock frequency that
you select.
Note:
The PLL reference clock pin may be placed in the address and command I/O bank or in
a data I/O bank, there is no implication on timing. However, for debug flexibility, it is
recommended to place the PLL reference clock in the address and command I/O bank.
Figure 37.
PLL Balanced Reference Clock Tree
I/O Bank
PLL
PHY
clock
tree
I/O Bank
I/O Column
ref_clk
Balanced Reference Clock Network
PLL
PHY
clock
tree
I/O Bank
PLL
PHY
clock
tree
I/O Bank
PLL
PHY
clock
tree
2.4.9 Stratix 10 EMIF Architecture: Clock Phase Alignment
In Stratix 10 external memory interfaces, a global clock network clocks registers
inside the FPGA core, and the PHY clock network clocks registers inside the FPGA
periphery. Clock phase alignment circuitry employs negative feedback to dynamically
adjust the phase of the core clock signal to match the phase of the PHY clock signal.
The clock phase alignment feature effectively eliminates the clock skew effect in all
transfers between the core and the periphery, facilitating timing closure. All Stratix 10
external memory interfaces employ clock phase alignment circuitry.
External Memory Interface Handbook Volume 3: Reference Material
79
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 38.
Clock Phase Alignment Illustration
FPGA Periphery
FPGA Core
PHY Clock
Network
PLL
Core Clock
Network
Clock Phase Alignment
p
Figure 39.
+t
Effect of Clock Phase Alignment
Without Clock Phase Alignment
With Clock Phase Alignment
Core clock
Core clock
PHY clock
PHY clock
Skew between core
and PHY clock network
Core and PHY clocks aligned dynamically
by clock phase alignment
2.5 Hardware Resource Sharing Among Multiple Stratix 10 EMIFs
Often, it is necessary or desirable to share certain hardware resources between
interfaces.
2.5.1 I/O SSM Sharing
The I/O SSM contains a hard Nios-II processor and dedicated memory storing the
calibration software code and data.
External Memory Interface Handbook Volume 3: Reference Material
80
2 Functional Description—Intel Stratix® 10 EMIF IP
When a column contains multiple memory interfaces, the hard Nios-II processor
calibrates each interface serially. Interfaces placed within the same I/O column always
share the same I/O SSM. The Quartus Prime Fitter handles I/O SSM sharing
automatically.
2.5.2 I/O Bank Sharing
Data lanes from multiple compatible interfaces can share a physical I/O bank to
achieve a more compact pin placement. To share an I/O bank, interfaces must use the
same memory protocol, rate, frequency, I/O standard, and PLL reference clock signal.
Rules for Sharing I/O Banks
•
A bank cannot serve as the address and command bank for more than one
interface. This means that lanes which implement address and command pins for
different interfaces cannot be allocated to the same physical bank.
Note: An exception to the above rule exists when two interfaces are configured in
a Ping-Pong PHY fashion. In such a configuration, two interfaces share the
same set of address and command pins, effectively meaning that they share
the same address and command tile.
•
Pins within a lane cannot be shared by multiple memory interfaces.
•
Pins that are not used by EMIF IP can serve as general-purpose I/Os of compatible
voltage and termination settings.
•
You can configure a bank as LVDS or as EMIF, but not both at the same time.
•
Interfaces that share banks must reside at consecutive bank locations.
The following diagram illustrates two x16 interfaces sharing an I/O bank. The two
interfaces share the same clock phase alignment block, so that one core clock signal
can interact with both interfaces. Without sharing, the two interfaces would occupy a
total of four physical banks instead of three.
External Memory Interface Handbook Volume 3: Reference Material
81
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 40.
I/O Bank Sharing
Memory
Controller
Sequencer
PLL
Clock Phase
Alignment
Memory
Controller
Sequencer
PLL
Clock Phase
Alignment
Memory
Controller
Sequencer
PLL
Clock Phase
Alignment
Output Path
I/O Lane 3
Input Path
Address/Command Lane 2
Output Path
I/O Lane 2
Input Path
Address/Command Lane 1
Output Path
I/O Lane 1
Input Path
Address/Command Lane 0
Output Path
I/O Lane 0
Input Path
DQ Group 0
Output Path
I/O Lane 3
Input Path
DQ Group 1
Output Path
I/O Lane 2
Interface 1
Input Path
Output Path
I/O Lane 1
Input Path
Output Path
I/O Lane 0
Input Path
DQ Group 1
Output Path
I/O Lane 3
Input Path
Address/Command Lane 2
Output Path
I/O Lane 2
Input Path
Address/Command Lane 1
Output Path
I/O Lane 1
Input Path
Address/Command Lane 0
Output Path
I/O Lane 0
Input Path
DQ Group 0
Interface 2
2.5.3 PLL Reference Clock Sharing
In Stratix 10, every I/O bank contains a PLL, meaning that it is not necessary to share
PLLs in the interest of conserving resources. Nonetheless, it is often desirable to share
PLLs for other reasons.
External Memory Interface Handbook Volume 3: Reference Material
82
2 Functional Description—Intel Stratix® 10 EMIF IP
You might want to share PLLs between interfaces for the following reasons:
•
To conserve pins.
•
When combined with the use of the balanced PLL reference clock tree, to allow the
clock signals at different interfaces to be synchronous and aligned to each other.
For this reason, interfaces that share core clock signals must also share the PLL
reference clock signal.
To implement PLL reference clock sharing, in your RTL code connect the PLL reference
clock signal at your design's top-level to the PLL reference clock port of multiple
interfaces.
To share a PLL reference clock, the following requirements must be met:
•
Interfaces must expect a reference clock signal of the same frequency.
•
Interfaces must be placed in the same column.
•
Interfaces must be placed at adjacent bank locations.
2.5.4 Core Clock Network Sharing
It is often desirable or necessary for multiple memory interfaces to be accessible using
a single clock domain in the FPGA core.
You might want to share core clock networks for the following reasons:
•
To minimize the area and latency penalty associated with clock domain crossing.
•
To minimize consumption of core clock networks.
Multiple memory interfaces can share the same core clock signals under the following
conditions:
•
The memory interfaces have the same protocol, rate, frequency, and PLL reference
clock source.
•
The interfaces reside in the same I/O column.
•
The interfaces reside in adjacent bank locations.
For multiple memory interfaces to share core clocks, you must specify one of the
interfaces as master and the remaining interfaces as slaves. Use the Core clocks
sharing setting in the parameter editor to specify the master and slaves.
In your RTL, connect the clks_sharing_master_out signal from the master
interface to the clks_sharing_slave_in signal of all the slave interfaces. Both the
master and slave interfaces expose their own output clock ports in the RTL (e.g.
emif_usr_clk, afi_clk), but the signals are equivalent, so it does not matter
whether a clock port from a master or a slave is used.
Core clock sharing necessitates PLL reference clock sharing; therefore, only the
master interface exposes an input port for the PLL reference clock. All slave interfaces
use the same PLL reference clock signal.
External Memory Interface Handbook Volume 3: Reference Material
83
2 Functional Description—Intel Stratix® 10 EMIF IP
2.6 Stratix 10 EMIF IP Component
The external memory interface IP component for Stratix 10 provides a complete
solution for implementing external memory interfaces. The EMIF IP also includes a
protocol-specific calibration algorithm that automatically determines the optimal delay
settings for a robust external memory interface.
The external memory interface IP comprises the following parts:
•
A set of synthesizable files that you can integrate into a larger design
•
A stand-alone synthesizable example design that you can use for hardware
validation
•
A set of simulation files that you can incorporate into a larger project
•
A stand-alone simulation example project that you can use to observe controller
and PHY operation
•
A set of timing scripts that you can use to determine the maximum operating
frequency of the memory interface based on external factors such as board skew,
trace delays, and memory component timing parameters
•
A customized data sheet specific to your memory interface configuration
2.6.1 Instantiating Your Stratix 10 EMIF IP in a Qsys Project
The following steps describe how to instantiate your Stratix 10 EMIF IP in a Qsys
project.
1.
Within the Qsys interface, select Memories and Memory Controllers in the
component Library tree.
2. Under Memories and Memory Controllers, select External Memory
Interfaces (Stratix 10).
3. Under External Memory Interfaces (Stratix 10), select the Stratix 10
External Memory Interface component.
External Memory Interface Handbook Volume 3: Reference Material
84
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 41.
Instantiating Stratix 10 EMIF IP in Qsys
2.6.1.1 Logical Connections
The following logical connections exist in a Stratix 10 EMIF IP core.
Table 24.
Logical Connections Table
Logical Connection
afi_conduit_end
(Conduit)
afi_clk_conduit_end
(Conduit)
afi_half_clk_conduit_end
(Conduit)
afi_reset_n_conduit_end
(Conduit)
cal_debug_avalon_slave
(Avalon Slave/Target)
cal_debug_clk_clock_sink
(Clock Input)
cal_debug_reset_reset_sink
(Reset Input)
Description
The Altera PHY Interface (AFI) connects a memory controller to the PHY. This
interface is exposed only when you configure the memory interface in PHY-Only
mode. The interface is synchronous to the afi_clk clock and afi_reset_n
reset.
Use this clock signal to clock the soft controller logic. The afi_clk is an output
clock coming from the PHY when the memory interface is in PHY-Only mode. The
phase of afi_clk is adjusted dynamically by hard circuitry for the best data
transfer between FPGA core logic and periphery logic with maximum timing
margin. Multiple memory interface instances can share a single afi_clk using
the Core Clocks Sharing option during IP generation.
This clock runs at half the frequency of afi_lk. It is exposed only when the
memory interface is in PHY-Only mode.
This single-bit reset provides a synchronized reset output. Use this signal to reset
all registers that are clocked by either afi_clk or afi_half_clk.
This interface is exposed when the EMIF Debug Toolkit/On-chip Debug Port
option is set to Export. This interface can be connected to a Stratix 10 external
memory interface debug component to allow EMIF Debug Toolkit access, or it can
be used directly by user logic to access calibration diagnostic data.
This clock is used for the cal_debug_avalon_slave interface. It can be
connected to the emif_usr_clk_clock_source interface.
This reset is used for the cal_debug_avalon_slave interface. It can be
connected to the emif_usr_reset_reset_source interface.
continued...
External Memory Interface Handbook Volume 3: Reference Material
85
2 Functional Description—Intel Stratix® 10 EMIF IP
Logical Connection
Description
(Avalon Master/Source)
This interface is exposed when the Enable Daisy-Chaining for EMIF Debug
Toolkit/On-Chip Debug Port option is enabled. Connect this interface to the
cal_debug_avalon_slave interface of the next EMIF instance in the same I/O
column.
cal_debug_out_clk_clock_sourc
e
This interface should be connected to the cal_debug_clk_clock_sink
interface of the next EMIF instance in the same I/O column (similar to
cal_debug_out_avalon_master).
cal_debug_out_avalon_master
(Clock Output)
cal_debug_out_reset_reset_sou
rce
(Reset Output)
This interface should be connected to the cal_debug_reset_reset_sink
interface of the next EMIF instance in the same I/O column (similar to
cal_debug_out_avalon_master).
(Avalon Slave/Target)
This interface allows access to the Efficiency Monitor CSR. For more information,
see the documentation on the UniPHY Efficiency Monitor.
global_reset_reset_sink
This single-wire input port is the asynchronous reset input for the EMIF core.
effmon_csr_avalon_slave
(Reset Input)
pll_ref_clk_clock_sink
(Clock Input)
oct_conduit_end
(Conduit)
mem_conduit_end
(Conduit)
status_conduit_end
(Conduit)
emif_usr_reset_reset_source
(Reset Output)
emif_usr_clk_clock_source
(Clock Output)
ctrl_amm_avalon_slave
(Avalon Slave/Target)
This single-wire input port connects the external PLL reference clock to the EMIF
core. Multiple EMIF cores may share a PLL reference clock source, provided the
restrictions outlined in the PLL Reference Clock Network section are observed.
This logical port is connected to an OCT pin and provides calibrated reference data
for EMIF cores with pins that use signaling standards that require on-chip
termination. Depending on the I/O standard, reference voltage, and memory
protocol, multiple EMIF cores may share a single OCT pin.
This logical conduit can attach an Intel memory model to an EMIF core for
simulation. Memory models for various protocols are available under the
Memories and Memory Controllers — External Memory Interfaces —
Memory Models section of the component library in Qsys. You must ensure that
all configuration parameters for the memory model match the configuration
parameters of the EMIF core.
The status conduit exports two signals that you can sample to determine whether
the calibration operation passed or failed for that core.
You should use this single-bit, synchronized, reset output to reset all components
that are synchronously connected to the EMIF core. Assertion of the global reset
input triggers an assertion of this output as well, therefore you should rely on this
signal only as a reset source for all components connected to the EMIF core.
Use this single-bit clock output to clock all logic connected to the EMIF core. The
phase of this clock signal is adjusted dynamically by circuitry in the EMIF core
such that data can be transferred between core logic and periphery registers with
maximum timing margin. Drive all logic connected to the EMIF core with this clock
signal. Other clock sources generated from the same reference clock or even the
same PLL may have unknown phase relationships. Multiple EMIF cores can share a
single core clock using the Core Clocks Sharing option described in the Example
Design tab of the parameter editor.
This Avalon target port initiates read or write commands to the controller. Refer to
the Avalon Interface Specification for more information on how to design cores
that comply to the Avalon Bus Specification.
For DDR3, DDR4, and LPDDR3 protocols with the hard PHY and hard controller
configuration and an AVL slave interface exposed, ctrl_amm_avalon_slave is
renamed to crtl_amm_avalon_slave_0.
For QDR II, QDR II+, and QDR II+ Xtreme interfaces with hard PHY and soft
controller, separate read and write connections are used.
ctrl_amm_avalon_slave_0 is the read port and ctrl_amm_avalon_slave_1
is the write port.
For QDR-IV interfaces with hard PHY and soft controller operating at quarter rate,
a total of eight separate Avalon interfaces (named ctrl_amm_avalon_slave_0
to ctrl_amm_avalon_slave_7) are used to maximize bus efficiency.
External Memory Interface Handbook Volume 3: Reference Material
86
2 Functional Description—Intel Stratix® 10 EMIF IP
Related Links
Avalon Interface Specifications
2.6.2 File Sets
The Stratix 10 EMIF IP core generates four output file sets for every EMIF IP core,
arranged according to the following directory structure.
Table 25.
Generated File Sets
Directory
<core_name>/*
Description
This directory contains only the files required to integrate a
generated EMIF core into a larger design. This directory contains:
• Synthesizable HDL source files
• Customized TCL timing scripts specific to the core (protocol and
topology)
• HEX files used by the calibration algorithm to identify the
interface parameters
• A customized data sheet that describes the operation of the
generated core
Note: The top-level HDL file is generated in the root folder as
<core_name>.v (or <core_name.vhd> for VHDL designs).
You can reopen this file in the parameter editor if you want
to modify the EMIF core parameters and regenerate the
design.
<core_name>_sim/*
This directory contains the simulation fileset for the generated EMIF
core. These files can be integrated into a larger simulation project.
For convenience, simulation scripts for compiling the core are
provided in the /mentor, /cadence, /synopsys, and /riviera
subdirectories.
The top-level HDL file, <core_name>.v (or <core_name>.vhd) is
located in this folder, and all remaining HDL files are placed in the /
altera_emif_arch_nf subfolder, with the customized data sheet.
The contents of this directory are not intended for synthesis.
emif_<instance_num>_example_design/*
This directory contains a set of TCL scripts, QSYS project files and
README files for the complete synthesis and simulation example
designs. You can invoke these scripts to generate a standalone fullysynthesizable project complete with an example driver, or a
standalone simulation design complete with an example driver and a
memory model.
2.6.3 Customized readme.txt File
When you generate your Stratix 10 EMIF IP, the system produces a customized
readme file, containing data indicative of the settings in your IP core.
External Memory Interface Handbook Volume 3: Reference Material
87
2 Functional Description—Intel Stratix® 10 EMIF IP
The readme file is <variation_name>/
altera_emif_arch_nf_<version_number>/<synth|sim>/
<variation_name>_altera_emif_arch_nf_<version_number>_<unique
ID>_readme.txt, and contains a summary of Stratix 10 EMIF information, and
details specific to your IP core, including:
•
Pin location guidelines
•
External port names, directions, and widths
•
Internal port names, directions, and widths
•
Avalon interface configuration details (if applicable)
•
Calibration mode
•
A brief summary of all configuration settings for the generated IP
You should review the generated readme file for implementation guidelines specific to
your IP core.
2.6.4 Clock Domains
The Stratix 10 EMIF IP core provides a single clock domain to drive all logic connected
to the EMIF core.
The frequency of the clock depends on the core-clock to memory-clock interface rate
ratio. For example, a quarter-rate interface with an 800 MHz memory clock would
provide a 200 MHz clock to the core (800 MHz / 4 = 200 MHz). The EMIF IP
dynamically adjusts the phase of the core clock with respect to the periphery clock to
maintain the optimum alignment for transferring data between the core and periphery.
Independent EMIF IP cores driven from the same reference clock have independent
core clock domains. You should employ one of the following strategies if you are
implementing multiple EMIF cores:
1. Treat all crossing between independent EMIF-generated clock domains as
asynchronous, even though they are generated from the same reference clock.
2. Use the Core clock sharing option to enforce that multiple EMIF cores share the
same core clock. You must enable this option before IP generation. This option is
applicable only for cores that reside in the same I/O column.
2.6.5 ECC in Stratix 10 EMIF IP
The ECC (error correction code) is a soft component of the Stratix 10 EMIF IP that
reduces the chance of errors when reading and writing to external memory. ECC
allows correction of single-bit errors and reduces the chances of system failure.
The ECC component includes an encoder, decoder, write FIFO buffer, and modification
logic, to allow read-modify-write operations. The ECC code employs standard
Hamming logic to correct single-bit errors and to detect double-bit errors. ECC is
available in 16, 24, 40, and 72 bit widths.
When writing data to memory, the encoder creates ECC bits and writes them together
with the regular data. When reading from memory, the decoder checks the ECC bits
and regular data, and passes the regular data unchanged if no errors are detected. If
a single-bit error is detected, the ECC logic corrects the error and passes the regular
data. If more than a single-bit error is detected, the ECC logic sets a flag to indicate
the error.
External Memory Interface Handbook Volume 3: Reference Material
88
2 Functional Description—Intel Stratix® 10 EMIF IP
Read-modify-write operations can occur in the following circumstances:
•
A partial write in data mask mode (with or without ECC), where at least one
memory burst of byte-enable is not all ones or all zeros.
•
Auto-correction with ECC enabled. This is usually a dummy write issued by the
auto-correction logic when a single-bit error is detected.
Read-modify-write operations can have a significant effect on memory interface
performance; for best performance, you should minimize partial writes.
2.6.5.1 ECC Components
The ECC logic communicates with user logic via an Avalon-MM interface, and with the
hard memory controller via an Avalon-ST interface
ECC Encoder
The ECC encoder consists of a x64/x72 encoder IP core capable of single-bit error
correction and double-bit error detection. The encoder takes 64 bits input and
converts it to 72 bits output, where the 8 additional bits are ECC code. The encoder
supports any input data width less than 64-bits. Any unused input data bits are set to
zero.
ECC Decoder
The ECC decoder consists of a x72/x64 decoder IP core capable of double-bit error
detection. The decoder takes 72 bits input and converts it to 64 bits output. The
decoder also produces single-bit error and double-bit error information. The decoder
controls the user read data valid signal; when read data is intended for partial write,
the user read data valid signal is deasserted, because the read data is meant for
merging, not for the user.
Partial Write Data FIFO Buffer
The Partial Write Data FIFO Buffer is implemented in soft logic to store partial write
data and byte enable. Data and byte enable are popped and merged with the returned
read data. A partial write can occur in the following situations:
•
At least one memory burst of byte enable is not all ones or all zeroes.
•
Non data masked mode, where all memory bursts of byte enable are not all ones.
•
A dummy write with auto-correction logic, where all memory bursts of byte enable
are all zeroes. (You might use a dummy write when correcting memory content
with a single-bit error.)
Merging Logic
Merge return partial read data with write data based on byte enabled popped from the
FIFO buffer, and send it to the ECC encoder.
Pointer FIFO Buffer
The pointer FIFO buffer is implemented in soft logic to store write data pointers. The
ECC logic refers to the pointers when sending write data to the data buffer control
(DBC). The pointers serve to overwrite existing write data in the data buffer during a
read-modify-write process.
External Memory Interface Handbook Volume 3: Reference Material
89
2 Functional Description—Intel Stratix® 10 EMIF IP
Partial Logic
Partial logic decodes byte enable information and distinguishes between normal and
partial writes.
Memory Mode Register Interface
The Memory Mode Register (MMR) interface is an Avalon-based interface through
which core logic can access debug signals and sideband operation requests in the hard
memory controller.
The MMR logic routes ECC-related operations to an MMR register implemented in soft
logic, and returns the ECC information via an Avalon-MM interface. The MMR logic
tracks single-bit and double-bit error status, and provides the following information:
•
Interrupt status.
•
Single-bit error and double-bit error status.
•
Single-bit error and double-bit error counts, to a maximum of 15. (If more than
15 errors occur, the count will overflow.)
•
Address of the last error.
2.6.5.2 ECC User Interface Controls
You can enable the ECC logic from the Configuration, Status, and Error Handling
section of the Controller tab in the parameter editor.
There are three user interface settings related to ECC:
•
Enabled Memory-Mapped Configuration and Status Register (MMR)
Interface: Allows run-time configuration of the memory controller. You can
enable this option together with ECC to retrieve error detection information from
the ECC logic.
•
Enabled Error Detection and Correction Logic: Enables ECC logic for single-bit
error correction and double-bit error detection.
•
Enable Auto Error Correction: Allows the controller to automatically correct
single-bit errors detected by the ECC logic.
2.7 Examples of External Memory Interface Implementations for
DDR4
The following figures are examples of external memory interface implementations for
different DDR4 memory widths. The figures show the locations of the address/
command and data pins in relation to the locations of the memory controllers.
External Memory Interface Handbook Volume 3: Reference Material
90
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 42.
DDR4 1x8 Implementation Example (One I/O Bank)
1 x 8 Pin (1 Bank)
Data
Address/Command
Controller
Address/Command
Address/Command
Figure 43.
DDR4 1x32 Implementation Example (Two I/O Banks)
1 x 32 Pin (2 Banks)
Data
Data
Data
Data
Address/Command
Controller
Address/Command
Address/Command
External Memory Interface Handbook Volume 3: Reference Material
91
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 44.
DDR4 1x72 Implementation Example (Three I/O Banks)
1 x 72 Pin (3 Banks)
Data
Data
Data
Data
Data
Address/Command
Controller
Address/Command
Address/Command
Data
Data
Data
Data
External Memory Interface Handbook Volume 3: Reference Material
92
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 45.
DDR4 2x16 Implementation Example with Controllers in Non-Adjacent Banks
(Three I/O Banks)
2 x 16 Pin (3 Banks)
Controller
1
Address/Command 1
Address/Command 1
Address/Command 1
Data 1
Data 1
Data 2
Data 2
Controller
2
Address/Command 2
Address/Command 2
Address/Command 2
External Memory Interface Handbook Volume 3: Reference Material
93
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 46.
DDR4 2x16 Implementation Example with Controllers in Adjacent Banks
(Three I/O Banks)
2 x 16 Pin (3 Banks)
Data 1
Controller
1
Address/Command 1
Address/Command 1
Address/Command 1
Data 1
Controller
2
Address/Command 2
Address/Command 2
Address/Command 2
Data 2
Data 2
External Memory Interface Handbook Volume 3: Reference Material
94
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 47.
DDR4 1x144 Implementation Example (Six I/O Banks)
1 x 144 Pin (6 Banks)
Data
Data
Data
Data
Data
Data
Data
Data
Data
Data
Address/Command
Controller
Address/Command
Address/Command
Data
Data
Data
Data
Data
Data
Data
Data
2.8 Stratix 10 EMIF Sequencer
The Stratix 10 EMIF sequencer is fully hardened in silicon, with executable code to
handle protocols and topologies. Hardened RAM contains the calibration algorithm.
External Memory Interface Handbook Volume 3: Reference Material
95
2 Functional Description—Intel Stratix® 10 EMIF IP
The Stratix 10 EMIF sequencer is responsible for the following operations:
Figure 48.
•
Initializes memory devices.
•
Calibrates the external memory interface.
•
Governs the hand-off of control to the memory controller.
•
Handles recalibration requests and debug requests.
•
Handles all supported protocols and configurations.
Stratix 10 EMIF Sequencer Operation
Start
Discover EMIFs in column
Sequencer
software
Processed all
interfaces?
Data
Yes
No
Initialize external memory
Calibrate interface
Hand-off
House-keeping
tasks
2.8.1 Stratix 10 EMIF DQS Tracking
DQS tracking is enabled for QDR II / II+ / QDR II+ Xtreme, RLDRAM 3, and LPDDR3
protocols. DQS tracking is not available for DDR3 and DDR4 protocols.
2.9 Stratix 10 EMIF Calibration
The calibration process compensates for skews and delays in the external memory
interface.
The following effects can be compensated for by the calibration process:
Note:
•
Timing and electrical constraints, such as setup/hold time and Vref variations.
•
Circuit board and package factors, such as skew, fly-by effects, and manufacturing
variations.
•
Environmental uncertainties, such as variations in voltage and temperature.
•
The demanding effects of small margins associated with high-speed operation.
The calibration process is intended to maximize margins for robust EMIF operation; it
cannot compensate for an inadequate PCB layout.
External Memory Interface Handbook Volume 3: Reference Material
96
2 Functional Description—Intel Stratix® 10 EMIF IP
2.9.1 Stratix 10 Calibration Stages
At a high level, the calibration routine consists of address and command calibration,
read calibration, and write calibration.
The stages of calibration vary, depending on the protocol of the external memory
interface.
Table 26.
Calibration Stages by Protocol
Stage
DDR4
DDR3
LPDDR3
RLDRAM 3
QDR-IV
QDR II/II+
Address and command
Leveling
Yes
Yes
—
—
—
—
Deskew
Yes
—
Yes
—
Yes
—
DQSen
Yes
Yes
Yes
Yes
Yes
Yes
Deskew
Yes
Yes
Yes
Yes
Yes
Yes
VREF-In
Yes
—
—
—
Yes
—
LFIFO
Yes
Yes
Yes
Yes
Yes
Yes
Leveling
Yes
Yes
Yes
Yes
Yes
—
Deskew
Yes
Yes
Yes
Yes
Yes
Yes
VREF-Out
Yes
—
—
—
—
—
Read
Write
2.9.2 Stratix 10 Calibration Stages Descriptions
The various stages of calibration perform address and command calibration, read
calibration, and write calibration.
Address and Command Calibration
The goal of address and command calibration is to delay address and command
signals as necessary to optimize the address and command window. This stage is not
available for all protocols, and cannot compensate for an inefficient board design.
Address and command calibration consists of the following parts:
•
Leveling calibration— Centers the CS# signal and the entire address and
command bus, relative to the CK clock. This operation is available for DDR3 and
DDR4 interfaces only.
•
Deskew calibration— Provides per-bit deskew for the address and command bus
(except CS#), relative to the CK clock. This operation is available for DDR4 and
QDR-IV interfaces only.
External Memory Interface Handbook Volume 3: Reference Material
97
2 Functional Description—Intel Stratix® 10 EMIF IP
Read Calibration
Read calibration consists of the following parts:
•
DQSen calibration— Calibrates the timing of the read capture clock gating and
ungating, so that the PHY can gate and ungate the read clock at precisely the
correct time—if too early or too late, data corruption can occur. The algorithm for
this stage varies, depending on the memory protocol.
•
Deskew calibration— Performs per-bit deskew of read data relative to the read
strobe or clock.
•
VREF-In calibration— Calibrates the VREF level at the FPGA.
•
LFIFO calibration: Normalizes differences in read delays between groups due to
fly-by, skews, and other variables and uncertainties.
Write Calibration
Write calibration consists of the following parts:
•
Leveling calibration— Aligns the write strobe and clock to the memory clock, to
compensate for skews, especially those associated with fly-by topology. The
algorithm for this stage varies, depending on the memory protocol.
•
Deskew calibration— Performs per-bit deskew of write data relative to the write
strobe and clock.
•
VREF-Out calibration— Calibrates the VREF level at the memory device.
2.9.3 Stratix 10 Calibration Algorithms
The calibration algorithms sometimes vary, depending on the targeted memory
protocol.
Address and Command Calibration
Address and command calibration consists of the following parts:
External Memory Interface Handbook Volume 3: Reference Material
98
2 Functional Description—Intel Stratix® 10 EMIF IP
•
Leveling calibration— (DDR3 and DDR4 only) Toggles the CS# and CAS# signals
to send read commands while keeping other address and command signals
constant. The algorithm monitors for incoming DQS signals, and if the DQS signal
toggles, it indicates that the read commands have been accepted. The algorithm
then repeats using different delay values, to find the optimal window.
•
Deskew calibration— (DDR4, QDR-IV, and LPDDR3 only)
—
(DDR4) Uses the DDR4 address and command parity feature. The FPGA sends
the address and command parity bit, and the DDR4 memory device responds
with an alert signal if the parity bit is detected. The alert signal from the
memory device tells the FPGA that the parity bit was received.
Deskew calibration requires use of the PAR/ALERT# pins, so you should not
omit these pins from your design. One limitation of deskew calibration is that
it cannot deskew ODT and CKE pins.
—
(QDR-IV) Uses the QDR-IV loopback mode. The FPGA sends address and
command signals, and the memory device sends back the address and
command signals which it captures, via the read data pins. The returned
signals indicate to the FPGA what the memory device has captured. Deskew
calibration can deskew all synchronous address and command signals.
—
(LPDDR3) Uses the LPDDR3 CA training mode. The FPGA sends signals onto
the LPDDR3 CA bus, and the memory device sends back those signals that it
captures, via the DQ pins. The returned signals indicate to the FPGA what the
memory device has captured. Deskew calibration can deskew all signals on the
CA bus. The remaining command signals (CS, CKE, and ODT) are calibrated
based on the average of the deskewed CA bus.
External Memory Interface Handbook Volume 3: Reference Material
99
2 Functional Description—Intel Stratix® 10 EMIF IP
Read Calibration
•
DQSen calibration— (DDR3, DDR4, LPDDR3, RLDRAMx and QDRx) DQSen
calibration occurs before Read deskew, therefore only a single DQ bit is required to
pass in order to achieve a successful read pass.
—
(DDR3, DDR4,and LPDDR3) The DQSen calibration algorithm searches the
DQS preamble using a hardware state machine. The algorithm sends many
back-to-back reads with a one clock cycle gap between. The hardware state
machine searches for the DQS gap while sweeping DQSen delay values. the
algorithm then increments the VFIFO value, and repeats the process until a
pattern is found. The process is then repeated for all other read DQS groups.
—
(RLDRAMx and QDRx) The DQSen calibration algorithm does not use a
hardware state machine; rather, it calibrates cycle-level delays using software
and subcycle delays using DQS tracking hardware. The algorithm requires
good data in memory, and therefore relies on guaranteed writes. (Writing a
burst of 0s to one location, and a burst of 1s to another; back-to-back reads
from these two locations are used for read calibration.)
The algorithm enables DQS tracking to calibrate the phase component of DQS
enable, and then issues a guaranteed write, followed by back-to-back reads.
The algorithm sweeps DQSen values cycle by cycle until the read operation
succeeds. The process is then repeated for all other read groups.
•
Deskew calibration— Read deskew calibration is performed before write leveling,
and must be performed at least twice: once before write calibration, using simple
data patterns from guaranteed writes, and again after write calibration, using
complex data patterns.
The deskew calibration algorithm performs a guaranteed write, and then sweeps
dqs_in delay values from low to high, to find the right edge of the read window.
The algorithm then sweeps dq-in delay values low to high, to find the left edge of
the read window. Updated dqs_in and dq_in delay values are then applied to
center the read window. The algorithm then repeats the process for all data pins.
•
Vref-In calibration— Read Vref-In calibration begins by programming Vref-In
with an arbitrary value. The algorithm then sweeps the Vref-In value from the
starting value to both ends, and measures the read window for each value. The
algorithm selects the Vref-In value which provides the maximum read window.
•
LFIFO calibration— Read LFIFO calibration normalizes read delays between groups.
The PHY must present all data to the controller as a single data bus. The LFIFO
latency should be large enough for the slowest read data group, and large enough
to allow proper synchronization across FIFOs.
External Memory Interface Handbook Volume 3: Reference Material
100
2 Functional Description—Intel Stratix® 10 EMIF IP
Write Calibration
•
Leveling calibration— Write leveling calibration aligns the write strobe and clock to
the memory clock, to compensate for skews. In general, leveling calibration tries a
variety of delay values to determine the edges of the write window, and then
selects an appropriate value to center the window. The details of the algorithm
vary, depending on the memory protocol.
—
(DDRx, LPDDR3) Write leveling occurs before write deskew, therefore only one
successful DQ bit is required to register a pass. Write leveling staggers the DQ
bus to ensure that at least one DQ bit falls within the valid write window.
—
(RLDRAMx) Optimizes for the CK versus DK relationship.
—
(QDR-IV) Optimizes for the CK versus DK relationship. Is covered by address
and command deskew using the loopback mode.
—
(QDR II/II+/Xtreme) The K clock is the only clock, therefore write leveling is
not required.
•
Deskew calibration— Performs per-bit deskew of write data relative to the write
strobe and clock. Write deskew calibration does not change dqs_out delays; the
write clock is aligned to the CK clock during write leveling.
•
VREF-Out calibration— (DDR4) Calibrates the VREF level at the memory device.
The VREF-Out calibration algorithm is similar to the VREF-In calibration algorithm.
2.9.4 Stratix 10 Calibration Flowchart
The following flowchart illustrates the Stratix 10 calibration flow.
Figure 49.
Calibration Flowchart
External Memory Interface Handbook Volume 3: Reference Material
101
2 Functional Description—Intel Stratix® 10 EMIF IP
2.10 Stratix 10 EMIF and SmartVID
Stratix 10 EMIF IP can be used with the SmartVID voltage management system, to
achieve reduced power consumption.
The SmartVID controller allows the FPGA to operate at a reduced Vcc, while
maintaining performance. Because the SmartVID controller can adjust Vcc up or down
in response to power requirements and temperature, it can have an impact on
external memory interface performance. When used with the SmartVID controller, the
EMIF IP implements a handshake protocol to ensure that EMIF calibration does not
begin until after voltage adjustment has completed.
In extended speed grade devices, voltage adjustment occurs once when the FPGA is
powered up, and no further voltage adjustments occur. The external memory
calibration occurs after this initial voltage adjustment is completed. EMIF specifications
are expected to be slightly lower in extended speed grade devices using SmartVID,
than in devices not using SmartVID.
In industrial speed grade devices, voltage adjustment occurs at power up, and may
also occur during operation, in response to temperature changes. External memory
interface calibration does not occur until after the initial voltage adjustment at power
up. However, the external memory interface is not recalibrated in response to
subsequent voltage adjustments that occur during operation. As a result, EMIF
specifications for industrial speed grade devices using SmartVID are expected to be
lower than for extended speed grade devices.
Using Stratix 10 EMIF IP with SmartVID
To employ Stratix 10 EMIF IP with SmartVID, follow these steps:
1.
Ensure that the Quartus Prime project and Qsys system are configured to use VID
components. This step exposes the vid_cal_done_persist interface on
instantiated EMIF IP, which is required for communicating with the SmartVID
controller.
2.
Instantiate the SmartVID controller, using an I/O PLL IP core to drive the 125MHz
vid_clk and the 25MHz jtag_core_clk inputs of the Smart VID controller.
Note: Do not connect the emif_usr_clk signal to either the vid_clk or
jtag_core_clk inputs. Doing so would hold both the EMIF IP and the
SmartVID controller in a perpetual reset condition.
3.
Instantiate the Stratix 10 EMIF IP.
4.
Connect the vid_cal_done_persist signal from the EMIF IP with the
cal_done_persistent signal on the SmartVID controller. This connection
enables handshaking between the EMIF IP and the SmartVID controller, which
allows the EMIF IP to delay memory calibration until after voltage levels are
stabilized.
Note: The EMIF vid_cal_done_persist interface becomes available only when
a VID-enabled device is selected.
2.11 Stratix 10 Hard Memory Controller Rate Conversion Feature
The hard memory controller's rate conversion feature allows the hard memory
controller and PHY to run at half-rate, even though user logic is configured to run at
quarter-rate.
External Memory Interface Handbook Volume 3: Reference Material
102
2 Functional Description—Intel Stratix® 10 EMIF IP
To facilitate timing closure, you may choose to clock your core user logic at quarterrate, resulting in easier timing closure at the expense of increased area and latency.
To improve efficiency and help reduce overall latency, you can run the hard memory
controller and PHY at half rate.
The rate conversion feature converts traffic from the FPGA core to the hard memory
controller from quarter-rate to half-rate, and traffic from the hard memory controller
to the FPGA core from half-rate to quarter-rate. From the perspective of user logic
inside the FPGA core, the effect is the same as if the hard memory controller were
running at quarter-rate.
The rate conversion feature is enabled automatically during IP generation whenever all
of the following conditions are met:
•
The hard memory controller is in use.
•
User logic runs at quarter-rate.
•
The interface targets either an ES2 or production device.
•
Running the hard memory controller at half-rate dpoes not exceed the fMax
specification of the hard memory controller and hard PHY.
When the rate conversion feature is enabled, you should see the following info
message displayed in the IP generation GUI:
PHY and controller running at 2x the frequency of user logic for
improved efficiency.
2.12 Differences Between User-Requested Reset in Stratix 10
versus Arria 10
The following table highlights differences between the user-requested reset
mechanism in the Arria 10 EMIF IP and the Stratix 10 EMIF IP.
Table 27.
Arria 10
Stratix 10
Reset-related signals
global_reset_n
local_reset_req
local_reset_done
When can user logic request a reset?
Any time after the FPGA enters user
mode.
local_reset_req has effect only
local_reset_done is high.
After device power-on, the
local_reset_done signal transitions
high upon completion of the first
calibration, whether the calibration is
successful or not.
Is user-requested reset a requirement?
A user-requested reset is typically
required to ensure the memory
interface begins from a known state.
A user-requested reset is optional. The
IOSSM (which is part of the device’s
CNOC) automatically ensures that the
memory interface begins from a known
state as part of the device power-on
sequence. A user-requested reset is
necessarily only if the user logic must
explicitly reset a memory interface
after the device power-on sequence.
continued...
External Memory Interface Handbook Volume 3: Reference Material
103
2 Functional Description—Intel Stratix® 10 EMIF IP
Arria 10
Stratix 10
When does a user-requested reset
actually happen?
As soon as global_reset_n is driven
low by user logic.
A reset request is handled by the
IOSSM. If the IOSSM receives a reset
request from multiple interfaces within
the same I/O column, it must serialize
the reset sequence of the individual
interfaces. You should avoid making
assumptions on when the reset
sequence will begin after a request is
issued.
Timing requirement and triggering
mechanism.
global_reset_n is an asynchronous,
Reset request is sent by transitioning
the local_reset_req signal from
low to high, then keeping the signal at
the high state for a minimum of 2 EMIF
core clock cycles, then transitioning the
signal from high to low.
local_reset_req is asynchronous in
that there is no setup/hold timing to
meet, but it must meet the minimum
pulse width requirement of 2 EMIF core
clock cycles.
How long can an external memory
interface be kept in reset?
The interface is kept in reset for as
long as global_reset_n is driven
low.
It is not possible to keep an external
memory interface in reset indefinitely.
Asserting local_reset_req high
continuously has no effect as a reset
request is completed by a full 0->1->0
pulse.
Delaying initial calibration.
Initial calibration can be delayed for as
long as desired by driving
global_reset_n immediately after
FPGA power-up.
Initial calibration cannot be skipped.
The local_reset_done signal is
driven high only after initial calibration
has completed.
Reset scope (within an external
memory interface).
All circuits involved in EMIF operations
are reset.
Only circuits that are required to
restore EMIF to power-up state are
reset. Excluded from the reset
sequence are the IOSSM, the IOPLL(s),
the DLL(s), and the CPA.
Reset scope (within an I/O column).
global_reset_n is a column-wide
local_reset_req is a per-interface
reset. It is not possible to reset a
subset of the memory interfaces within
an I/O column.
reset.
active-low reset signal. Reset assertion
and de-assertion is level-triggered.
2.12.1 Method for Initiating a User-requested Reset
Step 1 - Precondition
Before asserting local_reset_req, user logic must ensure that the
local_reset_done signal is high.
As part of the device power-on sequence, the local_reset_done signal
automatically transitions to high upon the completion of the interface calibration
sequence, regardless of whether calibration is successful or not.
Note:
When targeting a group of interfaces that share the same core clocks, user logic must
ensure that the local_reset_done signal of every interface is high.
External Memory Interface Handbook Volume 3: Reference Material
104
2 Functional Description—Intel Stratix® 10 EMIF IP
Step 2 - Reset Request
After the pre-condition is satisfied, user logic can send a reset request by driving the
local_cal_req signal from low to high and then low again (that is, by sending a
pulse of 1).
•
The 0-to-1 and 1-to-0 transitions need not happen in relation to any clock edges
(that is, they can occur asynchronously); however, the pulse must meet a
minimum pulse width of at least 2 EMIF core clock cycles. For example, if the
emif_usr_clk has a period of 4ns, then the local_reset_req pulse must last
at least 8ns (that is, two emif_usr_clk periods).
•
The reset request is considered complete only after the 1-to-0 transition. The EMIF
IP does not initiate the reset sequence when the local_reset_req is simply
held high.
•
Additional pulses to local_reset_req are ignored until the reset sequence is
completed.
Optional - Detecting local_reset_done deassertion and assertion
If you want, you can monitor the status of the local_reset_done signal to to
explicitly detect the status of the reset sequence.
•
After the EMIF IP receives a reset request, it deasserts the local_reset_done
signal. After initial power-up calibration, local_reset_done is de-asserted only
in response to a user-requested reset. The reset sequence is imminent when
local_reset_done has transitioned to low, although the exact timing depends
on the current state of the I/O Subsystem Manager (IOSSM). As part of the EMIF
reset sequence, the core reset signal (emif_usr_reset_n, afi_reset_n) is
driven low. Do not use a register reset by the core reset signal to sample
local_reset_done.
•
After the reset sequence has completed, local_reset_done is driven high
again. local_reset_done being driven high indicates the completion of the
reset sequence and the readiness to accept a new reset request; however, it does
not imply that calibration was successful or that the hard memory controller is
ready to accept requests. For these purposes, user logic must check signals such
as afi_cal_success, afi_cal_fail, and amm_ready.
2.13 Compiling Stratix 10 EMIF IP with the Quartus Prime Software
2.13.1 Instantiating the Stratix 10 EMIF IP
Depending on your work flow, you may instantiate your IP with Qsys or with the IP
Catalog.
Instantiating with Qsys
If you instantiate your IP as part of a system in Qsys, follow the Qsys documentation
for information on instantiating the IP in a Quartus Prime project.
External Memory Interface Handbook Volume 3: Reference Material
105
2 Functional Description—Intel Stratix® 10 EMIF IP
Instantiating with the IP Catalog
If you generated your IP with the IP Catalog, you must add the Quartus Prime IP file
(.qip) to your Quartus Prime project. The .qip file identifies the names and
locations of the files that compose the IP. After you add the .qip file to your project,
you can instantiate the memory interface in the RTL.
2.13.2 Setting I/O Assignments in Stratix 10 EMIF IP
The .qip file contains the I/O standard and I/O termination assignments required by
the memory interface pins for proper operation. The assignment values are based on
input that you provide during generation.
Unlike earlier device families, for Stratix 10 EMIF IP you do not need to run a
<instance_name)_pin_assignments.tcl script to add the assignments into the
Quartus Prime Settings File (.qsf). The system reads and applies the assignments
from the .qip file during every compilation, regardless of how you name the memory
interface pins in the top-level design component. No new assignments are created in
the project's .qsf file during compilation.
Note that I/O assignments in the .qsf file must specify the names of your top-level
pins as target (-to), and you must not include the -entity or -library options.
Consult the generated .qip file for the set of I/O assignments that are provided with
the IP.
Changing I/O Assignments
You should not make changes to the generated .qip file, because any changes are
overwritten and lost when you regenerate the IP. If you want to override an
assignment made in the .qip file, add the desired assignment to the project's .qsf
file. Assignments in the .qsf file always take precedence over assignments in
the .qip file.
2.14 Debugging Stratix 10 EMIF IP
You can debug hardware failures by connecting to the EMIF Debug Toolkit or by
exporting an Avalon-MM slave port, from which you can access information gathered
during calibration. You can also connect to this port to mask ranks and to request
recalibration.
You can access the exported Avalon-MM port in two ways:
•
Via the External Memory Interface Debug Toolkit
•
Via On-Chip Debug (core logic on the FPGA)
2.14.1 External Memory Interface Debug Toolkit
The External Memory Interface Debug Toolkit provides access to data collected by the
Nios II sequencer during memory calibration, and allows you to perform certain tasks.
External Memory Interface Handbook Volume 3: Reference Material
106
2 Functional Description—Intel Stratix® 10 EMIF IP
The External Memory Interface Debug Toolkit provides access to data including the
following:
•
General interface information, such as protocol and interface width
•
Calibration results per group, including pass/fail status, failure stage, and delay
settings
You can also perform the following tasks:
•
Mask ranks from calibration (you might do this to skip specific ranks)
•
Request recalibration of the interface
2.14.2 On-Chip Debug for Stratix 10
The On-Chip Debug feature allows user logic to access the same debug capabilities as
the External Memory Interface Toolkit. You can use On-Chip Debug to monitor the
calibration results of an external memory interface, without a connected computer.
To use On-Chip Debug, you need a C header file which is provided as part of the
external memory interface IP. The C header file defines data structures that contain
calibration data, and definitions of the commands that can be sent to the memory
interface.
The On-Chip Debug feature accesses the data structures through the Avalon-MM port
that is exposed by the EMIF IP when you turn on debugging features.
2.14.3 Configuring Your EMIF IP for Use with the Debug Toolkit
The Stratix 10 EMIF Debug Interface IP core contains the access point through which
the EMIF Debug Toolkit reads calibration data collected by the Nios II sequencer.
Connecting an EMIF IP Core to a Stratix 10 EMIF Debug Interface
For the EMIF Debug Toolkit to access the calibration data for a Stratix 10 EMIF IP core,
you must connect one of the EMIF cores in each I/O column to a Stratix 10 EMIF
Debug Interface IP core. Subsequent EMIF IP cores in the same column must connect
in a daisy chain to the first.
There are two ways that you can add the Stratix 10 EMIF Debug Interface IP core to
your design:
•
When you generate your EMIF IP core, on the Diagnostics tab, select Add EMIF
Debug Interface for the EMIF Debug Toolkit/On-Chip Debug Port; you do
not have to separately instantiate a Stratix 10 EMIF Debug Interface core. This
method does not export an Avalon-MM slave port. You can use this method if you
require only EMIF Debug Toolkit access to this I/O column; that is, if you do not
require On-Chip Debug Port access, or PHYLite reconfiguration access.
•
When you generate your EMIF IP core, on the Diagnostics tab, select Export for
the EMIF Debug Toolkit/On-Chip Debug Port. Then, separately instantiate a
Stratix 10 EMIF Debug Interface core and connect its to_iossm interface to the
cal_debug interface on the EMIF IP core. This method is appropriate if you want
to also have On-Chip Debug Port access to this I/O column, or PHYLite
reconfiguration access.
External Memory Interface Handbook Volume 3: Reference Material
107
2 Functional Description—Intel Stratix® 10 EMIF IP
For each of the above methods, you must assign a unique interface ID for each
external memory interface in the I/O column, to identify that interface in the Debug
Toolkit. You can assign an interface ID using the dropdown list that appears when you
enable the Debug Toolkit/On-Chip Debug Port option.
Daisy-Chaining Additional EMIF IP Cores for Debugging
After you have connected a Stratix 10 EMIF Debug Interface to one of the EMIF IP
cores in an I/O column, you must then connect subsequent EMIF IP cores in that
column in a daisy-chain manner. If you don't require debug capabilities for a particular
EMIF IP core, you do not have to connect that core to the daisy chain.
To create a daisy chain of EMIF IP cores, follow these steps:
1.
On the first EMIF IP core, select Enable Daisy-Chaining for EMIF Debug
Toolkit/On-Chip Debug Port to create an Avalon-MM interface called
cal_debug_out.
2. On the second EMIF IP core, select Export as the EMIF Debug Toolkit/On-Chip
Debug Port mode, to export an Avalon-MM interface called cal_debug.
3. Connect the cal_debug_out interface of the first EMIF core to the cal_debug
interface of the second EMIF core.
4. To connect more EMIF cores to the daisy chain, select the Enable DaisyChaining for EMIF Debug Toolkit/On-Chip Debug Port option on the second
core, connect it to the next core using the Export option as described above.
Repeat the process for subsequent EMIF cores.
If you place any PHYLite cores with dynamic reconfiguration enabled into the same I/O
column as an EMIF IP core, you should instantiate and connect the PHYLite cores in a
similar way. See the Altera PHYLite for Memory Megafunction User Guide for more
information.
Related Links
Altera PHYLite for Parallel Interfaces IP Core User Guide
2.14.4 Stratix 10 EMIF Debugging Examples
The following sections provide examples of debugging a single external memory
interface, and of adding additional EMIF instances to an I/O column.
Debugging a Single External Memory Interface
1.
Under EMIF Debug Toolkit/On-Chip Debug Port, select Add EMIF Debug
Interface.
(If you want to use the On-Chip Debug Port instead of the EMIF Debug Toolkit,
select Export instead.)
External Memory Interface Handbook Volume 3: Reference Material
108
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 50.
EMIF With Debug Interface Added (No Additional Ports)
emif_0
global_reset_reset_sink
pll_ref_clk_clock_sink
oct_conduit_end
mem_conduit_end
status_conduit_end
ctrl_amm_avalon_slave_0
2.
Figure 51.
emif_usr_clk_clock_source
emif_usr_reset_reset_source
emif
If you want to connect additional EMIF or PHYLite components in this I/O column,
select Enable Daisy Chaining for EMIF Debug Toolkit/On-Chip Debug Port.
EMIF With cal_debug Avalon Master Exported
emif_0
global_reset_reset_sink
pll_ref_clk_clock_sink
oct_conduit_end
mem_conduit_end
status_conduit_end
ctrl_amm_avalon_slave_0
emif_usr_clk_clock_source
emif_usr_reset_reset_source
cal_debug_out_reset_reset_source
cal_debug_out_clk_clock_source
cal_debug_out_avalon_master
emif
cal_debug Avalon
Master Exported
Adding Additional EMIF Instances to an I/O Column
1.
Figure 52.
Under EMIF Debug Toolkit/On-Chip Debug Port, select Export.
EMIF With cal_debug Avalon Slave Exported
emif_1
cal_debug Avalon
Slave Exported
global_reset_reset_sink
pll_ref_clk_clock_sink
oct_conduit_end
mem_conduit_end
status_conduit_end
cal_debug_reset_reset_sink
cal_debug_clk_clock_sink
ctrl_amm_avalon_slave_0
cal_debug_avalon_slave
emif_usr_clk_clock_source
emif_usr_reset_reset_source
emif
2. Specify a unique interface ID for this EMIF instance.
3. If you want to connect additional EMIF or PHYLite components in this I/O column,
select Enable Daisy Chaining for EMIF Debug Toolkit/On-Chip Debug Port.
External Memory Interface Handbook Volume 3: Reference Material
109
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 53.
EMIF With Both cal_debug Master and Slave Exported
emif_1
cal_debug Avalon
Slave Exported
4.
Figure 54.
global_reset_reset_sink
pll_ref_clk_clock_sink
oct_conduit_end
mem_conduit_end
status_conduit_end
cal_debug_reset_reset_sink
cal_debug_clk_clock_sink
ctrl_amm_avalon_slave_0
cal_debug_avalon_slave
emif_usr_clk_clock_source
emif_usr_reset_reset_source
cal_debug_out_reset_reset_source
cal_debug_out_clk_clock_source
cal_debug_out_avalon_master
cal_debug Avalon
Master Exported
emif
Connect the cal_debug Avalon Master, clock, and reset interfaces of the previous
component to the cal_debug Avalon Slave, clock, and reset interfaces of this
component.
EMIF Components Connected
emif_1
emif_0
global_reset_reset_sink
pll_ref_clk_clock_sink
oct_conduit_end
mem_conduit_end
status_conduit_end
ctrl_amm_avalon_slave_0
emif_usr_clk_clock_source
emif_usr_reset_reset_source
cal_debug_out_reset_reset_source
cal_debug_out_clk_clock_source
cal_debug_out_avalon_master
emif
global_reset_reset_sink
pll_ref_clk_clock_sink
oct_conduit_end
mem_conduit_end
status_conduit_end
cal_debug_reset_reset_sink
cal_debug_clk_clock_sink
ctrl_amm_avalon_slave_0
cal_debug_avalon_slave
emif_usr_clk_clock_source
emif_usr_reset_reset_source
cal_debug_out_reset_reset_source
cal_debug_out_clk_clock_source
cal_debug_out_avalon_master
emif
2.15 Stratix 10 EMIF for Hard Processor Subsystem
The Stratix 10 EMIF IP can enable the Stratix 10 Hard Processor Subsystem (HPS) to
access external DRAM memory devices.
To enable connectivity between the Stratix 10 HPS and the Stratix 10 EMIF IP, you
must create and configure an instance of the Stratix 10 External Memory Interface for
HPS IP core, and use Qsys to connect it to the Stratix 10 Hard Processor Subsystem
instance in your system.
Supported Modes
The Stratix 10 Hard Processor Subsystem is compatible with the following external
memory configurations:
Protocol
DDR3, DDR4, LPDDR3
Maximum memory clock frequency
DDR3: 1.067 GHz
DDR4: 1.333 GHz
LPDDR3: 800 MHz
Configuration
Hard PHY with hard memory controller
Clock rate of PHY and hard memory controller
Half-rate
Data width (without ECC)
16-bit, 32-bit, 64-bit
Data width (with ECC)
24-bit, 40-bit, 72-bit
DQ width per group
x8
continued...
External Memory Interface Handbook Volume 3: Reference Material
110
2 Functional Description—Intel Stratix® 10 EMIF IP
Maximum number of I/O lanes for address/command
3
Memory format
Discrete, UDIMM, SODIMM, RDIMM
Ranks / CS# width
Up to 2
2.15.1 Restrictions on I/O Bank Usage for Stratix 10 EMIF IP with HPS
You can use only certain Stratix 10 I/O banks to implement Stratix 10 EMIF IP with
the Stratix 10 Hard Processor Subsystem.
External Memory Interface Handbook Volume 3: Reference Material
111
2 Functional Description—Intel Stratix® 10 EMIF IP
The restrictions on I/O bank usage result from the Stratix 10 HPS having hard-wired
connections to the EMIF circuits in the I/O banks closest to the HPS. For any given
EMIF configuration, the pin-out of the EMIF-to-HPS interface is fixed.
The following diagram illustrates the use of I/O banks and lanes for various EMIF-HPS
data widths:
16 bit, no ECC
16 bit, with ECC
32 bit, no ECC
32 bit, with ECC
64 bit, no ECC
Stratix 10 External Memory Interfaces I/O Bank and Lanes Usage
64 bit, with ECC
Figure 55.
ECC
8
bits
Data
32
bits
Data
32
bits
Data
16
bits
Data
16
bits
Lane 0
Lane 3
ECC
8
bits
ECC
8
bits
Lane 1
Lane 2
Addr
/ cmd
Addr
/ cmd
Addr
/ cmd
Addr
/ cmd
Addr
/ cmd
Addr
/ cmd
Lane 1
Lane 0
Lane 3
Data
32
bits
Data
32
bits
Lane 2
Lane 1
Lane 0
External Memory Interface Handbook Volume 3: Reference Material
112
I/O Bank 2M
(Addr/Cmd + ECC data)
Data
32
bits
I/O Bank 2L
(Data bits 63:32)
Lane 2
Data
32
bits
I/O Bank 2N
(Data bits 31:0)
Lane 3
HPS
2 Functional Description—Intel Stratix® 10 EMIF IP
The HPS EMIF uses the closest located external memory interfaces I/O banks to
connect to SDRAM. These banks include:
•
Bank 2N—used for data I/Os (Data bits 31:0)
•
Bank 2M—used for address, command and ECC data I/Os
•
Bank 2L—used for data I/Os (Data bits 63:32)
If no HPS EMIF is used in a system, the entire HPS EMIF bank can be used as FPGA
GPIO. If there is a HPS EMIF in a system, the unused HPS EMIF pins can be used as
FPGA general I/O with restrictions:
•
•
Bank 2M:
—
Lane 3 is used for SDRAM ECC data. Unused pins in lane 3 can be used as
FPGA inputs only.
—
Lanes 2, 1, and 0 are used for SDRAM address and command. Unused pins in
these lanes can be used as FPGA inputs or outputs.
Bank 2N and Bank 2L :
—
Lanes 3, 2, 1, and 0 are used for data bits.
—
With 64-bit data widths, unused pins in these banks can be used as FPGA
inputs only.
—
With 32-bit data widths, unused pins in Bank 2N can be used as FPGA inputs
only.Unused pins for Bank 2L can be used as FPGA inputs or outputs.
—
With 16-bit data widths, Quartus® Prime assigns lane 0 and lane 1 as data
lanes in bank 2N. Unused pins in lane 0 and lane 1 can be used as FPGA
inputs only. The other two lanes are available to use as FPGA inputs or
outputs.
By default, the Stratix 10 External Memory Interface for HPS IP core together with the
Quartus Prime Fitter automatically implement the correct pin-out for HPS EMIF without
you having to apply additional constraints. If you must modify the default pin-out for
any reason, you must adhere to the following requirements, which are specific to HPS
EMIF:
1. Within a single data lane (which implements a single x8 DQS group):
•
DQ pins must use pins at indices 1, 2, 3, 6, 7, 8, 9, 10. You may swap the
locations between the DQ bits (that is, you may swap location of DQ[0] and
DQ[3]) so long as the resulting pin-out uses pins at these indices only.
•
DM/DBI pin must use pin at index 11. There is no flexibility.
•
DQS/DQS# must use pins at index 4 and 5. There is no flexibility.
2.
Assignment of data lanes must be as illustrated in the above figure. You are
allowed to swap the locations of entire byte lanes (that is, you may swap locations
of byte 0 and byte 3) so long as the resulting pin-out uses only the lanes
permitted by your HPS EMIF configuration, as shown in the above figure.
3.
You must not change placement of the address and command pins from the
default.
4. You may place the alert# pin at any available pin location in either a data lane or
an address and command lane.
External Memory Interface Handbook Volume 3: Reference Material
113
2 Functional Description—Intel Stratix® 10 EMIF IP
To override the default generated pin assignments, comment out the relevant
HPS_LOCATION assignments in the .qip file, and add your own location assignments
(using set_location_assignment) in the .qsf file.
2.16 Stratix 10 EMIF Ping Pong PHY
Ping Pong PHY allows two memory interfaces to share the address and command bus
through time multiplexing. Compared to having two independent interfaces that
allocate address and command lanes separately, Ping Pong PHY achieves the same
throughput with fewer resources, by sharing the address and command lanes.
In Stratix 10 EMIF, Ping Pong PHY supports both half-rate and quarter-rate interfaces
for DDR3, and quarter-rate for DDR4.
2.16.1 Stratix 10 Ping Pong PHY Feature Description
Conventionally, the address and command buses of a DDR3 or DDR4 half-rate or
quarter-rate interface use 2T time—meaning that commands are issued for two fullrate clock cycles, as illustrated below.
Figure 56.
2T Command Timing
CK
CSn
Addr, ba
Extra Setup Time
2T Command Issued
Active Period
With the Ping Pong PHY, address and command signals from two independent
controllers are multiplexed onto shared buses by delaying one of the controller
outputs by one full-rate clock cycle. The result is 1T timing, with a new command
being issued on each full-rate clock cycle. The following figure shows address and
command timing for the Ping Pong PHY.
The command signals CS, ODT, and CKE have two signals (one for ping and one for
pong); the other address and command signals are shared.
External Memory Interface Handbook Volume 3: Reference Material
114
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 57.
1T Command Timing Use by Ping Pong PHY
CK
CSn[0]
CSn[1]
Addr, ba
Cmd
Dev1
Cmd
Dev0
2.16.2 Stratix 10 Ping Pong PHY Architecture
In Stratix 10 EMIF, the Ping Pong PHY feature can be enabled only with the hard
memory controller, where two hard memory controllers are instantiated—one for the
primary interface and one for the secondary interface.
The hard memory controller I/O bank of the primary interface is used for address and
command and is always adjacent and above the hard memory controller bank of the
secondary interface. All four lanes of the primary hard memory controller bank are
used for address and command.
The following example shows a 2x16 Ping Pong PHY bank-lane configuration. The
upper bank (I/O bank N) is the address and command bank, which serves both the
primary and secondary interfaces. The primary hard memory controller is linked to the
secondary interface by the Ping Pong bus. The lower bank (I/O bank N-1) is the
secondary interface bank, which carries the data buses for both primary and
secondary interfaces. In the 2x16 case a total of four I/O banks are required for data,
hence two banks in total are sufficient for the implementation.
The data for the primary interface is routed down to the top two lanes of the
secondary I/O bank, and the data for the secondary interface is routed to the bottom
two lanes of the secondary I/O bank.
External Memory Interface Handbook Volume 3: Reference Material
115
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 58.
2x16 Ping Pong PHY I/O Bank-Lane Configuration
I/O Tile N
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
Primary HMC
I/O Tile N - 1
Secondary HMC
Address/
Command
Primaary
Interface
Data Bus
Secondary
Interface
Data Bus
A 2x32 interface can be implemented similarly, with the additional data lanes placed
above and below the primary and secondary I/O banks, such that primary data lanes
are placed above the primary bank and secondary data lanes are placed below the
secondary bank.
External Memory Interface Handbook Volume 3: Reference Material
116
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 59.
2x32 Ping Pong PHY I/O Bank-Lane Configuration.
I/O Tile N + 1
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
Control Path
I/O Tile N
Primary HMC
I/O Tile N - 1
Secondary HMC
I/O Tile N - 2
Primary
Interface
Data Bus
Address/
Command
Primaary
Interface
Data Bus
Secondary
Interface
Data Bus
Control Path
Secondary
Interface
Data Bus
2.16.3 Stratix 10 Ping Pong PHY Limitations
Ping Pong PHY supports up to two ranks per memory interface. In addition, the
maximum data width is x72, which is half the maximum width of x144 for a single
interface.
External Memory Interface Handbook Volume 3: Reference Material
117
2 Functional Description—Intel Stratix® 10 EMIF IP
Ping Pong PHY uses all lanes of the address and command I/O bank as address and
command. For information on the pin allocations of the DDR3 and DDR4 address and
command I/O bank, refer to DDR3 Scheme 1 and DDR4 Scheme 3, in External
Memory Interface Pin Information for Stratix 10 Devices, on www.altera.com.
An additional limitation is that I/O lanes may be left unused when you instantiate
multiple pairs of Ping Pong PHY interfaces. The following diagram shows two pairs of
x8 Pin Pong controllers (a total of 4 interfaces). Lanes highlighted in yellow are not
driven by any memory interfaces (unused lanes and pins can still serve as general
purpose I/Os). Even with some I/O lanes left unused, the Ping Pong PHY approach is
still beneficial in terms of resource usage, compared to independent interfaces.
Memory widths of 24 bits and 40 bits have a similar situation, while 16 bit, 32 bit, and
64 bit memory widths do not suffer this limitation.
External Memory Interface Handbook Volume 3: Reference Material
118
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 60.
Two Pairs of x8 Pin-Pong PHY Controllers
I/O Tile N
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
Control Path
I/O Tile N - 1
Primary HMC
I/O Tile N - 2
Primary
Interface
Data Bus
Address/
Command
Secondary
Interface
Data Bus
Secondary HMC
I/O Tile N - 3
Primary HMC
I/O Tile N - 4
Primary
Interface
Data Bus
Address/
Command
Secondary
Interface
Data Bus
Secondary HMC
2.16.4 Stratix 10 Ping Pong PHY Calibration
A Ping Pong PHY interface is calibrated as a regular interface of double width.
Calibration of a Ping Pong PHY interface incorporates two sequencers, one on the
primary hard memory controller I/O bank, and one on the secondary hard memory
controller I/O bank. To ensure that the two sequencers issue instructions on the same
memory clock cycle, the Nios II processor configures the sequencer on the primary
hard memory controller to receive a token from the secondary interface, ignoring any
External Memory Interface Handbook Volume 3: Reference Material
119
2 Functional Description—Intel Stratix® 10 EMIF IP
commands from the Avalon bus. Additional delays are programmed on the secondary
interface to allow for the passing of the token from the sequencer on the secondary
hard memory controller tile to the sequencer on the primary hard memory controller
tile. During calibration, the Nios II processor assumes that commands are always
issued from the sequencer on the primary hard memory controller I/O bank. After
calibration, the Nios II processor adjusts the delays for use with the primary and
secondary hard memory controllers.
2.16.5 Using the Ping Pong PHY
The following steps describe how to use the Ping Pong PHY for Stratix 10 EMIF.
1.
Configure a single memory interface according to your requirements.
2.
Select Instantiate two controllers sharing a Ping Pong PHY on the General
tab in the parameter editor.
The Quartus Prime software replicates the interface, resulting in two memory
controllers and a shared PHY. The system configures the I/O bank-lane structure,
without further input from you.
2.16.6 Ping Pong PHY Simulation Example Design
The following figure illustrates a top-level block diagram of a generated Ping Pong PHY
simulation example design, using two I/O banks.
Functionally, the IP interfaces with user traffic separately, as it would with two
independent memory interfaces. You can also generate synthesizable example
designs, where the external memory interface IP interfaces with a traffic generator.
Figure 61.
Ping Pong PHY Simulation Example Design
Simulation Example Design
EMIF
Tile N
Traffic
Generator 0
Sim
Checker
Primary
HMC
Tile N - 1
Lane 3
CS, ODT, CKE
Lane 2
CAS, RAS, WE, ADDR, BA, BG, ...
Lane 1
CS, ODT, CKE
Lane 0
Lane 3
DQ, DQS, DM
Lane 2
Traffic
Generator 1
Secondary
HMC
Memory
0
Lane 1
DQ, DQS, DM
Memory
1
Lane 0
2.17 AFI 4.0 Specification
The Altera PHY interface (AFI) 4.0 defines communication between the controller and
physical layer (PHY) in the external memory interface IP.
External Memory Interface Handbook Volume 3: Reference Material
120
2 Functional Description—Intel Stratix® 10 EMIF IP
The AFI is a single-data-rate interface, meaning that data is transferred on the rising
edge of each clock cycle. Most memory interfaces, however, operate at double-datarate, transferring data on both the rising and falling edges of the clock signal. If the
AFI interface is to directly control a double-data-rate signal, two single-data-rate bits
must be transmitted on each clock cycle; the PHY then sends out one bit on the rising
edge of the clock and one bit on the falling edge.
The AFI convention is to send the low part of the data first and the high part second,
as shown in the following figure.
Figure 62.
Single Versus Double Data Rate Transfer
clock
Single-data-rate
Double-data-rate
A High , A Low
A Low
A High
B High , B Low
B Low
B High
2.17.1 Bus Width and AFI Ratio
In cases where the AFI clock frequency is one-half or one-quarter of the memory clock
frequency, the AFI data must be twice or four times as wide, respectively, as the
corresponding memory data. The ratio between AFI clock and memory clock
frequencies is referred to as the AFI ratio. (A half-rate AFI interface has an AFI ratio of
2, while a quarter-rate interface has an AFI ratio of 4.)
In general, the width of the AFI signal depends on the following three factors:
•
The size of the equivalent signal on the memory interface. For example, if
a[15:0] is a DDR3 address input and the AFI clock runs at the same speed as
the memory interface, the equivalent afi_addr bus will be 16-bits wide.
•
The data rate of the equivalent signal on the memory interface. For example, if
d[7:0] is a double-data-rate QDR II input data bus and the AFI clock runs at the
same speed as the memory interface, the equivalent afi_write_data bus will
be 16-bits wide.
•
The AFI ratio. For example, if cs_n is a single-bit DDR3 chip select input and the
AFI clock runs at half the speed of the memory interface, the equivalent
afi_cs_n bus will be 2-bits wide.
The following formula summarizes the three factors described above:
AFI_width = memory_width * signal_rate * AFI_RATE_RATIO
Note:
The above formula is a general rule, but not all signals obey it. For definite signal-size
information, refer to the specific table.
2.17.2 AFI Parameters
The following tables list Altera PHY interface (AFI) parameters for AFI 4.0.
The parameters described in the following tables affect the width of AFI signal buses.
Parameters prefixed by MEM_IF_ refer to the signal size at the interface between the
PHY and memory device.
External Memory Interface Handbook Volume 3: Reference Material
121
2 Functional Description—Intel Stratix® 10 EMIF IP
Table 28.
Ratio Parameters
Parameter Name
Description
AFI_RATE_RATIO
The ratio between the AFI clock frequency and the memory clock frequency.
For full-rate interfaces this value is 1, for half-rate interfaces the value is 2,
and for quarter-rate interfaces the value is 4.
DATA_RATE_RATIO
The number of data bits transmitted per clock cycle. For single-date rate
protocols this value is 1, and for double-data rate protocols this value is 2.
ADDR_RATE_RATIO
The number of address bits transmitted per clock cycle. For single-date rate
address protocols this value is 1, and for double-data rate address protocols
this value is 2.
Table 29.
Memory Interface Parameters
Parameter Name
Description
MEM_IF_ADDR_WIDTH
The width of the address bus on the memory device(s).
MEM_IF_BANKADDR_WIDTH
The width of the bank address bus on the interface to the memory device(s).
Typically, the log 2 of the number of banks.
MEM_IF_CS_WIDTH
The number of chip selects on the interface to the memory device(s).
MEM_IF_WRITE_DQS_WIDTH
The number of DQS (or write clock) signals on the write interface. For
example, the number of DQS groups.
MEM_IF_CLK_PAIR_COUNT
The number of CK/CK# pairs.
MEM_IF_DQ_WIDTH
The number of DQ signals on the interface to the memory device(s). For
single-ended interfaces such as QDR II, this value is the number of D or Q
signals.
MEM_IF_DM_WIDTH
The number of data mask pins on the interface to the memory device(s).
MEM_IF_READ_DQS_WIDTH
The number of DQS signals on the read interface. For example, the number of
DQS groups.
Table 30.
Derived AFI Parameters
Parameter Name
Derivation Equation
AFI_ADDR_WIDTH
MEM_IF_ADDR_WIDTH * AFI_RATE_RATIO * ADDR_RATE_RATIO
AFI_BANKADDR_WIDTH
MEM_IF_BANKADDR_WIDTH * AFI_RATE_RATIO * ADDR_RATE_RATIO
AFI_CONTROL_WIDTH
AFI_RATE_RATIO * ADDR_RATE_RATIO
AFI_CS_WIDTH
MEM_IF_CS_WIDTH * AFI_RATE_RATIO
AFI_DM_WIDTH
MEM_IF_DM_WIDTH * AFI_RATE_RATIO * DATA_RATE_RATIO
AFI_DQ_WIDTH
MEM_IF_DQ_WIDTH * AFI_RATE_RATIO * DATA_RATE_RATIO
AFI_WRITE_DQS_WIDTH
MEM_IF_WRITE_DQS_WIDTH * AFI_RATE_RATIO
AFI_LAT_WIDTH
6
AFI_RLAT_WIDTH
AFI_LAT_WIDTH
AFI_WLAT_WIDTH
AFI_LAT_WIDTH * MEM_IF_WRITE_DQS_WIDTH
AFI_CLK_PAIR_COUNT
MEM_IF_CLK_PAIR_COUNT
AFI_WRANK_WIDTH
Number of ranks * MEM_IF_WRITE_DQS_WIDTH *AFI_RATE_RATIO
AFI_RRANK_WIDTH
Number of ranks * MEM_IF_READ_DQS_WIDTH *AFI_RATE_RATIO
External Memory Interface Handbook Volume 3: Reference Material
122
2 Functional Description—Intel Stratix® 10 EMIF IP
2.17.3 AFI Signals
The following tables list Altera PHY interface (AFI) signals grouped according to their
functions.
In each table, the Direction column denotes the direction of the signal relative to the
PHY. For example, a signal defined as an output passes out of the PHY to the
controller. The AFI specification does not include any bidirectional signals.
Not all signals are used for all protocols.
2.17.3.1 AFI Clock and Reset Signals
The AFI interface provides up to two clock signals and an asynchronous reset signal.
Table 31.
Clock and Reset Signals
Signal Name
Direction
Width
Description
afi_clk
Output
1
Clock with which all data exchanged on the AFI bus
is synchronized. In general, this clock is referred to
as full-rate, half-rate, or quarter-rate, depending on
the ratio between the frequency of this clock and
the frequency of the memory device clock.
afi_half_clk
Output
1
Clock signal that runs at half the speed of the
afi_clk. The controller uses this signal when the
half-rate bridge feature is in use. This signal is
optional.
afi_reset_n
Output
1
Asynchronous reset output signal. You must
synchronize this signal to the clock domain in which
you use it.
2.17.3.2 AFI Address and Command Signals
The address and command signals for AFI 4.0 encode read/write/configuration
commands to send to the memory device. The address and command signals are
single-data rate signals.
Table 32.
Address and Command Signals
Signal Name
Direction
Width
Description
afi_addr
Input
AFI_ADDR_WIDTH
Address or CA bus (LPDDR3 only).
ADDR_RATE_RATIO is 2 for
LPDDR3 CA bus.
afi_bg
Input
AFI_BANKGROUP_WIDTH
Bank group (DDR4 only).
afi_ba
Input
AFI_BANKADDR_WIDTH
Bank address. (Not applicable for
LPDDR3.)
afi_cke
Input
AFI_CLK_EN_WIDTH
Clock enable.
afi_cs_n
Input
AFI_CS_WIDTH
Chip select signal. (The number of
chip selects may not match the
number of ranks; for example,
RDIMMs and LRDIMMs require a
minimum of 2 chip select signals
for both single-rank and dual-rank
configurations. Consult your
memory device data sheet for
continued...
External Memory Interface Handbook Volume 3: Reference Material
123
2 Functional Description—Intel Stratix® 10 EMIF IP
Signal Name
Direction
Width
Description
information about chip select signal
width.) (Matches the number of
ranks for LPDDR3.)
afi_ras_n
Input
AFI_CONTROL_WIDTH
RAS# (for DDR2 and DDR3
memory devices.)
afi_we_n
Input
AFI_CONTROL_WIDTH
WE# (for DDR2, DDR3, and
RLDRAM II memory devices.)
afi_rw_n
Input
AFI_CONTROL_WIDTH * 2
RWA/B# (QDR-IV).
afi_cas_n
Input
AFI_CONTROL_WIDTH
CAS# (for DDR2 and DDR3
memory devices.)
afi_act_n
Input
AFI_CONTROL_WIDTH
ACT# (DDR4).
afi_ref_n
Input
AFI_CONTROL_WIDTH
REF# (for RLDRAM II memory
devices.)
afi_rst_n
Input
AFI_CONTROL_WIDTH
RESET# (for DDR3 and DDR4
memory devices.)
afi_odt
Input
AFI_CLK_EN_WIDTH
On-die termination signal for DDR2,
DDR3, and LPDDR3 memory
devices. (Do not confuse this
memory device signal with the
FPGA’s internal on-chip termination
signal.)
afi_par
Input
AFI_CS_WIDTH
Address and command parity input.
(DDR4)
Address parity input. (QDR-IV)
afi_ainv
Input
AFI_CONTROL_WIDTH
Address inversion. (QDR-IV)
afi_mem_clk_disable
Input
AFI_CLK_PAIR_COUNT
When this signal is asserted,
mem_clk and mem_clk_n are
disabled. This signal is used in lowpower mode.
afi_wps_n
Output
AFI_CS_WIDTH
WPS (for QDR II/II+ memory
devices.)
afi_rps_n
Output
AFI_CS_WIDTH
RPS (for QDR II/II+ memory
devices.)
2.17.3.3 AFI Write Data Signals
Write Data Signals for AFI 4.0 control the data, data mask, and strobe signals passed
to the memory device during write operations.
Table 33.
Write Data Signals
Signal Name
afi_dqs_burst
Direction
Input
Width
AFI_RATE_RATIO
Description
Controls the enable on the strobe
(DQS) pins for DDR2, DDR3,
LPDDR2, and LPDDR3 memory
devices. When this signal is
asserted, mem_dqs and mem_dqsn
are driven.
continued...
External Memory Interface Handbook Volume 3: Reference Material
124
2 Functional Description—Intel Stratix® 10 EMIF IP
Signal Name
Direction
Width
Description
This signal must be asserted before
afi_wdata_valid to implement the
write preamble, and must be driven
for the correct duration to generate
a correctly timed mem_dqs signal.
afi_wdata_valid
Input
AFI_RATE_RATIO
Write data valid signal. This signal
controls the output enable on the
data and data mask pins.
afi_wdata
Input
AFI_DQ_WIDTH
Write data signal to send to the
memory device at double-data
rate. This signal controls the PHY’s
mem_dq output.
afi_dm
Input
AFI_DM_WIDTH
Data mask. This signal controls the
PHY’s mem_dm signal for DDR2,
DDR3, LPDDR2, LPDDR3, and
RLDRAM II memory devices.)
Also directly controls the PHY's
mem_dbi signal for DDR4.
The mem_dm and mem_dbi
features share the same port on
the memory device.
afi_bws_n
Input
AFI_DM_WIDTH
Data mask. This signal controls the
PHY’s mem_bws_n signal for
QDR II/II+ memory devices.
afi_dinv
Input
AFI_WRITE_DQS_WIDTH * 2
Data inversion. It directly controls
the PHY's mem_dinva/b signal for
QDR-IV devices.
2.17.3.4 AFI Read Data Signals
Read Data Signals for AFI 4.0 control the data sent from the memory device during
read operations.
Table 34.
Read Data Signals
Signal Name
Direction
Width
Description
afi_rdata_en_full
Input
AFI_RATE_RATIO
Read data enable full. Indicates that the
memory controller is currently performing
a read operation. This signal is held high
for the entire read burst.If this signal is
aligned to even clock cycles, it is possible
to use 1-bit even in half-rate mode (i.e.,
AFI_RATE=2).
afi_rdata
Output
AFI_DQ_WIDTH
Read data from the memory device. This
data is considered valid only when
afi_rdata_valid is asserted by the PHY.
afi_rdata_valid
Output
AFI_RATE_RATIO
Read data valid. When asserted, this
signal indicates that the afi_rdata bus is
valid.If this signal is aligned to even clock
cycles, it is possible to use 1-bit even in
half-rate mode (i.e., AFI_RATE=2).
External Memory Interface Handbook Volume 3: Reference Material
125
2 Functional Description—Intel Stratix® 10 EMIF IP
2.17.3.5 AFI Calibration Status Signals
The PHY instantiates a sequencer which calibrates the memory interface with the
memory device and some internal components such as read FIFOs and valid FIFOs.
The sequencer reports the results of the calibration process to the controller through
the Calibration Status Signals in the AFI interface.
Table 35.
Calibration Status Signals
Signal Name
Direction
Width
Description
afi_cal_success
Output
1
Asserted to indicate that calibration has
completed successfully.
afi_cal_fail
Output
1
Asserted to indicate that calibration has
failed.
afi_cal_req
Input
1
Effectively a synchronous reset for the
sequencer. When this signal is asserted,
the sequencer returns to the reset state;
when this signal is released, a new
calibration sequence begins.
afi_wlat
Output
AFI_WLAT_WIDTH
The required write latency in afi_clk
cycles, between address/command and
write data being issued at the PHY/
controller interface. The afi_wlat value
can be different for different groups; each
group’s write latency can range from 0 to
63. If write latency is the same for all
groups, only the lowest 6 bits are
required.
afi_rlat
Output
AFI_RLAT_WIDTH
The required read latency in afi_clk cycles
between address/command and read
data being returned to the PHY/controller
interface. Values can range from 0 to 63.
(1)
Note to Table:
1. The afi_rlat signal is not supported for PHY-only designs. Instead, you can sample the afi_rdata_valid signal to
determine when valid read data is available.
2.17.3.6 AFI Tracking Management Signals
When tracking management is enabled, the sequencer can take control over the AFI
4.0 interface at given intervals, and issue commands to the memory device to track
the internal DQS Enable signal alignment to the DQS signal returning from the
memory device. The tracking management portion of the AFI 4.0 interface provides a
means for the sequencer and the controller to exchange handshake signals.
Table 36.
Tracking Management Signals
Signal Name
Direction
Width
Description
afi_ctl_refresh_done
Input
4
Handshaking signal from controller to
tracking manager, indicating that a
refresh has occurred and waiting for a
response.
afi_seq_busy
Output
4
Handshaking signal from sequencer to
controller, indicating when DQS tracking
is in progress.
afi_ctl_long_idle
Input
4
Handshaking signal from controller to
tracking manager, indicating that it has
exited low power state without a periodic
refresh, and waiting for response.
External Memory Interface Handbook Volume 3: Reference Material
126
2 Functional Description—Intel Stratix® 10 EMIF IP
2.17.3.7 AFI Shadow Register Management Signals
Shadow registers are a feature that enables high-speed multi-rank support. Shadow
registers allow the sequencer to calibrate each rank separately, and save the
calibrated settings—such as deskew delay-chain configurations—of each rank in its
own set of shadow registers.
During a rank-to-rank switch, the correct set of calibrated settings is restored just in
time to optimize the data valid window. The PHY relies on additional AFI signals to
control which set of shadow registers to activate.
Table 37.
Shadow Register Management Signals
Signal Name
Direction
Width
Description
afi_wrank
Input
AFI_WRANK_WIDTH
Signal from controller
specifying which rank the
write data is going to. The
signal timing is identical to
that of afi_dqs_burst. That
is, afi_wrank must be
asserted at the same time
and must last the same
duration as the
afi_dqs_burst signal.
afi_rrank
Output
AFI_RRANK_WIDTH
Signal from controller
specifying which rank is
being read. The signal must
be asserted at the same
time as the afi_rdata_en
signal when issuing a read
command, but unlike
afi_rdata_en, afi_rrank is
stateful. That is, once
asserted, the signal value
must remain unchanged
until the controller issues a
new read command to a
different rank.
Both the afi_wrank and afi_rrank signals encode the rank being accessed using the
one-hot scheme (e.g. in a quad-rank interface, 0001, 0010, 0100, 1000 refer to the
1st, 2nd, 3rd, 4th rank respectively). The ordering within the bus is the same as other
AFI signals. Specifically the bus is ordered by time slots, for example:
Half-rate afi_w/rrank = {T1, T0}
Quarter-rate afi_w/rrank = {T3, T2, T1, T0}
Where Tx is a number of rank-bit words that one-hot encodes the rank being accessed
at the yth full-rate cycle.
External Memory Interface Handbook Volume 3: Reference Material
127
2 Functional Description—Intel Stratix® 10 EMIF IP
Additional Requirements for Stratix 10 Shadow Register Support
To ensure that the hardware has enough time to switch from one shadow register to
another, the controller must satisfy the following minimum rank-to-rank-switch delays
(tRTRS):
•
Two read commands going to different ranks must be separated by a minimum of
3 full-rate cycles (in addition to the burst length delay needed to avoid collision of
data bursts).
•
Two write commands going to different rank must be separated by a minimum of
4 full-rate cycles (in addition to the burst length delay needed to avoid collision of
data bursts).
The Stratix 10 device family supports a maximum of 4 sets of shadow registers, each
for an independent set of timings. More than 4 ranks are supported if those ranks
have four or fewer sets of independent timing. For example, the rank multiplication
mode of an LRDIMM allows more than one physical rank to share a set of timing data
as a single logical rank. Therefore Stratix 10 devices can support up to 4 logical ranks,
though that means more than 4 physical ranks.
2.17.4 AFI 4.0 Timing Diagrams
2.17.4.1 AFI Address and Command Timing Diagrams
Depending on the ratio between the memory clock and the PHY clock, different
numbers of bits must be provided per PHY clock on the AFI interface. The following
figures illustrate the AFI address/command waveforms in full, half and quarter rate
respectively.
The waveforms show how the AFI command phase corresponds to the memory
command output. AFI command 0 corresponds to the first memory command slot, AFI
command 1 corresponds to the second memory command slot, and so on.
External Memory Interface Handbook Volume 3: Reference Material
128
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 63.
AFI Address and Command Full-Rate
Memory Interface
mem_clk
mem_cs_n
mem_cke
mem_ras_n
mem_cas_n
mem_we_n
AFI Interface
afi_clk
afi_cs_n
afi_cke
afi_ras_n
afi_cas_n
afi_we_n
External Memory Interface Handbook Volume 3: Reference Material
129
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 64.
AFI Address and Command Half-Rate
Memory Interface
mem_clk
mem_cs_n
mem_cke
mem_ras_n
mem_cas_n
mem_we_n
AFI Interface
afi_clk
afi_cs_n[1]
afi_cs_n[0]
1
0
0
1
afi_cke[1]
afi_cke[0]
1
1
1
1
afi_ras_n[1]
afi_ras_n[0]
1
0
1
1
afi_cas_n[1]
afi_cas_n[0]
1
1
0
1
afi_we_n[1]
afi_we_n[0]
1
1
0
1
External Memory Interface Handbook Volume 3: Reference Material
130
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 65.
AFI Address and Command Quarter-Rate
Memory Interface
mem_clk
mem_cs_n
mem_cke
mem_ras_n
mem_cas_n
mem_we_n
AFI Interface
afi_clk
afi_cs_n[3]
0
1
afi_cs_n[2]
afi_cs_n[1]
afi_cs_n[0]
1
0
0
1
1
0
afi_cke[3]
1
1
afi_cke[2]
afi_cke[1]
afi_cke[0]
1
1
1
1
1
1
1
1
afi_ras_n[3]
afi_ras_n[2]
afi_ras_n[1]
1
1
0
1
afi_ras_n[0]
1
0
afi_cas_n[3]
afi_cas_n[2]
afi_cas_n[1]
afi_cas_n[0]
0
1
1
0
1
1
1
1
afi_we_n[3]
afi_we_n[2]
0
1
1
0
afi_we_n[1]
afi_we_n[0]
1
1
1
1
2.17.4.2 AFI Write Sequence Timing Diagrams
The following timing diagrams illustrate the relationships between the write command
and corresponding write data and write enable signals, in full, half, and quarter rate.
For half rate and quarter rate, when the write command is sent on the first memory
clock in a PHY clock (for example, afi_cs_n[0] = 0), that access is called aligned
access; otherwise it is called unaligned access. You may use either aligned or
unaligned access, or you may use both, but you must ensure that the distance
External Memory Interface Handbook Volume 3: Reference Material
131
2 Functional Description—Intel Stratix® 10 EMIF IP
between the write command and the corresponding write data are constant on the
AFI interface. For example, if a command is sent on the second memory clock in a PHY
clock, the write data must also start at the second memory clock in a PHY clock.
Write sequences with wlat=0
Figure 66.
AFI Write Data Full-Rate, wlat=0
afi_clk
afi_command
WR
WR
WR
afi_wdata_valid
afi_wdata
A
B
C
D
E
F
afi_dm
M
N
O
P
Q
R
The following diagrams illustrate both aligned and unaligned access. The first three
write commands are aligned accesses where they were issued on LSB of
afi_command. The fourth write command is unaligned access where it was issued on
a different command slot. AFI signals must be shifted accordingly, based on the
command slot.
Figure 67.
AFI Write Data Half-Rate, wlat=0
afi_clk
afi_command[1]
NOP
NOP
NOP
NOP
afi_command[0]
WR
WR
WR
WR
afi_wdata_valid[1]
1
1
1
1
0
afi_wdata_valid[0]
1
1
1
0
1
afi_wdata[1]
B
D
F
G
afi_wdata[0]
A
C
E
afi_dm[1]
N
P
R
afi_dm[0]
M
O
Q
External Memory Interface Handbook Volume 3: Reference Material
132
H
S
T
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 68.
AFI Write Data Quarter-Rate, wlat=0
afi_clk
afi_command[3]
NOP
NOP
NOP
WR
afi_command[2]
NOP
NOP
NOP
NOP
afi_command[1]
NOP
NOP
NOP
NOP
afi_command[0]
WR
WR
WR
NOP
afi_wdata_valid[3]
1
1
1
1
0
afi_wdata_valid[2]
1
1
1
0
1
afi_wdata_valid[1]
1
1
1
0
1
afi_wdata_valid[0]
1
1
1
0
1
afi_wdata[3]
D
H
L
A
afi_wdata[2]
C
G
K
D
afi_wdata[1]
B
F
J
C
afi_wdata[0]
A
E
I
B
afi_dm[3]
P
T
X
afi_dm[2]
O
S
W
P
afi_dm[1]
N
R
V
O
afi_dm[0]
M
Q
U
N
M
Write sequences with wlat=non-zero
The afi_wlat is a signal from the PHY. The controller must delay afi_dqs_burst,
afi_wdata_valid, afi_wdata and afi_dm signals by a number of PHY clock cycles
equal to afi_wlat, which is a static value determined by calibration before the PHY
asserts cal_success to the controller. The following figures illustrate the cases when
wlat=1. Note that wlat is in the number of PHY clocks and therefore wlat=1 equals 1,
2, and 4 memory clocks delay, respectively, on full, half and quarter rate.
External Memory Interface Handbook Volume 3: Reference Material
133
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 69.
AFI Write Data Full-Rate, wlat=1
afi_clk
afi_command
WR
WR
WR
afi_wdata_valid
Figure 70.
afi_wdata
A
B
C
D
E
F
afi_dm
M
N
O
P
Q
R
AFI Write Data Half-Rate, wlat=1
afi_clk
afi_command[1]
NOP
NOP
NOP
NOP
afi_command[0]
WR
WR
WR
WR
afi_wdata_valid[1]
1
1
1
1
0
afi_wdata_valid[0]
1
1
1
0
1
afi_wdata[1]
B
D
F
G
afi_wdata[0]
A
C
E
afi_dm[1]
N
P
R
afi_dm[0]
M
O
Q
External Memory Interface Handbook Volume 3: Reference Material
134
H
S
T
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 71.
AFI Write Data Quarter-Rate, wlat=1
afi_clk
afi_command[3]
NOP
NOP
NOP
WR
afi_command[2]
NOP
NOP
NOP
NOP
afi_command[1]
NOP
NOP
NOP
NOP
afi_command[0]
WR
WR
WR
NOP
afi_wdata_valid[3]
1
1
1
1
0
afi_wdata_valid[2]
1
1
1
0
1
afi_wdata_valid[1]
1
1
1
0
1
afi_wdata_valid[0]
1
1
1
0
1
afi_wdata[3]
D
H
L
A
afi_wdata[2]
C
G
K
D
afi_wdata[1]
B
F
J
C
afi_wdata[0]
A
E
I
B
afi_dm[3]
P
T
X
afi_dm[2]
O
S
W
P
afi_dm[1]
N
R
V
O
afi_dm[0]
M
Q
U
N
M
DQS burst
The afi_dqs_burst signal must be asserted one or two complete memory clock
cycles earlier to generate DQS preamble. DQS preamble is equal to one-half and onequarter AFI clock cycles in half and quarter rate, respectively.
A DQS preamble of two is required in DDR4, when the write preamble is set to two
clock cycles.
The following diagrams illustrate how afi_dqs_burst must be asserted in full, half, and
quarter-rate configurations.
External Memory Interface Handbook Volume 3: Reference Material
135
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 72.
AFI DQS Burst Full-Rate, wlat=1
afi_clk
afi_command
WR
WR
WR
afi_dqs_burst
afi_wdata_valid
Figure 73.
afi_wdata
A
B
C
D
E
F
afi_dm
M
N
O
P
Q
R
AFI DQS Burst Half-Rate, wlat=1
afi_clk
afi_command[1]
NOP
NOP
NOP
NOP
afi_command[0]
WR
WR
WR
WR
afi_dqs_burst[1]
1
1
1
1
1
1
0
afi_dqs_burst[0]
0
1
0
1
1
1
1
afi_wdata_valid[1]
1
1
1
1
0
afi_wdata_valid[0]
1
1
1
0
1
afi_wdata[1]
B
D
F
G
afi_wdata[0]
A
C
E
afi_dm[1]
N
P
R
afi_dm[0]
M
O
Q
External Memory Interface Handbook Volume 3: Reference Material
136
H
S
T
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 74.
AFI DQS Burst Quarter-Rate, wlat=1
afi_clk
afi_command[3]
NOP
NOP
NOP
WR
afi_command[2]
NOP
NOP
NOP
NOP
afi_command[1]
NOP
NOP
NOP
NOP
afi_command[0]
WR
WR
WR
NOP
afi_dqs_burst[3]
1
1
1
1
1
1
0
afi_dqs_burst[2]
0
1
0
1
1
1
1
afi_dqs_burst[1]
0
1
0
1
1
0
1
afi_dqs_burst[0]
0
1
0
1
1
0
1
afi_wdata_valid[3]
1
1
1
1
0
afi_wdata_valid[2]
1
1
1
0
1
afi_wdata_valid[1]
1
1
1
0
1
afi_wdata_valid[0]
1
1
1
0
1
afi_wdata[3]
D
H
L
A
afi_wdata[2]
C
G
K
D
afi_wdata[1]
B
F
J
C
afi_wdata[0]
A
E
I
B
afi_dm[3]
P
T
X
afi_dm[2]
O
S
W
P
afi_dm[1]
N
R
V
O
afi_dm[0]
M
Q
U
N
M
Write data sequence with DBI (DDR4 and QDRIV only)
The DDR4 write DBI feature is supported in the PHY, and when it is enabled, the PHY
sends and receives the DBI signal without any controller involvement. The sequence is
identical to non-DBI scenarios on the AFI interface.
External Memory Interface Handbook Volume 3: Reference Material
137
2 Functional Description—Intel Stratix® 10 EMIF IP
Write data sequence with CRC (DDR4 only)
When the CRC feature of the PHY is enabled and used, the controller ensures at least
one memory clock cycle between write commands, during which the PHY inserts the
CRC data. Sending back to back write command would cause functional failure. The
following figures show the legal sequences in CRC mode.
Entries marked as 0 and RESERVE must be observed by the controller; no information
is allowed on those entries.
Figure 75.
AFI Write Data with CRC Half-Rate, wlat=2
afi_clk
afi_command[1]
NOP
NOP
NOP
afi_command[0]
WR
WR
WR
afi_dqs_burst[1]
1
1
1
afi_dqs_burst[0]
0
1
1
afi_wdata_valid[1]
1
1
afi_wdata_valid[0]
1
1
afi_wdata[1]
B
D
afi_wdata[0]
A
C
afi_dm[1]
N
P
afi_dm[0]
M
O
External Memory Interface Handbook Volume 3: Reference Material
138
0
0
Reserve
Reserve
1
1
1
1
1
0
0
1
1
0
1
1
1
1
1
1
0
1
1
0
1
1
F
H
I
H
Reserve
E
G
Reserve
J
L
R
T
U
W
Reserve
Q
S
Reserve
V
X
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 76.
AFI Write Data with CRC Quarter-Rate, wlat=2
afi_clk
afi_command[1]
NOP
NOP
NOP
afi_command[0]
WR
WR
WR
afi_dqs_burst[3]
1
1
1
1
1
1
1
0
afi_dqs_burst[2]
0
1
0
1
1
1
0
1
afi_dqs_burst[1]
0
1
0
1
1
0
1
1
afi_dqs_burst[0]
0
1
0
1
0
1
1
1
0
afi_wdata_valid[3]
1
1
1
1
1
0
afi_wdata_valid[2]
1
1
1
1
0
1
afi_wdata_valid[1]
1
1
1
0
1
1
afi_wdata_valid[0]
1
1
0
1
1
1
afi_wdata[3]
D
D
G
J
M
Reserve
afi_wdata[2]
C
C
F
I
Reserve
P
afi_wdata[1]
B
B
E
Reserve
L
O
afi_wdata[0]
A
A
Reserve
H
K
N
afi_dm[3]
D
D
G
J
M
Reserve
afi_dm[2]
C
C
F
I
Reserve
P
afi_dm[1]
B
B
E
Reserve
L
O
afi_dm[0]
A
A
Reserve
H
K
N
0
Reserve
Reserve
2.17.4.3 AFI Read Sequence Timing Diagrams
The following waveforms illustrate the AFI write data waveform in full, half, and
quarter-rate, respectively.
The afi_rdata_en_full signal must be asserted for the entire read burst
operation. The afi_rdata_en signal need only be asserted for the intended read
data.
Aligned and unaligned access for read commands is similar to write commands;
however, the afi_rdata_en_full signal must be sent on the same memory clock in
a PHY clock as the read command. That is, if a read command is sent on the second
memory clock in a PHY clock, afi_rdata_en_full must also be asserted, starting
from the second memory clock in a PHY clock.
External Memory Interface Handbook Volume 3: Reference Material
139
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 77.
AFI Read Data Full-Rate
afi_clk
afi_command
RD
RD
RD
afi_rdata_en_full
afi_rdata
A
B
C
D
E
F
afi_rdata_valid
The following figure illustrates that the second and third reads require only the first
and second half of data, respectively. The first three read commands are aligned
accesses where they are issued on the LSB of afi_command. The fourth read
command is unaligned access, where it is issued on a different command slot. AFI
signals must be shifted accordingly, based on command slot.
Figure 78.
AFI Read Data Half-Rate
afi_clk
afi_command[1]
NOP
NOP
NOP
RD
afi_command[0]
RD
RD
RD
NOP
afi_rdata_en_full[1]
1
1
1
1
0
afi_rdata_en_full[0]
1
1
1
0
1
afi_rdata[1]
B
D
F
afi_rdata[0]
A
C
E
afi_rdata_valid[1]
1
1
1
afi_rdata_valid[0]
1
1
1
G
H
1
1
In the following figure, the first three read commands are aligned accesses where
they are issued on the LSB of afi_command. The fourth read command is unaligned
access, where it is issued on a different command slot. AFI signals must be shifted
accordingly, based on command slot.
External Memory Interface Handbook Volume 3: Reference Material
140
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 79.
AFI Read Data Quarter-Rate
afi_clk
afi_command[3]
NOP
NOP
NOP
RD
afi_command[2]
NOP
NOP
NOP
NOP
afi_command[1]
NOP
NOP
NOP
NOP
afi_command[0]
RD
RD
RD
NOP
afi_rdata_en_full[3]
1
1
1
1
0
afi_rdata_en_full[2]
1
1
1
0
1
afi_rdata_en_full[1]
1
1
1
0
1
afi_rdata_en_full[0]
1
1
1
0
1
afi_rdata[3]
D
H
L
afi_rdata[2]
C
G
K
P
afi_rdata[1]
B
F
J
O
afi_rdata[0]
A
E
I
N
afi_rdata_valid[3]
1
1
1
1
0
afi_rdata_valid[2]
1
1
1
0
1
afi_rdata_valid[1]
1
1
1
0
1
afi_rdata_valid[0]
1
1
1
0
1
M
2.17.4.4 AFI Calibration Status Timing Diagram
The controller interacts with the PHY during calibration at power-up and at
recalibration.
At power-up, the PHY holds afi_cal_success and afi_cal_fail 0 until calibration
is done, when it asserts afi_cal_success, indicating to controller that the PHY is
ready to use and afi_wlat and afi_rlat signals have valid values.
At recalibration, the controller asserts afi_cal_req, which triggers the same
sequence as at power-up, and forces recalibration of the PHY.
External Memory Interface Handbook Volume 3: Reference Material
141
2 Functional Description—Intel Stratix® 10 EMIF IP
Figure 80.
Calibration
PHY Status Calibrating
Controller Working
Re-Calibrating
Controller Working
AFI Interface
afi_cal_success
afi_cal_fail
afi_cal_req
afi_wlat
9
9
afi_rlat
9
9
2.18 Stratix 10 Resource Utilization
The following tables provide resource utilization information for external memory
interfaces on Stratix 10 devices.
2.18.1 QDR-IV Resource Utilization in Stratix 10 Devices
The following table shows typical resource usage of QDR-IV controllers for Stratix 10
devices.
Table 38.
QDR-IV Resource Utilization in Stratix 10 Devices
Memory Width
(Bits)
Combinational
ALUTs
Dedicated Logic
Registers
Block Memory
Bits
M20Ks
Soft Controller
18
2123
4592
18432
8
1
36
2127
6023
36864
16
1
72
2114
8826
73728
32
1
2.19 Stratix 10 EMIF Latency
The following latency data applies to all memory protocols supported by the Stratix 10
EMIF IP.
Table 39.
Rate
Latency in Full-Rate Memory Clock Cycles
1
Controller
Address &
Command
PHY
Address &
Command
Memory
Read
Latency 2
PHY Read
Data
Return
Controller
Read Data
Return
Round Trip
Round Trip
Without
Memory
—
Half:Write
12
2
3-23
—
—
—
Half:Read
8
2
3-23
6
8
27-47
24
continued...
External Memory Interface Handbook Volume 3: Reference Material
142
2 Functional Description—Intel Stratix® 10 EMIF IP
Rate
1
Controller
Address &
Command
PHY
Address &
Command
Memory
Read
Latency 2
PHY Read
Data
Return
Controller
Read Data
Return
Round Trip
Round Trip
Without
Memory
Quarter:Writ
e
14
2
3-23
—
—
—
—
Quarter:Rea
d
10
2
3-23
6
14
35-55
32
Half:Write
(ECC)
14
2
3-23
—
—
—
—
Half:Read
(ECC)
12
2
3-23
6
8
31-51
28
Quarter:Writ
e (ECC)
14
2
3-23
—
—
—
—
Quarter:Rea
d (ECC)
12
2
3-23
6
14
37-57
34
1. User interface rate; the controller always operates in half rate.
2. Minimum and maximum read latency range for DDR3, DDR4, and LPDDR3.
2.20 Integrating a Custom Controller with the Hard PHY
If you want to use your own custom memory controller, you must integrate the
controller with the hard PHY to achieve a complete memory solution.
Observe the following general guidelines:
•
When you configure your external memory interface IP, ensure that you select
Configuration ➤ Hard PHY Only on the General tab in the parameter editor.
•
Consult the AFI 4.0 Specification, for detailed information on the AFI interface to
the PHY.
Related Links
AFI 4.0 Specification on page 120
The Altera PHY interface (AFI) 4.0 defines communication between the controller
and physical layer (PHY) in the external memory interface IP.
2.21 Document Revision History
Date
May 2017
Version
2017.05.08
Changes
Initial release.
External Memory Interface Handbook Volume 3: Reference Material
143
3 Functional Description—Intel Arria 10 EMIF IP
3 Functional Description—Intel Arria 10 EMIF IP
Intel Arria 10 devices can interface with external memory devices clocking at
frequencies of up to 1.3 GHz. The external memory interface IP component for Arria
10 devices provides a single parameter editor for creating external memory interfaces,
regardless of memory protocol. Unlike earlier EMIF solutions which used protocolspecific parameter editors to create memory interfaces via a complex RTL generation
method, the Arria 10 EMIF solution captures the protocol-specific hardened EMIF logic
of the Arria 10 device together with more generic soft logic.
The Arria 10 EMIF solution is designed with the following implementations in mind:
Hard Memory Controller and Hard PHY
This implementation provides a complete external memory interface, based on the
hard memory controller and hard PHY that are part of the Arria 10 silicon. An AvalonMM interface is available for integration with user logic.
Soft Memory Controller and Hard PHY
This implementation provides a complete external memory interface, using an Intelprovided soft-logic-based memory controller and the hard PHY that is part of the Arria
10 silicon. An Avalon-MM interface is available for integration with user logic.
Custom Memory Controller and Hard PHY (PHY only)
This implementation provides access to the AFI interface, to allow use of a custom or
third-party memory controller with the hard PHY that is part of the Arria 10 silicon.
Because only the PHY component is provided by Intel, this configuration is also known
as PHY only.
Related Links
•
Arria 10 EMIF Architecture: Introduction on page 146
The Arria 10 EMIF architecture contains many new hardware features designed
to meet the high-speed requirements of emerging memory protocols, while
consuming the smallest amount of core logic area and power.
•
Hardware Resource Sharing Among Multiple EMIFs on page 158
Often, it is necessary or desirable to share resources between interfaces.
•
Arria 10 EMIF IP Component on page 162
The external memory interface IP component for Arria 10 provides a complete
solution for implementing DDR3, DDR4, and QDR-IV external memory
interfaces.
•
Compiling Arria 10 EMIF IP with the Quartus Prime Software on page 185
•
Debugging Arria 10 EMIF IP on page 186
Intel Corporation. All rights reserved. Intel, the Intel logo, Altera, Arria, Cyclone, Enpirion, MAX, Nios, Quartus
and Stratix words and logos are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other
countries. Intel warrants performance of its FPGA and semiconductor products to current specifications in
accordance with Intel's standard warranty, but reserves the right to make changes to any products and services
at any time without notice. Intel assumes no responsibility or liability arising out of the application or use of any
information, product, or service described herein except as expressly agreed to in writing by Intel. Intel
customers are advised to obtain the latest version of device specifications before relying on any published
information and before placing orders for products or services.
*Other names and brands may be claimed as the property of others.
ISO
9001:2008
Registered
3 Functional Description—Intel Arria 10 EMIF IP
You can debug hardware failures by connecting to the EMIF Debug Toolkit or by
exporting an Avalon-MM slave port, from which you can access information
gathered during calibration.
•
Intel Arria 10 Core Fabric and General Purpose I/Os Handbook
3.1 Supported Memory Protocols
The following table lists the external memory protocols supported by Arria 10 devices.
Table 40.
Supported Memory Protocols
Protocol
Hard Controller and Hard
PHY
Soft Controller and Hard
PHY
PHY Only
DDR4
Yes
—
Yes
DDR3
Yes
—
Yes
LPDDR3
Yes
—
Yes
RLDRAM 3
—
Third party
Yes
QDR II/II+/II+ Xtreme
—
Yes
—
QDR-IV
—
Yes
—
Memory protocols not listed above are not supported by the Arria 10 EMIF IP;
however, you can implement a custom memory interface for these protocols using the
Altera PHYLite Megafunction.
LPDDR3 is supported for simulation, compilation, and timing. Hardware support for
LPDDR3 will be provided in a future release.
Note:
To achieve maximum top-line spec performance for a DDR4 interface, both read data
bus inversion and periodic OCT calibration must be enabled. For maximum top-line
spec performance for a QDR-IV interface, read data bus inversion must be enabled.
Related Links
Altera PHYLite for Memory Megafunction User Guide
3.2 Key Differences Compared to UniPHY IP and Previous Device
Families
The Arria 10 EMIF IP has a new design which bears several notable differences
compared to UniPHY-based IP. If you are familiar with the UniPHY-based IP, you should
review the following differences, as they affect the way you generate, instantiate, and
use the Arria 10 EMIF IP.
External Memory Interface Handbook Volume 3: Reference Material
145
3 Functional Description—Intel Arria 10 EMIF IP
•
Unlike the UniPHY-based IP, which presents a protocol-specific parameter editor
for each supported memory protocol, the Arria 10 EMIF IP uses one parameter
editor for all memory protcols.
•
With UniPHY-based IP, you must run the
<variation_name>_pin_assignments.tcl script following synthesis, to apply
I/O assignments to the project's .qsf file. In Arria 10 EMIF IP, the
<variation_name>_pin_assignments.tcl script is no longer necessary. All the
I/O assignments are included in the generated .qip file, which the Quartus Prime
software processes during compilation. Assignments that you make in the .qsf
file override those in the .qip file.
•
The Arria 10 EMIF IP includes a <variation_name>readme.txt file, located in
the / altera_emif_arch_nf_<version> directory. This file contains important
information about the implementation of the IP, including pin location guidelines,
information on resource sharing, and signal descriptions.
•
To generate the synthesis example design or the simulation example design, you
need to run additional scripts after generation.
3.3 Migrating from Previous Device Families
There is no automatic migration mechanism for external memory interface IP
generated for previous device families.
To migrate an existing EMIF IP from an earlier device family to Arria 10, you must
reparameterize and regenerate the IP targeting Arria 10, using either the IP Catalog or
Qsys. If you attempt to recompile an existing IP generated for a previous device
family, you will encounter errors in the Quartus Prime software.
UniPHY-based IP continues to be supported for previous device families.
3.4 Arria 10 EMIF Architecture: Introduction
The Arria 10 EMIF architecture contains many new hardware features designed to
meet the high-speed requirements of emerging memory protocols, while consuming
the smallest amount of core logic area and power.
The following are key hardware features of the Arria 10 EMIF architecture:
Hard Sequencer
The sequencer employs a hard Nios II processor, and can perform memory calibration
for a wide range of protocols. You can share the sequencer among multiple memory
interfaces of the same or different protocols.
Hard PHY
The hard PHY in Arria 10 devices can interface with external memories running at
speeds of up to 1.3 GHz. The PHY circuitry is hardened in the silicon, which simplifies
the challenges of achieving timing closure and minimal power consumption.
Hard Memory Controller
The hard memory controller reduces latency and minimizes core logic consumption in
the external memory interface. The hard memory controller supports the DDR3,
DDR4, and LPDDR3 memory protocols.
External Memory Interface Handbook Volume 3: Reference Material
146
3 Functional Description—Intel Arria 10 EMIF IP
PHY-Only Mode
Protocols that use a hard controller (DDR4, DDR3, and LPDDR3) as well as RLDRAM 3,
provide a "PHY-only" option. When selected, this option generates only the PHY and
sequencer, but not the controller. This PHY-Only mode provides a mechanism by which
to integrate your own custom soft controller.
High-Speed PHY Clock Tree
Dedicated high speed PHY clock networks clock the I/O buffers in Arria 10 EMIF IP.
The PHY clock trees exhibit low jitter and low duty cycle distortion, maximizing the
data valid window.
Automatic Clock Phase Alignment
Automatic clock phase alignment circuitry dynamically adjust the clock phase of core
clock networks to match the clock phase of the PHY clock networks. The clock phase
alignment circuitry minimizes clock skew that can complicate timing closure in
transfers between the FPGA core and the periphery.
Resource Sharing
The Arria 10 architecture simplifies resource sharing between memory interfaces.
Resources such as the OCT calibration block, PLL reference clock pin, and core clock
can be shared. The hard Nios processor in the I/O AUX must be shared across all
interfaces in a column.
Related Links
Intel Arria 10 Core Fabric and General Purpose I/Os Handbook
3.4.1 Arria 10 EMIF Architecture: I/O Subsystem
The I/O subsystem consists of two columns inside the core of Arria 10 devices.
Each column can be thought of as loosely analogous to an I/O bank.
Figure 81.
Arria 10 I/O Subsystem
Core Fabric
I/O Column
Transceivers (if applicable)
External Memory Interface Handbook Volume 3: Reference Material
147
3 Functional Description—Intel Arria 10 EMIF IP
The I/O subsystem provides the following features:
•
General-purpose I/O registers and I/O buffers
•
On-chip termination control (OCT)
•
I/O PLLs for external memory interfaces and user logic
•
Low-voltage differential signaling (LVDS)
•
External memory interface components, as follows:
—
Hard memory controller
—
Hard PHY
—
Hard Nios processor and calibration logic
—
DLL
Related Links
Arria 10 Core Fabric and General Purpose I/Os Handbook
3.4.2 Arria 10 EMIF Architecture: I/O Column
Arria 10 devices have two I/O columns, which contain the hardware related to
external memory interfaces.
Each I/O column contains the following major parts:
•
A hardened Nios processor with dedicated memory. This Nios block is referred to
as the I/O AUX.
•
Up to 13 I/O banks. Each I/O bank contains the hardware necessary for an
external memory interface.
External Memory Interface Handbook Volume 3: Reference Material
148
3 Functional Description—Intel Arria 10 EMIF IP
Figure 82.
I/O Column
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
Transceiver
Block
2L
3H
2K
3G
2J
3F
2I
3E
2H
3D
2G
3C
2F
3B
2A
3A
I/O
Column
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
I/O Center
Transceiver Block
Transceiver Block
Individual
I/O Banks
I/O PLL
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
LVDS I/O Buffer Pair
I/O
Column
I/O Lane
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
I/O Lane
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
I/O DLL
I/O CLK
OCT
VR
Hard Memory Controller
and
PHY Sequencer
I/O Lane
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
I/O Lane
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
SERDES & DPA
Bank
Control
3.4.3 Arria 10 EMIF Architecture: I/O AUX
Each column includes one I/O AUX, which contains a hardened Nios II processor with
dedicated memory. The I/O AUX is responsible for calibration of all the EMIFs in the
column.
The I/O AUX includes dedicated memory which stores both the calibration algorithm
and calibration run-time data. The hardened Nios II processor and the dedicated
memory can be used only by an external memory interface, and cannot be employed
for any other use. The I/O AUX can interface with soft logic, such as the debug toolkit,
via an Avalon-MM bus.
The I/O AUX is clocked by an on-die oscillator, and therefore does not consume a PLL.
3.4.4 Arria 10 EMIF Architecture: I/O Bank
A single I/O bank contains all the hardware needed to build an external memory
interface. Each I/O column contains up to 13 I/O banks; the exact number of banks
depends on device size and pin package. You can make a wider interface by
connecting multiple banks together.
External Memory Interface Handbook Volume 3: Reference Material
149
3 Functional Description—Intel Arria 10 EMIF IP
Each I/O bank resides in an I/O column, and contains the following components:
Figure 83.
•
Hard memory controller
•
Sequencer components
•
PLL and PHY clock trees
•
DLL
•
Input DQS clock trees
•
48 pins, organized into four I/O lanes of 12 pins each
I/O Bank Architecture in Arria 10 Devices
to / from bank above
I/O Bank
Output Path
Input Path
I/O Lane 3
S equencer
Output Path
Input Path
I/O Lane 2
Output Path
Input Path
I/O Lane 1
PLL
C lock Phase
Alignment
Output Path
Input Path
I/O Lane 0
Memory
C ontroller
to / from
FPGA core
to / from bank below
I/O Bank Usage
The pins in an I/O bank can serve as address and command pins, data pins, or clock
and strobe pins for an external memory interface. You can implement a narrow
interface, such as a DDR3 or DDR4 x8 interface, with only a single I/O bank. A wider
interface, such as x72 or x144, can be implemented by configuring multiple adjacent
banks in a multi-bank interface. Any pins in a bank which are not used by the external
memory interface remain available for use as general purpose I/O pins (of the same
voltage standard).
External Memory Interface Handbook Volume 3: Reference Material
150
3 Functional Description—Intel Arria 10 EMIF IP
Every I/O bank includes a hard memory controller which you can configure for DDR3
or DDR4. In a multi-bank interface, only the controller of one bank is active;
controllers in the remaining banks are turned off to conserve power.
To use a multi-bank Arria 10 EMIF interface, you must observe the following rules:
•
Designate one bank as the address and command bank.
•
The address and command bank must contain all the address and command pins.
•
The locations of individual address and command pins within the address and
command bank must adhere to the pin map defined in the pin table— regardless
of whether you use the hard memory controller or not.
•
If you do use the hard memory controller, the address and command bank
contains the active hard controller.
All the I/O banks in a column are capable of functioning as the address and command
bank. However, for minimal latency, you should select the center-most bank of the
interface as the address and command bank.
3.4.4.1 Implementing a x8 Interface with Hard Memory Controller
The following diagram illustrates the use of a single I/O bank to implement a DDR3 or
DDR4 x8 interface using the hard memory controller.
Figure 84.
Single Bank x8 Interface With Hard Controller
Memory
Controller
Sequencer
PLL
Clock Phase
Alignment
Output Path
I/O Lane 3
Input Path
Address/Command Lane 3
Output Path
I/O Lane 2
Input Path
Address/Command Lane 2
Output Path
I/O Lane 1
Input Path
Address/Command Lane 1
Output Path
I/O Lane 0
Input Path
DQ Group 0
In the above diagram, shaded cells indicate resources that are in use.
Note:
For information on the I/O lanes and pins in use, consult the pin table for your device
or the <variation_name>/altera_emif_arch_nf_140/<synth|sim>/
<variation_name>_altera_emif_arch_nf_140_<unique ID>_readme.txt file
generated with your IP.
Related Links
Intel Arria 10 Core Fabric and General Purpose I/Os Handbook
External Memory Interface Handbook Volume 3: Reference Material
151
3 Functional Description—Intel Arria 10 EMIF IP
3.4.4.2 Implementing a x72 Interface with Hard Memory Controller
The following diagram illustrates one possible implementation of a DDR3 or DDR4 x72
interface using the hard memory controller.
Note that only the hard memory controller in the address and command bank is used.
Similarly, only the clock phase alignment block of the address and command bank is
used to generate clock signals for the FPGA core.
Figure 85.
Multi-Bank x72 Interface With Hard Controller
Memory
Controller
Sequencer
PLL
Clock Phase
Alignment
Memory
Controller
Sequencer
PLL
Clock Phase
Alignment
Memory
Controller
Sequencer
PLL
Clock Phase
Alignment
Output Path
I/O Lane 3
Input Path
DQ Group 8
Output Path
I/O Lane 2
Input Path
DQ Group 7
Output Path
I/O Lane 1
Input Path
DQ Group 6
Output Path
I/O Lane 0
Input Path
DQ Group 5
Output Path
I/O Lane 3
Input Path
Address/Command Lane 3
Output Path
I/O Lane 2
Input Path
Address/Command Lane 2
Output Path
I/O Lane 1
Input Path
Address/Command Lane 1
Output Path
I/O Lane 0
Input Path
DQ Group 4
Output Path
I/O Lane 3
Input Path
DQ Group 3
Output Path
I/O Lane 2
Input Path
DQ Group 2
Output Path
I/O Lane 1
Input Path
DQ Group 1
Output Path
I/O Lane 0
Input Path
DQ Group 0
In the above diagram, shaded cells indicate resources that are in use.
External Memory Interface Handbook Volume 3: Reference Material
152
3 Functional Description—Intel Arria 10 EMIF IP
Note:
For information on the I/O lanes and pins in use, consult the pin table for your device
or the <variation_name>/altera_emif_arch_nf_140/<synth|sim>/
<variation_name>_altera_emif_arch_nf_140_<unique ID>_readme.txt file
generated with your IP.
Related Links
Arria 10 Core Fabric and General Purpose I/Os Handbook
3.4.5 Arria 10 EMIF Architecture: I/O Lane
An I/O bank contains 48 I/O pins, organized into four I/O lanes of 12 pins each.
Each I/O lane can implement one x8/x9 read capture group (DQS group), with two
pins functioning as the read capture clock/strobe pair (DQS/DQS#), and up to 10 pins
functioning as data pins (DQ and DM pins). To implement x18 and x36 groups, you
can use multiple lanes within the same bank.
It is also possible to implement a pair of x4 groups in a lane. In this case, four pins
function as clock/strobe pair, and 8 pins function as data pins. DM is not available for
x4 groups. There must be an even number of x4 groups for each interface.
For x4 groups, DQS0 and DQS1 must be placed in the same I/O lane as a pair.
Similarly, DQS2 and DQS3 must be paired. In general, DQS(x) and DQS(x+1) must be
paired in the same I/O lane.
Table 41.
Lanes Used Per Group
Group Size
Number of Lanes Used
Maximum Number of Data Pins per
Group
x8 / x9
1
10
x18
2
22
x36
4
46
pair of x4
1
4 per group, 8 per lane
External Memory Interface Handbook Volume 3: Reference Material
153
3 Functional Description—Intel Arria 10 EMIF IP
Figure 86.
x4 Group
Memory
Controller
Sequencer
PLL
Clock Phase
Alignment
Figure 87.
Output Path
I/O Lane 3
Input Path
X4 Groups 6 and 7
Output Path
I/O Lane 2
Input Path
X4 Groups 4 and 5
Output Path
I/O Lane 1
Input Path
X4 Groups 2 and 3
Output Path
I/O Lane 0
Input Path
X4 Groups 0 and 1
Output Path
I/O Lane 3
Input Path
X8 Group 3
Output Path
I/O Lane 2
Input Path
X8 Group 2
Output Path
I/O Lane 1
Input Path
X8 Group 1
Output Path
I/O Lane 0
Input Path
X8 Group 0
x8 Group
Memory
Controller
Sequencer
PLL
Clock Phase
Alignment
External Memory Interface Handbook Volume 3: Reference Material
154
3 Functional Description—Intel Arria 10 EMIF IP
Figure 88.
x18 Group
M emo ry
Controller
Sequ encer
PLL
Clock Phase
Alignment
Figure 89.
Output Path
Input Path
X18 Group 0
Output Path
Input Path
I/O Lane 3
I/O Lane 2
Output Path
Input Path
X18 Group 1
Output Path
Input Path
I/O Lane 1
Output Path
Input Path
I/O Lane 3
Output Path
Input Path
X36 Group 0
Output Path
Input Path
I/O Lane 2
Output Path
Input Path
I/O Lane 0
I/O Lane 0
x36 Group
M emo ry
Controller
Sequ encer
PLL
Clock Phase
Alignment
I/O Lane 1
3.4.6 Arria 10 EMIF Architecture: Input DQS Clock Tree
The input DQS clock tree is a balanced clock network that distributes the read capture
clock and strobe from the external memory device to the read capture registers inside
the I/Os.
You can configure an input DQS clock tree in x4 mode, x8/x9 mode, x18 mode, or x36
mode.
External Memory Interface Handbook Volume 3: Reference Material
155
3 Functional Description—Intel Arria 10 EMIF IP
Within every bank, only certain physical pins at specific locations can drive the input
DQS clock trees. The pin locations that can drive the input DQS clock trees vary,
depending on the size of the group.
Table 42.
Pins Usable as Read Capture Clock / Strobe Pair
Group Size
Index of Lanes Spanned
by Clock Tree
In-Bank Index of Pins Usable as Read Capture Clock /
Strobe Pair
Positive Leg
Negative Leg
x4
0A
4
5
x4
0B
8
9
x4
1A
16
17
x4
1B
20
21
x4
2A
28
29
x4
2B
32
33
x4
3A
40
41
x4
3B
44
45
x8 / x9
0
4
5
x8 / x9
1
16
17
x8 / x9
2
28
29
x8 / x9
3
40
41
x18
0, 1
12
13
x18
2, 3
36
37
x36
0, 1, 2, 3
20
21
3.4.7 Arria 10 EMIF Architecture: PHY Clock Tree
Dedicated high-speed clock networks drive I/Os in Arria 10 EMIF. Each PHY clock
network spans only one bank.
The relatively short span of the PHY clock trees results in low jitter and low duty-cycle
distortion, maximizing the data valid window.
The PHY clock tree in Arria 10 devices can run as fast as 1.3 GHz. All Arria 10 external
memory interfaces use the PHY clock trees.
3.4.8 Arria 10 EMIF Architecture: PLL Reference Clock Networks
Each I/O bank includes a PLL that can drive the PHY clock trees of that bank, through
dedicated connections. In addition to supporting EMIF-specific functions, such PLLs
can also serve as general-purpose PLLs for user logic.
Arria 10 external memory interfaces that span multiple banks use the PLL in each
bank. (Previous device families relied on a single PLL with clock signals broadcast to
all I/Os via a clock network.) The Arria 10 architecture allows for relatively short PHY
clock networks, reducing jitter and duty-cycle distortion.
External Memory Interface Handbook Volume 3: Reference Material
156
3 Functional Description—Intel Arria 10 EMIF IP
In a multi-bank interface, the clock outputs of individual PLLs must remain in phase;
this is achieved by the following mechanisms:
•
A single PLL reference clock source feeds all PLLs. The reference clock signal
reaches the PLLs by a balanced PLL reference clock tree. The Quartus Prime
software automatically configures the PLL reference clock tree so that it spans the
correct number of banks.
•
The IP sets the PLL M and N values appropriately to maintain synchronization
among the clock dividers across the PLLs. This requirement restricts the legal PLL
reference clock frequencies for a given memory interface frequency and clock rate.
The Arria 10 EMIF IP parameter editor automatically calculates and displays the
set of legal PLL reference clock frequencies. If you plan to use an on-board
oscillator, you must ensure that its frequency matches the PLL reference clock
frequency that you select from the displayed list. The correct M and N values of
the PLLs are set automatically based on the PLL reference clock frequency that
you select.
Note:
The PLL reference clock pin may be placed in the address and command I/O bank or in
a data I/O bank, there is no implication on timing.
Figure 90.
PLL Balanced Reference Clock Tree
I/O Bank
PLL
PHY
clock
tree
I/O Bank
I/O Column
ref_clk
Balanced Reference Clock Network
PLL
PHY
clock
tree
I/O Bank
PLL
PHY
clock
tree
I/O Bank
PLL
PHY
clock
tree
3.4.9 Arria 10 EMIF Architecture: Clock Phase Alignment
In Arria 10 external memory interfaces, a global clock network clocks registers inside
the FPGA core, and the PHY clock network clocks registers inside the FPGA periphery.
Clock phase alignment circuitry employs negative feedback to dynamically adjust the
phase of the core clock signal to match the phase of the PHY clock signal.
The clock phase alignment feature effectively eliminates the clock skew effect in all
transfers between the core and the periphery, facilitating timing closure. All Arria 10
external memory interfaces employ clock phase alignment circuitry.
External Memory Interface Handbook Volume 3: Reference Material
157
3 Functional Description—Intel Arria 10 EMIF IP
Figure 91.
Clock Phase Alignment Illustration
FPGA Periphery
FPGA Core
PHY Clock
Network
PLL
Core Clock
Network
Clock Phase Alignment
p
Figure 92.
+t
Effect of Clock Phase Alignment
Without Clock Phase Alignment
With Clock Phase Alignment
Core clock
Core clock
PHY clock
PHY clock
Skew between core
and PHY clock network
Core and PHY clocks aligned dynamically
by clock phase alignment
3.5 Hardware Resource Sharing Among Multiple EMIFs
Often, it is necessary or desirable to share resources between interfaces.
The following topics explain which hardware resources can be shared, and provide
guidance for doing so.
3.5.1 I/O Aux Sharing
The I/O Aux contains a hard Nios-II processor and dedicated memory storing the
calibration software code and data.
External Memory Interface Handbook Volume 3: Reference Material
158
3 Functional Description—Intel Arria 10 EMIF IP
When a column contains multiple memory interfaces, the hard Nios-II processor
calibrates each interface serially. Interfaces placed within the same I/O column always
share the same I/O Aux. The Quartus Prime Fitter handles I/O Aux sharing
automatically.
3.5.2 I/O Bank Sharing
Data lanes from multiple compatible interfaces can share a physical I/O bank to
achieve a more compact pin placement. To share an I/O bank, interfaces must use the
same memory protocol, rate, frequency, I/O standard, and PLL reference clock signal.
Rules for Sharing I/O Banks
•
A bank cannot serve as the address and command bank for more than one
interface. This means that lanes which implement address and command pins for
different interfaces cannot be allocated to the same physical bank.
Note: An exception to the above rule exists when two interfaces are configured in
a Ping-Pong PHY fashion. In such a configuration, two interfaces share the
same set of address and command pins, effectively meaning that they share
the same address and command tile.
•
Pins within a lane cannot be shared by multiple memory interfaces.
•
Pins that are not used by EMIF IP can serve as general-purpose I/Os of compatible
voltage and termination settings.
•
You can configure a bank as LVDS or as EMIF, but not both at the same time.
•
Interfaces that share banks must reside at consecutive bank locations.
The following diagram illustrates two x16 interfaces sharing an I/O bank. The two
interfaces share the same clock phase alignment block, so that one core clock signal
can interact with both interfaces. Without sharing, the two interfaces would occupy a
total of four physical banks instead of three.
External Memory Interface Handbook Volume 3: Reference Material
159
3 Functional Description—Intel Arria 10 EMIF IP
Figure 93.
I/O Bank Sharing
Memory
Controller
Sequencer
PLL
Clock Phase
Alignment
Memory
Controller
Sequencer
PLL
Clock Phase
Alignment
Memory
Controller
Sequencer
PLL
Clock Phase
Alignment
Output Path
I/O Lane 3
Input Path
Address/Command Lane 2
Output Path
I/O Lane 2
Input Path
Address/Command Lane 1
Output Path
I/O Lane 1
Input Path
Address/Command Lane 0
Output Path
I/O Lane 0
Input Path
DQ Group 0
Output Path
I/O Lane 3
Input Path
DQ Group 1
Output Path
I/O Lane 2
Interface 1
Input Path
Output Path
I/O Lane 1
Input Path
Output Path
I/O Lane 0
Input Path
DQ Group 1
Output Path
I/O Lane 3
Input Path
Address/Command Lane 2
Output Path
I/O Lane 2
Input Path
Address/Command Lane 1
Output Path
I/O Lane 1
Input Path
Address/Command Lane 0
Output Path
I/O Lane 0
Input Path
DQ Group 0
Interface 2
3.5.3 PLL Reference Clock Sharing
In Arria 10, every I/O bank contains a PLL, meaning that it is not necessary to share
PLLs in the interest of conserving resources. Nonetheless, it is often desirable to share
PLLs for other reasons.
External Memory Interface Handbook Volume 3: Reference Material
160
3 Functional Description—Intel Arria 10 EMIF IP
You might want to share PLLs between interfaces for the following reasons:
•
To conserve pins.
•
When combined with the use of the balanced PLL reference clock tree, to allow the
clock signals at different interfaces to be synchronous and aligned to each other.
For this reason, interfaces that share core clock signals must also share the PLL
reference clock signal.
To implement PLL reference clock sharing, open your RTL and connect the PLL
reference clock signal at your design's top-level to the PLL reference clock port of
multiple interfaces.
To share a PLL reference clock, the following requirements must be met:
•
Interfaces must expect a reference clock signal of the same frequency.
•
Interfaces must be placed in the same column.
•
Interfaces must be placed at adjacent bank locations.
3.5.4 Core Clock Network Sharing
It is often desirable or necessary for multiple memory interfaces to be accessible using
a single clock domain in the FPGA core.
You might want to share core clock networks for the following reasons:
•
To minimize the area and latency penalty associated with clock domain crossing.
•
To minimize consumption of core clock networks.
Multiple memory interfaces can share the same core clock signals under the following
conditions:
•
The memory interfaces have the same protocol, rate, frequency, and PLL reference
clock source.
•
The interfaces reside in the same I/O column.
•
The interfaces reside in adjacent bank locations.
For multiple memory interfaces to share core clocks, you must specify one of the
interfaces as master and the remaining interfaces as slaves. Use the Core clocks
sharing setting in the parameter editor to specify the master and slaves.
In your RTL, connect the clks_sharing_master_out signal from the master
interface to the clks_sharing_slave_in signal of all the slave interfaces. Both the
master and slave interfaces expose their own output clock ports in the RTL (e.g.
emif_usr_clk, afi_clk), but the signals are equivalent, so it does not matter
whether a clock port from a master or a slave is used.
Core clock sharing necessitates PLL reference clock sharing; therefore, only the
master interface exposes an input port for the PLL reference clock. All slave interfaces
use the same PLL reference clock signal.
External Memory Interface Handbook Volume 3: Reference Material
161
3 Functional Description—Intel Arria 10 EMIF IP
3.6 Arria 10 EMIF IP Component
The external memory interface IP component for Arria 10 provides a complete solution
for implementing DDR3, DDR4, and QDR-IV external memory interfaces. The EMIF IP
also includes a protocol-specific calibration algorithm that automatically determines
the optimal delay settings for a robust external memory interface.
The external memory interface IP comprises the following parts:
•
A set of synthesizable files that you can integrate into a larger design
•
A stand-alone synthesizable example design that you can use for hardware
validation
•
A set of simulation files that you can incorporate into a larger project
•
A stand-alone simulation example project that you can use to observe controller
and PHY operation
•
A set of timing scripts that you can use to determine the maximum operating
frequency of the memory interface based on external factors such as board skew,
trace delays, and memory component timing parameters
•
A customized data sheet specific to your memory interface configuration
3.6.1 Instantiating Your Arria 10 EMIF IP in a Qsys Project
The following steps describe how to instantiate your Arria 10 EMIF IP in a Qsys
project.
1.
Within the Qsys interface, select Memories and Memory Controllers in the
component Library tree.
2. Under Memories and Memory Controllers, select External Memory
Interfaces (Arria 10).
3. Under External Memory Interfaces (Arria 10), select the Arria 10 External
Memory Interface component.
External Memory Interface Handbook Volume 3: Reference Material
162
3 Functional Description—Intel Arria 10 EMIF IP
Figure 94.
Instantiating Arria 10 EMIF IP in Qsys
3.6.1.1 Logical Connections
The following logical connections exist in an Arria 10 EMIF IP core.
Table 43.
Logical Connections Table
Logical Connection
afi_conduit_end
(Conduit)
afi_clk_conduit_end
(Conduit)
afi_half_clk_conduit_end
(Conduit)
afi_reset_n_conduit_end
(Conduit)
cal_debug_avalon_slave
(Avalon Slave/Target)
cal_debug_clk_clock_sink
(Clock Input)
cal_debug_reset_reset_sink
(Reset Input)
Description
The Altera PHY Interface (AFI) connects a memory controller to the PHY. This
interface is exposed only when you configure the memory interface in PHY-Only
mode. The interface is synchronous to the afi_clk clock and afi_reset_n
reset.
Use this clock signal to clock the soft controller logic. The afi_clk is an output
clock coming from the PHY when the memory interface is in PHY-Only mode. The
phase of afi_clk is adjusted dynamically by hard circuitry for the best data
transfer between FPGA core logic and periphery logic with maximum timing
margin. Multiple memory interface instances can share a single afi_clk using
the Core Clocks Sharing option during IP generation.
This clock runs at half the frequency of afi_lk. It is exposed only when the
memory interface is in PHY-Only mode.
This single-bit reset provides a synchronized reset output. Use this signal to reset
all registers that are clocked by either afi_clk or afi_half_clk.
This interface is exposed when the EMIF Debug Toolkit/On-chip Debug Port
option is set to Export. This interface can be connected to an Arria 10 External
Memory Interface Debug Component to allow EMIF Debug Toolkit access, or it can
be used directly by user logic to access calibration diagnostic data.
This clock is used for the cal_debug_avalon_slave interface. It can be
connected to the emif_usr_clk_clock_source interface.
This reset is used for the cal_debug_avalon_slave interface. It can be
connected to the emif_usr_reset_reset_source interface.
continued...
External Memory Interface Handbook Volume 3: Reference Material
163
3 Functional Description—Intel Arria 10 EMIF IP
Logical Connection
Description
(Avalon Master/Source)
This interface is exposed when the Enable Daisy-Chaining for EMIF Debug
Toolkit/On-Chip Debug Port option is enabled. Connect this interface to the
cal_debug_avalon_slave interface of the next EMIF instance in the same I/O
column.
cal_debug_out_clk_clock_sourc
e
This interface should be connected to the cal_debug_clk_clock_sink
interface of the next EMIF instance in the same I/O column (similar to
cal_debug_out_avalon_master).
cal_debug_out_avalon_master
(Clock Output)
cal_debug_out_reset_reset_sou
rce
(Reset Output)
This interface should be connected to the cal_debug_reset_reset_sink
interface of the next EMIF instance in the same I/O column (similar to
cal_debug_out_avalon_master).
(Avalon Slave/Target)
This interface allows access to the Efficiency Monitor CSR. For more information,
see the documentation on the UniPHY Efficiency Monitor.
global_reset_reset_sink
This single-wire input port is the asynchronous reset input for the EMIF core.
effmon_csr_avalon_slave
(Reset Input)
pll_ref_clk_clock_sink
(Clock Input)
oct_conduit_end
(Conduit)
mem_conduit_end
(Conduit)
status_conduit_end
(Conduit)
emif_usr_reset_reset_source
(Reset Output)
emif_usr_clk_clock_source
(Clock Output)
ctrl_amm_avalon_slave
(Avalon Slave/Target)
This single-wire input port connects the external PLL reference clock to the EMIF
core. Multiple EMIF cores may share a PLL reference clock source, provided the
restrictions outlined in the PLL and PLL Reference Clock Network section are
observed.
This logical port is connected to an OCT pin and provides calibrated reference data
for EMIF cores with pins that use signaling standards that require on-chip
termination. Depending on the I/O standard, reference voltage, and memory
protocol, multiple EMIF cores may share a single OCT pin.
This logical conduit can attach an Altera Memory Model to an EMIF core for
simulation. Memory models for various protocols are available under the
Memories and Memory Controllers — External Memory Interfaces —
Memory Models section of the component library in Qsys. You must ensure that
all configuration parameters for the memory model match the configuration
parameters of the EMIF core.
The status conduit exports two signals that can be sampled to determine if the
calibration operation passed or failed for that core.
This single-bit reset output provides a synchronized reset output that should be
used to reset all components that are synchronously connected to the EMIF core.
Assertion of the global reset input triggers an assertion of this output as well,
therefore you should rely on this signal only as a reset source for all components
connected to the EMIF core.
Use this single-bit clock output to clock all logic connected to the EMIF core. The
phase of this clock signal is adjusted dynamically by circuitry in the EMIF core
such that data can be transferred between core logic and periphery registers with
maximum timing margin. Drive all logic connected to the EMIF core with this clock
signal. Other clock sources generated from the same reference clock or even the
same PLL may have unknown phase relationships. Multiple EMIF cores can share a
single core clock using the Core Clocks Sharing option described in the Example
Design tab of the parameter editor.
This Avalon target port initiates read or write commands to the controller. Refer to
the Avalon Interface Specification for more information on how to design cores
that comply to the Avalon Bus Specification.
For DDR3, DDR4, and LPDDR3 protocols with the hard PHY and hard controller
configuration and an AVL slave interface exposed, ctrl_amm_avalon_slave is
renamed to crtl_amm_avalon_slave_0.
For QDR II, QDR II+, and QDR II+ Xtreme interfaces with hard PHY and soft
controller, separate read and write connections are used.
ctrl_amm_avalon_slave_0 is the read port and ctrl_amm_avalon_slave_1
is the write port.
For QDR-IV interfaces with hard PHY and soft controller operating at quarter rate,
a total of eight separate Avalon interfaces (named ctrl_amm_avalon_slave_0
to ctrl_amm_avalon_slave_7) are used to maximize bus efficiency.
External Memory Interface Handbook Volume 3: Reference Material
164
3 Functional Description—Intel Arria 10 EMIF IP
3.6.2 File Sets
The Arria 10 EMIF IP core generates four output file sets for every EMIF IP core,
arranged according to the following directory structure.
Table 44.
Generated File Sets
Directory
Description
This directory contains only the files required to integrate a
generated EMIF core into a larger design. This directory contains:
• Synthesizable HDL source files
• Customized TCL timing scripts specific to the core (protocol and
topology)
• HEX files used by the calibration algorithm to identify the
interface parameters
• A customized data sheet that describes the operation of the
generated core
<core_name>/*
Note: The top-level HDL file is generated in the root folder as
<core_name>.v (or <core_name.vhd> for VHDL designs).
You can reopen this file in the parameter editor if you want
to modify the EMIF core parameters and regenerate the
design.
<core_name>_sim/*
This directory contains the simulation fileset for the generated EMIF
core. These files can be integrated into a larger simulation project.
For convenience, simulation scripts for compiling the core are
provided in the /mentor, /cadence, /synopsys, and /riviera
subdirectories.
The top-level HDL file, <core_name>.v (or <core_name>.vhd) is
located in this folder, and all remaining HDL files are placed in the /
altera_emif_arch_nf subfolder, with the customized data sheet.
The contents of this directory are not intended for synthesis.
emif_<instance_num>_example_design/*
This directory contains a set of TCL scripts, QSYS project files and
README files for the complete synthesis and simulation example
designs. You can invoke these scripts to generate a standalone fullysynthesizable project complete with an example driver, or a
standalone simulation design complete with an example driver and a
memory model.
3.6.3 Customized readme.txt File
When you generate your Arria 10 EMIF IP, the system produces a customized readme
file, containing data indicative of the settings in your IP core.
The readme file is <variation_name>/
altera_emif_arch_nf_<version_number>/<synth|sim>/
<variation_name>_altera_emif_arch_nf_<version_number>_<unique
ID>_readme.txt, and contains a summary of Arria 10 EMIF information, and details
specific to your IP core, including:
•
Pin location guidelines
•
External port names, directions, and widths
•
Internal port names, directions, and widths
•
Avalon interface configuration details (if applicable)
•
Calibration mode
•
A brief summary of all configuration settings for the generated IP
External Memory Interface Handbook Volume 3: Reference Material
165
3 Functional Description—Intel Arria 10 EMIF IP
You should review the generated readme file for implementation guidelines specific to
your IP core.
3.6.4 Clock Domains
The Arria 10 EMIF IP core provides a single clock domain to drive all logic connected to
the EMIF core.
The frequency of the clock depends on the core-clock to memory-clock interface rate
ratio. For example, a quarter-rate interface with an 800 MHz memory clock would
provide a 200 MHz clock to the core (800 MHz / 4 = 200 MHz). The EMIF IP
dynamically adjusts the phase of the core clock with respect to the periphery clock to
maintain the optimum alignment for transferring data between the core and periphery.
Independent EMIF IP cores driven from the same reference clock have independent
core clock domains. You should employ one of the following strategies if you are
implementing multiple EMIF cores:
1. Treat all crossing between independent EMIF-generated clock domains as
asynchronous, even though they are generated from the same reference clock.
2. Use the Core clock sharing option to enforce that multiple EMIF cores share the
same core clock. You must enable this option during IP generation. This option is
applicable only for cores that reside in the same I/O column.
3.6.5 ECC in Arria 10 EMIF IP
The ECC (error correction code) is a soft component of the Arria 10 EMIF IP that
reduces the chance of errors when reading and writing to external memory. ECC
allows correction of single-bit errors and reduces the chances of system failure.
The ECC component includes an encoder, decoder, write FIFO buffer, and modification
logic, to allow read-modify-write operations. The ECC code employs standard
Hamming logic to correct single-bit errors and to detect double-bit errors. ECC is
available in 16, 24, 40, and 72 bit widths.
When writing data to memory, the encoder creates ECC bits and writes them together
with the regular data. When reading from memory, the decoder checks the ECC bits
and regular data, and passes the regular data unchanged if no errors are detected. If
a single-bit error is detected, the ECC logic corrects the error and passes the regular
data. If more than a single-bit error is detected, the ECC logic sets a flag to indicate
the error.
Read-modify-write operations can occur in the following circumstances:
•
A partial write in data mask mode with ECC enabled, where at least one memory
burst of byte-enable is not all ones or all zeros.
•
Auto-correction with ECC enabled. This is usually a dummy write issued by the
auto-correction logic to correct the memory content when a single-bit error is
detected. The read-modify-write reads back the data, corrects the single-bit error,
and writes the data back.
The additional overhead associated with read-modify-write operations can severely
reduce memory interface efficiency. For best efficiency, you should design traffic
patterns to avoid read-modify-write operations wherever possible, such as by
minimizing the number of partial writes in ECC mode.
External Memory Interface Handbook Volume 3: Reference Material
166
3 Functional Description—Intel Arria 10 EMIF IP
3.6.5.1 ECC Components
The ECC logic communicates with user logic via an Avalon-MM interface, and with the
hard memory controller via an Avalon-ST interface
ECC Encoder
The ECC encoder consists of a x64/x72 encoder IP core capable of single-bit error
correction and double-bit error detection. The encoder takes 64 bits input and
converts it to 72 bits output, where the 8 additional bits are ECC code. The encoder
supports any input data width less than 64-bits. Any unused input data bits are set to
zero.
ECC Decoder
The ECC decoder consists of a x72/x64 decoder IP core capable of double-bit error
detection. The decoder takes 72 bits input and converts it to 64 bits output. The
decoder also produces single-bit error and double-bit error information. The decoder
controls the user read data valid signal; when read data is intended for partial write,
the user read data valid signal is deasserted, because the read data is meant for
merging, not for the user.
Partial Write Data FIFO Buffer
The Partial Write Data FIFO Buffer is implemented in soft logic to store partial write
data and byte enable. Data and byte enable are popped and merged with the returned
read data. A partial write can occur in the following situations:
•
At least one memory burst of byte enable is not all ones or all zeroes.
•
Non data masked mode, where all memory bursts of byte enable are not all ones.
•
A dummy write with auto-correction logic, where all memory bursts of byte enable
are all zeroes. (You might use a dummy write when correcting memory content
with a single-bit error.
Merging Logic
Merge return partial read data with write data based on byte enabled popped from the
FIFO buffer, and send it to the ECC encoder.
Pointer FIFO Buffer
The pointer FIFO buffer is implemented in soft logic to store write data pointers. The
ECC logic refers to the pointers when sending write data to DBC. The pointers serve to
overwrite existing write data in the data buffer during a read-modify-write process.
Partial Logic
Partial logic decodes byte enable information and distinguishes between normal and
partial writes.
Memory Mode Register Interface
The Memory Mode Register interface is an Avalon-based interface through which core
logic can access debug signals and sideband operation requests in the hard memory
controller.
External Memory Interface Handbook Volume 3: Reference Material
167
3 Functional Description—Intel Arria 10 EMIF IP
The MMR logic routes ECC-related operations to an MMR register implemented in soft
logic, and returns the ECC information via an Avalon-MM interface. The MMR logic
tracks single-bit and double-bit error status, and provides the following information:
•
Interrupt status.
•
Single-bit error and double-bit error status.
•
Single-bit error and double-bit error counts (to a maximum of 15; if more than 15
errors occur, the count will overflow).
•
Address of the last error.
3.6.5.2 ECC User Interface Controls
You can enable the ECC logic from the Configuration, Status, and Error Handling
section of the Controller tab in the parameter editor.
There are three user interface settings related to ECC:
•
Enabled Memory-Mapped Configuration and Status Register (MMR)
Interface: Allows run-time configuration of the memory controller. You can
enable this option together with ECC to retrieve error detection information from
the ECC logic.
•
Enabled Error Detection and Correction Logic: Enables ECC logic for single-bit
error correction and double-bit error detection.
•
Enable Auto Error Correction: Allows the controller to automatically correct
single-bit errors detected by the ECC logic.
3.7 Examples of External Memory Interface Implementations for
DDR4
The following figures are examples of external memory interface implementations for
different DDR4 memory widths. The figures show the locations of the address/
command and data pins in relation to the locations of the memory controllers.
Figure 95.
DDR4 1x8 Implementation Example (One I/O Bank)
1 x 8 Pin (1 Bank)
Data
Address/Command
Controller
Address/Command
Address/Command
External Memory Interface Handbook Volume 3: Reference Material
168
3 Functional Description—Intel Arria 10 EMIF IP
Figure 96.
DDR4 1x32 Implementation Example (Two I/O Banks)
1 x 32 Pin (2 Banks)
Data
Data
Data
Data
Address/Command
Controller
Address/Command
Address/Command
External Memory Interface Handbook Volume 3: Reference Material
169
3 Functional Description—Intel Arria 10 EMIF IP
Figure 97.
DDR4 1x72 Implementation Example (Three I/O Banks)
1 x 72 Pin (3 Banks)
Data
Data
Data
Data
Data
Address/Command
Controller
Address/Command
Address/Command
Data
Data
Data
Data
External Memory Interface Handbook Volume 3: Reference Material
170
3 Functional Description—Intel Arria 10 EMIF IP
Figure 98.
DDR4 2x16 Implementation Example with Controllers in Non-Adjacent Banks
(Three I/O Banks)
2 x 16 Pin (3 Banks)
Controller
1
Address/Command 1
Address/Command 1
Address/Command 1
Data 1
Data 1
Data 2
Data 2
Controller
2
Address/Command 2
Address/Command 2
Address/Command 2
External Memory Interface Handbook Volume 3: Reference Material
171
3 Functional Description—Intel Arria 10 EMIF IP
Figure 99.
DDR4 2x16 Implementation Example with Controllers in Adjacent Banks
(Three I/O Banks)
2 x 16 Pin (3 Banks)
Data 1
Controller
1
Address/Command 1
Address/Command 1
Address/Command 1
Data 1
Controller
2
Address/Command 2
Address/Command 2
Address/Command 2
Data 2
Data 2
External Memory Interface Handbook Volume 3: Reference Material
172
3 Functional Description—Intel Arria 10 EMIF IP
Figure 100. DDR4 1x144 Implementation Example (Six I/O Banks)
1 x 144 Pin (6 Banks)
Data
Data
Data
Data
Data
Data
Data
Data
Data
Data
Address/Command
Controller
Address/Command
Address/Command
Data
Data
Data
Data
Data
Data
Data
Data
3.8 Arria 10 EMIF Sequencer
The Arria 10 EMIF sequencer is fully hardened in silicon, with executable code to
handle protocols and topologies. Hardened RAM contains the calibration algorithm.
External Memory Interface Handbook Volume 3: Reference Material
173
3 Functional Description—Intel Arria 10 EMIF IP
The Arria 10 EMIF sequencer is responsible for the following operations:
•
Initializes memory devices.
•
Calibrates the external memory interface.
•
Governs the hand-off of control to the memory controller.
•
Handles recalibration requests and debug requests.
•
Handles all supported protocols and configurations.
Figure 101. Arria 10 EMIF Sequencer Operation
Start
Discover EMIFs in column
Sequencer
software
Processed all
interfaces?
Data
Yes
No
Initialize external memory
Calibrate interface
Hand-off
House-keeping
tasks
3.8.1 Arria 10 EMIF DQS Tracking
DQS tracking is enabled for QDR II / II+ / QDR II+ Xtreme, RLDRAM 3, and LPDDR3
protocols. DQS tracking is not available for DDR3 and DDR4 protocols.
3.9 Arria 10 EMIF Calibration
The calibration process compensates for skews and delays in the external memory
interface.
The calibration process enables the system to compensate for the effects of factors
such as the following:
•
Timing and electrical constraints, such as setup/hold time and Vref variations.
•
Circuit board and package factors, such as skew, fly-by effects, and manufacturing
variations.
•
Environmental uncertainties, such as variations in voltage and temperature.
•
The demanding effects of small margins associated with high-speed operation.
3.9.1 Calibration Stages
At a high level, the calibration routine consists of address and command calibration,
read calibration, and write calibration.
External Memory Interface Handbook Volume 3: Reference Material
174
3 Functional Description—Intel Arria 10 EMIF IP
The stages of calibration vary, depending on the protocol of the external memory
interface.
Table 45.
Calibration Stages by Protocol
Stage
DDR4
DDR3
LPDDR3
RLDRAM II/3
QDR-IV
QDR II/II+
Address and command
Leveling
Yes
Yes
—
—
—
—
Deskew
Yes
—
Yes
—
Yes
—
DQSen
Yes
Yes
Yes
Yes
Yes
Yes
Deskew
Yes
Yes
Yes
Yes
Yes
Yes
VREF-In
Yes
—
—
—
Yes
—
LFIFO
Yes
Yes
Yes
Yes
Yes
Yes
Leveling
Yes
Yes
Yes
Yes
Yes
—
Deskew
Yes
Yes
Yes
Yes
Yes
Yes
VREF-Out
Yes
—
—
—
—
—
Read
Write
3.9.2 Calibration Stages Descriptions
The various stages of calibration perform address and command calibration, read
calibration, and write calibration.
Address and Command Calibration
The goal of address and command calibration is to delay address and command
signals as necessary to optimize the address and command window. This stage is not
available for all protocols, and cannot compensate for an inefficient board design.
Address and command calibration consists of the following parts:
•
Leveling calibration— Centers the CS# signal and the entire address and
command bus, relative to the CK clock. This operation is available only for DDR3
and DDR4 interfaces.
•
Deskew calibration— Provides per-bit deskew for the address and command bus
(except CS#), relative to the CK clock. This operation is available for DDR4 and
QDR-IV interfaces only.
External Memory Interface Handbook Volume 3: Reference Material
175
3 Functional Description—Intel Arria 10 EMIF IP
Read Calibration
Read calibration consists of the following parts:
•
DQSen calibration— Calibrates the timing of the read capture clock gating and
ungating, so that the PHY can gate and ungate the read clock at precisely the
correct time—if too early or too late, data corruption can occur. The algorithm for
this stage varies, depending on the memory protocol.
•
Deskew calibration— Performs per-bit deskew of read data relative to the read
strobe or clock.
•
VREF-IN calibration— Calibrates the Vref level at the FPGA.
•
LFIFO calibration: Normalizes differences in read delays between groups due to
fly-by, skews, and other variables and uncertainties.
Write Calibration
Write calibration consists of the following parts:
•
Leveling calibration— Aligns the write strobe and clock to the memory clock, to
compensate for skews, especially those associated with fly-by topology. The
algorithm for this stage varies, depending on the memory protocol.
•
Deskew calibration— Performs per-bit deskew of write data relative to the write
strobe and clock.
•
VREF-Out calibration— Calibrates the VREF level at the memory device.
3.9.3 Calibration Algorithms
The calibration algorithms sometimes vary, depending on the targeted memory
protocol.
Address and Command Calibration
Address and command calibration consists of the following parts:
External Memory Interface Handbook Volume 3: Reference Material
176
3 Functional Description—Intel Arria 10 EMIF IP
•
Leveling calibration— (DDR3 and DDR4 only) Toggles the CS# and CAS# signals
to send read commands while keeping other address and command signals
constant. The algorithm monitors for incoming DQS signals, and if the DQS signal
toggles, it indicates that the read commands have been accepted. The algorithm
then repeats using different delay values, to find the optimal window.
•
Deskew calibration— (DDR4, QDR-IV, and LPDDR3 only)
—
(DDR4) Uses the DDR4 address and command parity feature. The FPGA sends
the address and command parity bit, and the DDR4 memory device responds
with an alert signal if the parity bit is detected. The alert signal from the
memory device tells the FPGA that the parity bit was received.
Deskew calibration requires use of the PAR/ALERT# pins, so you should not
omit these pins from your design. One limitation of deskew calibration is that
it cannot deskew ODT and CKE pins.
—
(QDR-IV) Uses the QDR-IV loopback mode. The FPGA sends address and
command signals, and the memory device sends back the address and
command signals which it captures, via the read data pins. The returned
signals indicate to the FPGA what the memory device has captured. Deskew
calibration can deskew all synchronous address and command signals.
—
(LPDDR3) Uses the LPDDR3 CA training mode. The FPGA sends signals onto
the LPDDR3 CA bus, and the memory device sends back those signals that it
captures, via the DQ pins. The returned signals indicate to the FPGA what the
memory device has captured. Deskew calibration can deskew all signals on the
CA bus. The remaining command signals (CS, CKE, and ODT) are calibrated
based on the average of the deskewed CA bus.
External Memory Interface Handbook Volume 3: Reference Material
177
3 Functional Description—Intel Arria 10 EMIF IP
Read Calibration
•
DQSen calibration— (DDR3, DDR4, LPDDR3, RLDRAMx and QDRx) DQSen
calibration occurs before Read deskew, therefore only a single DQ bit is required to
pass in order to achieve a successful read pass.
—
(DDR3, DDR4,and LPDDR3) The DQSen calibration algorithm searches the
DQS preamble using a hardware state machine. The algorithm sends many
back-to-back reads with a one clock cycle gap between. The hardware state
machine searches for the DQS gap while sweeping DQSen delay values. the
algorithm then increments the VFIFO value, and repeats the process until a
pattern is found. The process is then repeated for all other read DQS groups.
—
(RLDRAMx and QDRx) The DQSen calibration algorithm does not use a
hardware state machine; rather, it calibrates cycle-level delays using software
and subcycle delays using DQS tracking hardware. The algorithm requires
good data in memory, and therefore relies on guaranteed writes. (Writing a
burst of 0s to one location, and a burst of 1s to another; back-to-back reads
from these two locations are used for read calibration.)
The algorithm enables DQS tracking to calibrate the phase component of DQS
enable. It then issues a guaranteed write, followed by back-to-back reads. The
algorithm sweeps DQSen values cycle by cycle until the read operation
succeeds. The process is then repeated for all other read groups.
•
Deskew calibration— Read deskew calibration is performed before write leveling,
and must be performed at least twice: once before write calibration, using simple
data patterns from guaranteed writes, and again after write calibration, using
complex data patterns.
The deskew calibration algorithm performs a guaranteed write, and then sweeps
dqs_in delay values from low to high, to find the right edge of the read window.
The algorithm then sweeps dq-in delay values low to high, to find the left edge of
the read window. Updated dqs_in and dq_in delay values are then applied to
center the read window. The algorithm then repeats the process for all data pins.
•
Vref-In calibration— Read Vref-In calibration begins by programming Vref-In
with an arbitrary value. The algorithm then sweeps the Vref-In value from the
starting value to both ends, and measures the read window for each value. The
algorithm selects the Vref-In value which provides the maximum read window.
•
LFIFO calibration— Read LFIFO calibration normalizes read delays between groups.
The PHY must present all data to the controller as a single data bus. The LFIFO
latency should be large enough for the slowest read data group, and large enough
to allow proper synchronization across FIFOs.
External Memory Interface Handbook Volume 3: Reference Material
178
3 Functional Description—Intel Arria 10 EMIF IP
Write Calibration
•
Leveling calibration— Write leveling calibration aligns the write strobe and clock to
the memory clock, to compensate for skews. In general, leveling calibration tries a
variety of delay values to determine the edges of the write window, and then
selects an appropriate value to center the window. The details of the algorithm
vary, depending on the memory protocol.
—
(DDRx, LPDDR3) Write leveling occurs before write deskew, therefore only one
successful DQ bit is required to register a pass. Write leveling staggers the DQ
bus to ensure that at least one DQ bit falls within the valid write window.
—
(RLDRAMx) Optimizes for the CK versus DK relationship.
—
(QDR-IV) Optimizes for the CK versus DK relationship. Is covered by address
and command deskew using the loopback mode.
—
(QDR II/II+/Xtreme) The K clock is the only clock, therefore write leveling is
not required.
•
Deskew calibration— Performs per-bit deskew of write data relative to the write
strobe and clock. Write deskew calibration does not change dqs_out delays; the
write clock is aligned to the CK clock during write leveling.
•
VREF-Out calibration— (DDR4) Calibrates the VREF level at the memory device.
The VREF-Out calibration algorithm is similar to the VREF-In calibration algorithm.
3.9.4 Calibration Flowchart
The following flowchart illustrates the calibration flow.
Figure 102. Calibration Flowchart
External Memory Interface Handbook Volume 3: Reference Material
179
3 Functional Description—Intel Arria 10 EMIF IP
3.9.5 Periodic OCT Recalibration
Periodic OCT recalibration improves the accuracy of the on-chip termination values
used by DDR4 Pseudo-open Drain (POD) I/Os. This feature periodically invokes the
user-mode OCT calibration engine and updates the I/O buffer termination settings to
compensate for variations in calibrated OCT settings caused by large changes in
device operating temperature.
This feature is automatically enabled for DDR4 memory interfaces unless the IP does
not meet the technical requirements, or if you explicitly disable the feature in the
parameter editor.
3.9.5.1 Operation
The Periodic OCT recalibration engine refreshes the calibrated OCT settings for DDR4
I/O buffers every 500ms. To ensure data integrity, there is a momentary pause in user
traffic as the OCT settings are refreshed; however, the process of OCT calibration is
decoupled from the actual update to the I/O buffers, to minimize disruption of user
traffic.
The calibration process uses the external RZQ reference resistor to determine the
optimal settings for the I/O buffer, to meet the specified calibrated I/O standards on
the FPGA. OCT Calibration only affects the I/O pin that is connected to the RZQ
resistor; therefore, memory traffic is not interrupted during the calibration phase.
Upon completion of the calibration process, the updated calibration settings are
applied to the I/O buffers. The memory traffic is halted momentarily by placing the
memory into self-refresh mode; this ensures that the data bus is idle and no glitches
are created by the I/O buffers during the buffer update. The buffer is updated as soon
as the memory enters self-refresh mode. The memory interface exits self-refresh
mode when the buffer update is complete and new read or write requests are detected
on the Avalon bus. The controller remains in self-refresh mode until a new command
is detected. OCT calibration continues to occur even if the memory is still in self
refresh mode. Upon detection of a new command, the controller issues a self-refresh
exit command to the memory, followed by a memory-side ZQ calibration short
duration (ZQCS) command. Memory traffic resumes when the memory DLL has relocked.
If you disable the periodic OCT recalibration engine, the calibration process occurs
only once during device configuration. In this operating mode, the calibrated OCT
settings can vary across temperature as specified by the calibration accuracy ranges
listed in the Arria 10 Device Handbook. The DDR external timing report automatically
factors in the effect of enabling or disabling the periodic OCT recalibration engine
when calculating the total amount of external I/O transfer margin.
3.9.5.2 Technical Restrictions
Certain criteria must be met in order to use periodic OCT recalibration.
External Memory Interface Handbook Volume 3: Reference Material
180
3 Functional Description—Intel Arria 10 EMIF IP
The periodic OCT recalibration engine is enabled only when all of the following criteria
are met:
•
The memory interface is configured to use the Altera Hard Memory Controller for
DDR4.
•
The memory interface is configured for either DDR4 UDIMM or component
topologies. RDIMM and LRDIMM topologies are not supported.
•
The memory interface is not used with the hardened processor subsystem.
•
The memory interface does not use Ping-Pong PHY.
•
The memory interface does not use calibrated I/O standards for address,
command, or clock signals.
•
The memory interface uses calibrated I/O standards for the data bus.
•
The memory does not use the memory mapped register (MMR) interface of the
HMC, including ECC modes.
•
You have not explicitly disabled periodic OCT recalibration in the parameter editor.
•
The specified device is a production level device (that is, not an ES/ES2/ES3 class
device).
Periodic OCT recalibration requires that each EMIF instance in the design employ a
dedicated RZQ resistor. Because this restriction cannot be detected at IP generation
time, you must explicitly disable the periodic OCT recalibration engine for a given
interface if it shares an RZQ resistor with another interface. Ensure that you observe
this restriction when automatically upgrading EMIF IP from older versions of the
Quartus Prime software.
3.9.5.3 Efficiency Impact
The Periodic OCT recalibration engine must interrupt user traffic for a short period of
time in order to update I/O buffer termination settings.
External Memory Interface Handbook Volume 3: Reference Material
181
3 Functional Description—Intel Arria 10 EMIF IP
The exact flow of operations executed by the recalibration engine that affects memory
traffic is described below:
1.
Enter Self-Refresh Mode. The EMIF calibration CPU triggers self-refresh entry on
the hard memory controller. The controller flushes all pending operations,
precharges all banks and issues the self-refresh command. This operation
introduces a delay of approximately 25 Memory clock cycles (precharge all and
self-refresh entry commands).
2. Confirm Self-Refresh Mode. The EMIF calibration CPU polls the hard memory
controller to confirm that the clocks have stopped. This operation introduces no
delay.
3.
Issue codeword update. The EMIF calibration CPU triggers user-mode OCT logic to
update code words. This operation introduces a delay of 50-100ns, depending on
the device speed grade.
4.
Allow Exit Self-Refresh Mode. The EMIF calibration CPU enables automatic selfrefresh exit logic. This operation introduces a delay of 50-100ns, depending on the
device speed grade.
5. Wait for Memory Traffic. The hard memory controller waits for an incoming read or
write command on the Avalon bus. The delay introduced by this operation varies,
depending on the user application.
6.
Exit Self Refresh Mode. The hard memory controller issues the Self-Refresh Exit
command and a simultaneous memory-side RZQ calibration (ZQCS) command.
The delay introduced by this operation varies according to the device speed bin
(up to ~1000 memory clock cycles for fastest memory devices).
The efficiency impact on throughput-sensitive work loads is less than one percent,
even under worst-case scenarios with all banks active. However, be aware that the
first command issued after the hard memory controller exits self-refresh mode will
incur the latency overhead of waiting for the memory DLL to re-lock when the SelfRefresh Exit command is issued by the hard memory controller. Contact Intel FPGA
Technical Services for information on how to manually trigger or inhibit periodic OCT
updates for applications that are sensitive to latency.
3.10 Back-to-Back User-Controlled Refresh Usage in Arria 10
The following diagram illustrates the back-to-back refresh model for optimized hard
memory controller (HMC) performance in Arria 10 devices.
For optimal performance, ensure that you deassert the Refresh request after receiving
the acknowledgement pulse. You can implement a timer to track tRFC before asserting
the next Refresh request. Failure to deassert the Refresh request can delay memory
access to the rank not in refresh.
External Memory Interface Handbook Volume 3: Reference Material
182
3 Functional Description—Intel Arria 10 EMIF IP
3.11 Arria 10 EMIF and SmartVID
Arria 10 EMIF IP can be used with the SmartVID voltage management system, to
achieve reduced power consumption.
Note:
Arria 10 HPS EMIF IP does not currently support SmartVID.
The SmartVID controller allows the FPGA to operate at a reduced Vcc, while
maintaining performance. Because the SmartVID controller can adjust Vcc up or down
in response to power requirements and temperature, it can have an impact on
external memory interface performance. When used with the SmartVID controller, the
EMIF IP implements a handshake protocol to ensure that EMIF calibration does not
begin until after voltage adjustment has completed.
In extended speed grade devices, voltage adjustment occurs once when the FPGA is
powered up, and no further voltage adjustments occur. The external memory
calibration occurs after this initial voltage adjustment is completed. EMIF specifications
are expected to be slightly lower in extended speed grade devices using SmartVID,
than in devices not using SmartVID.
In industrial speed grade devices, voltage adjustment occurs at power up, and may
also occur during operation, in response to temperature changes. External memory
interface calibration does not occur until after the initial voltage adjustment at power
up. However, the external memory interface is not recalibrated in response to
subsequent voltage adjustments that occur during operation. As a result, EMIF
specifications for industrial speed grade devices using SmartVID are expected to be
lower than for extended speed grade devices.
Using Arria 10 EMIF IP with SmartVID
To employ Arria 10 EMIF IP with SmartVID, follow these steps:
1.
Ensure that the Quartus Prime project and Qsys system are configured to use VID
components. This step exposes the vid_cal_done_persist interface on
instantiated EMIF IP, which is required for communicating with the SmartVID
controller.
2. Instantiate the SmartVID controller, using an I/O PLL IP core to drive the 125MHz
vid_clk and the 25MHz jtag_core_clk inputs of the Smart VID controller.
Note: Do not connect the emif_usr_clk signal to either the vid_clk or
jtag_core_clk inputs. Doing so would hold both the EMIF IP and the
SmartVID controller in a perpetual reset condition.
3. Instantiate the Arria 10 EMIF IP.
4. Connect the vid_cal_done_persist signal from the EMIF IP with the
cal_done_persistent signal on the SmartVID controller. This connection
enables handshaking between the EMIF IP and the SmartVID controller, which
allows the EMIF IP to delay memory calibration until after voltage levels are
stabilized.
Note: The EMIF vid_cal_done_persist interface becomes available only when
a VID-enabled device is selected.
Related Links
SmartVID Controller IP Core User Guide
External Memory Interface Handbook Volume 3: Reference Material
183
3 Functional Description—Intel Arria 10 EMIF IP
3.12 Hard Memory Controller Rate Conversion Feature
The hard memory controller's rate conversion feature allows the hard memory
controller and PHY to run at half-rate, even though user logic is configured to run at
quarter-rate.
To facilitate timing closure, you may choose to clock your core user logic at quarterrate, resulting in easier timing closure at the expense of increased area and latency.
To improve efficiency and help reduce overall latency, you can run the hard memory
controller and PHY at half rate.
The rate conversion feature converts traffic from the FPGA core to the hard memory
controller from quarter-rate to half-rate, and traffic from the hard memory controller
to the FPGA core from half-rate to quarter-rate. From the perspective of user logic
inside the FPGA core, the effect is the same as if the hard memory controller were
running at quarter-rate.
The rate conversion feature is enabled automatically during IP generation whenever all
of the following conditions are met:
•
The hard memory controller is in use.
•
User logic runs at quarter-rate.
•
The interface targets either an ES2 or production device.
•
Running the hard memory controller at half-rate dpoes not exceed the fMax
specification of the hard memory controller and hard PHY.
When the rate conversion feature is enabled, you should see the following info
message displayed in the IP generation GUI:
PHY and controller running at 2x the frequency of user logic for
improved efficiency.
Related Links
Arria 10 Core Fabric and General Purpose I/Os Handbook
3.13 Back-to-Back User-Controlled Refresh for Hard Memory
Controller
The following waveform illustrates the recommended Arria 10 model for back-to-back
user-controlled refreshes, for optimized hard memory controller performance.
Figure 103.
External Memory Interface Handbook Volume 3: Reference Material
184
3 Functional Description—Intel Arria 10 EMIF IP
You should deassert the refresh request after the refresh acknowledgement pulse is
received. You can implement a timer to keep track of the tRFC status before asserting
a refresh request. Failure to deassert the Refresh request can delay access to the rank
not in refresh.
3.14 Compiling Arria 10 EMIF IP with the Quartus Prime Software
3.14.1 Instantiating the Arria 10 EMIF IP
Depending on your work flow, you may instantiate your IP with Qsys or with the IP
Catalog.
Instantiating with Qsys
If you instantiate your IP as part of a system in Qsys, follow the Qsys documentation
for information on instantiating the IP in a Quartus Prime project.
Instantiating with the IP Catalog
If you generated your IP with the IP Catalog, you must add the Quartus Prime IP file
(.qip) to your Quartus Prime project. The .qip file identifies the names and
locations of the files that compose the IP. After you add the .qip file to your project,
you can instantiate the memory interface in the RTL.
3.14.2 Setting I/O Assignments in Arria 10 EMIF IP
The .qip file contains the I/O standard and I/O termination assignments required by
the memory interface pins for proper operation. The assignment values are based on
input that you provide during generation.
Unlike earlier device families, for Arria 10 EMIF IP you do not need to run a
<instance_name)_pin_assignments.tcl script to add the assignments into the
Quartus Prime Settings File (.qsf). The system reads and applies the assignments
from the .qip file during every compilation, regardless of how you name the memory
interface pins in the top-level design component. No new assignments are created in
the project's .qsf file during compilation.
Note that I/O assignments in the .qsf file must specify the names of your top-level
pins as target (-to), and you must not include the -entity or -library options.
Consult the generated .qip file for the set of I/O assignments that are provided with
the IP.
Changing I/O Assignments
You should not make changes to the generated .qip file, because any changes are
overwritten and lost when you regenerate the IP. If you want to override an
assignment made in the .qip file, add the desired assignment to the project's .qsf
file. Assignments in the .qsf file always take precedence over assignments in
the .qip file.
External Memory Interface Handbook Volume 3: Reference Material
185
3 Functional Description—Intel Arria 10 EMIF IP
3.15 Debugging Arria 10 EMIF IP
You can debug hardware failures by connecting to the EMIF Debug Toolkit or by
exporting an Avalon-MM slave port, from which you can access information gathered
during calibration. You can also connect to this port to mask ranks and to request
recalibration.
You can access the exported Avalon-MM port in two ways:
•
Via the External Memory Interface Debug Toolkit
•
Via On-Chip Debug (core logic on the FPGA)
3.15.1 External Memory Interface Debug Toolkit
The External Memory Interface Debug Toolkit provides access to data collected by the
Nios II sequencer during memory calibration, and allows you to perform certain tasks.
The External Memory Interface Debug Toolkit provides access to data including the
following:
•
General interface information, such as protocol and interface width
•
Calibration results per group, including pass/fail status, failure stage, and delay
settings
You can also perform the following tasks:
•
Mask ranks from calibration (you might do this to skip specific ranks)
•
Request recalibration of the interface
3.15.2 On-Chip Debug for Arria 10
The On-Chip Debug feature allows user logic to access the same debug capabilities as
the External Memory Interface Toolkit. You can use On-Chip Debug to monitor the
calibration results of an external memory interface, without a connected computer.
To use On-Chip Debug, you need a C header file which is provided as part of the
external memory interface IP. The C header file defines data structures that contain
calibration data, and definitions of the commands that can be sent to the memory
interface.
The On-Chip Debug feature accesses the data structures through the Avalon-MM port
that is exposed by the EMIF IP when you turn on debugging features.
3.15.3 Configuring Your EMIF IP for Use with the Debug Toolkit
The Arria 10 EMIF Debug Interface IP core contains the access point through which
the EMIF Debug Toolkit reads calibration data collected by the Nios II sequencer.
Connecting an EMIF IP Core to an Arria 10 EMIF Debug Interface
For the EMIF Debug Toolkit to access the calibration data for an Arria 10 EMIF IP core,
you must connect one of the EMIF cores in each I/O column to an Arria 10 EMIF
Debug Interface IP core. Subsequent EMIF IP cores in the same column must connect
in a daisy chain to the first.
External Memory Interface Handbook Volume 3: Reference Material
186
3 Functional Description—Intel Arria 10 EMIF IP
There are two ways that you can add the Arria 10 EMIF Debug Interface IP core to
your design:
•
When you generate your EMIF IP core, on the Diagnostics tab, select Add EMIF
Debug Interface for the EMIF Debug Toolkit/On-Chip Debug Port; you do
not have to separately instantiate an Arria 10 EMIF Debug Interface core. This
method does not export an Avalon-MM slave port. You can use this method if you
require only EMIF Debug Toolkit access to this I/O column; that is, if you do not
require On-Chip Debug Port access, or PHYLite reconfiguration access.
•
When you generate your EMIF IP core, on the Diagnostics tab, select Export for
the EMIF Debug Toolkit/On-Chip Debug Port. Then, separately instantiate an
Arria 10 EMIF Debug Interface core and connect its to_ioaux interface to the
cal_debug interface on the EMIF IP core. This method is appropriate if you want
to also have On-Chip Debug Port access to this I/O column, or PHYLite
reconfiguration access.
For each of the above methods, you must assign a unique interface ID for each
external memory interface in the I/O column, to identify that interface in the Debug
Toolkit. You can assign an interface ID using the dropdown list that appears when you
enable the Debug Toolkit/On-Chip Debug Port option.
Daisy-Chaining Additional EMIF IP Cores for Debugging
After you have connected an Arria 10 EMIF Debug Interface to one of the EMIF IP
cores in an I/O column, you must then connect subsequent EMIF IP cores in that
column in a daisy-chain manner. If you don't require debug capabilities for a particular
EMIF IP core, you do not have to connect that core to the daisy chain.
To create a daisy chain of EMIF IP cores, follow these steps:
1. On the first EMIF IP core, select Enable Daisy-Chaining for EMIF Debug
Toolkit/On-Chip Debug Port to create an Avalon-MM interface called
cal_debug_out.
2. On the second EMIF IP core, select Export as the EMIF Debug Toolkit/On-Chip
Debug Port mode, to export an Avalon-MM interface called cal_debug.
3.
Connect the cal_debug_out interface of the first EMIF core to the cal_debug
interface of the second EMIF core.
4.
To connect more EMIF cores to the daisy chain, select the Enable DaisyChaining for EMIF Debug Toolkit/On-Chip Debug Port option on the second
core, connect it to the next core using the Export option as described above.
Repeat the process for subsequent EMIF cores.
If you place any PHYLite cores with dynamic reconfiguration enabled into the same I/O
column as an EMIF IP core, you should instantiate and connect the PHYLite cores in a
similar way. See the Altera PHYLite for Memory Megafunction User Guide for more
information.
3.15.4 Arria 10 EMIF Debugging Examples
This topic provides examples of debugging a single external memory interface, and of
adding additional EMIF instances to an I/O column.
External Memory Interface Handbook Volume 3: Reference Material
187
3 Functional Description—Intel Arria 10 EMIF IP
Debugging a Single External Memory Interface
1.
Under EMIF Debug Toolkit/On-Chip Debug Port, select Add EMIF Debug
Interface.
(If you want to use the On-Chip Debug Port instead of the EMIF Debug Toolkit,
select Export instead.)
Figure 104. EMIF With Debug Interface Added (No Additional Ports)
emif_0
global_reset_reset_sink
pll_ref_clk_clock_sink
oct_conduit_end
mem_conduit_end
status_conduit_end
ctrl_amm_avalon_slave_0
2.
emif_usr_clk_clock_source
emif_usr_reset_reset_source
emif
If you want to connect additional EMIF or PHYLite components in this I/O column,
select Enable Daisy Chaining for EMIF Debug Toolkit/On-Chip Debug Port.
Figure 105. EMIF With cal_debug Avalon Master Exported
emif_0
global_reset_reset_sink
pll_ref_clk_clock_sink
oct_conduit_end
mem_conduit_end
status_conduit_end
ctrl_amm_avalon_slave_0
emif_usr_clk_clock_source
emif_usr_reset_reset_source
cal_debug_out_reset_reset_source
cal_debug_out_clk_clock_source
cal_debug_out_avalon_master
emif
cal_debug Avalon
Master Exported
Adding Additional EMIF Instances to an I/O Column
1.
Under EMIF Debug Toolkit/On-Chip Debug Port, select Export.
Figure 106. EMIF With cal_debug Avalon Slave Exported
emif_1
cal_debug Avalon
Slave Exported
global_reset_reset_sink
pll_ref_clk_clock_sink
oct_conduit_end
mem_conduit_end
status_conduit_end
cal_debug_reset_reset_sink
cal_debug_clk_clock_sink
ctrl_amm_avalon_slave_0
cal_debug_avalon_slave
emif_usr_clk_clock_source
emif_usr_reset_reset_source
emif
2. Specify a unique interface ID for this EMIF instance.
3. If you want to connect additional EMIF or PHYLite components in this I/O column,
select Enable Daisy Chaining for EMIF Debug Toolkit/On-Chip Debug Port.
External Memory Interface Handbook Volume 3: Reference Material
188
3 Functional Description—Intel Arria 10 EMIF IP
Figure 107. EMIF With Both cal_debug Master and Slave Exported
emif_1
cal_debug Avalon
Slave Exported
4.
global_reset_reset_sink
pll_ref_clk_clock_sink
oct_conduit_end
mem_conduit_end
status_conduit_end
cal_debug_reset_reset_sink
cal_debug_clk_clock_sink
ctrl_amm_avalon_slave_0
cal_debug_avalon_slave
emif_usr_clk_clock_source
emif_usr_reset_reset_source
cal_debug_out_reset_reset_source
cal_debug_out_clk_clock_source
cal_debug_out_avalon_master
cal_debug Avalon
Master Exported
emif
Connect the cal_debug Avalon Master, clock, and reset interfaces of the previous
component to the cal_debug Avalon Slave, clock, and reset interfaces of this
component.
Figure 108. EMIF Components Connected
emif_1
emif_0
global_reset_reset_sink
pll_ref_clk_clock_sink
oct_conduit_end
mem_conduit_end
status_conduit_end
ctrl_amm_avalon_slave_0
emif_usr_clk_clock_source
emif_usr_reset_reset_source
cal_debug_out_reset_reset_source
cal_debug_out_clk_clock_source
cal_debug_out_avalon_master
emif
global_reset_reset_sink
pll_ref_clk_clock_sink
oct_conduit_end
mem_conduit_end
status_conduit_end
cal_debug_reset_reset_sink
cal_debug_clk_clock_sink
ctrl_amm_avalon_slave_0
cal_debug_avalon_slave
emif_usr_clk_clock_source
emif_usr_reset_reset_source
cal_debug_out_reset_reset_source
cal_debug_out_clk_clock_source
cal_debug_out_avalon_master
emif
3.16 Arria 10 EMIF for Hard Processor Subsystem
The Arria 10 EMIF IP can enable the Arria 10 Hard Processor Subsystem (HPS) to
access external DRAM memory devices.
To enable connectivity between the Arria 10 HPS and the Arria 10 EMIF IP, you must
create and configure an instance of the Arria 10 External Memory Interface for HPS IP
core, and use Qsys to connect it to the Arria 10 Hard Processor Subsystem instance in
your system.
Supported Modes
The Arria 10 Hard Processor Subsystem is compatible with the following external
memory configurations:
Protocol
DDR3, DDR4
Maximum memory clock frequency
DDR3: 1.067 GHz
DDR4: 1.333 GHz
Configuration
Hard PHY with hard memory controller
Clock rate of PHY and hard memory controller
Half-rate
Data width (without ECC)
16-bit, 32-bit, 64-bit
2
Data width (with ECC)
24-bit, 40-bit, 72-bit
2
DQ width per group
x8
Maximum number of I/O lanes for address/command
3
continued...
External Memory Interface Handbook Volume 3: Reference Material
189
3 Functional Description—Intel Arria 10 EMIF IP
Memory format
Discrete, UDIMM, SODIMM, RDIMM
Ranks / CS# width
Up to 2
Notes to table:
1. Only Arria 10 devices with a special ordering code support 64-bit and 72-bit data widths; all other devices support only
to 32-bit data widths.
Note:
Arria 10 HPS EMIF IP does not currently support SmartVID.
3.16.1 Restrictions on I/O Bank Usage for Arria 10 EMIF IP with HPS
Only certain Arria 10 I/O banks can be used to implement Arria 10 EMIF IP with the
Arria 10 Hard Processor Subsystem. If both Arria 10 HPS EMIF IP and non-HPS Arria
10 EMIF IP are implemented, you must place the non-HPS EMIF IP in a different I/O
column than the HPS EMIF IP.
External Memory Interface Handbook Volume 3: Reference Material
190
3 Functional Description—Intel Arria 10 EMIF IP
The restrictions on I/O bank usage result from the Arria 10 HPS having hard-wired
connections to the EMIF circuits in the I/O banks closest to the HPS. For any given
EMIF configuration, the pin-out of the EMIF-to-HPS interface is fixed.
The following diagram illustrates the use of I/O banks and lanes for various EMIF-HPS
data widths:
I/O Bank 2L
(not used by HPS EMIF)
16 bit, with ECC
32 bit, no ECC
16 bit, no ECC
Lane 3
ECC
8
bits
ECC
8
bits
Lane 2
Addr
/ cmd
Addr
/ cmd
Addr
/ cmd
Addr
/ cmd
Addr
/ cmd
HPS
Addr
/ cmd
Lane 1
I/O Bank 2K
ECC
8
bits
32 bit, with ECC
64 bit, no ECC
64 bit, with ECC
Figure 109. I/O Banks and Lanes Usage
Lane 0
Lane 2
Data
32
bits
Data
16
bits
Data
64
bits
Data
64
bits
Data
16
bits
Lane 1
Lane 0
Lane 3
Lane 2
Lane 1
I/O Bank 2I
Data
32
bits
I/O Band 2J
Lane 3
Lane 0
External Memory Interface Handbook Volume 3: Reference Material
191
3 Functional Description—Intel Arria 10 EMIF IP
You should refer to the pinout file for your device and package for detailed
information. Banks and pins used for HPS access to a DDR interface are labeled
HPS_DDR in the HPS Function column of the pinout file.
By default, the Arria 10 External Memory Interface for HPS IP core together with the
Quartus Prime Fitter automatically implement the correct pin-out for HPS EMIF without
you having to implement additional constraints. If, for any reason, you must modify
the default pin-out, you must adhere to the following requirements, which are specific
to HPS EMIF:
1.
Within a single data lane (which implements a single x8 DQS group):
a.
DQ pins must use pins at indices 1, 2, 3, 6, 7, 8, 9, 10. You may swap the
locations between the DQ bits (that is, you may swap location of DQ[0] and
DQ[3]) so long as the resulting pin-out uses pins at these indices only.
b.
DM/DBI pin must use pin at index 11. There is no flexibility.
c.
DQS/DQS# must use pins at index 4 and 5. There is no flexibility.
2. Assignment of data lanes must be as illustrated in the above figure. You are
allowed to swap the locations of entire byte lanes (that is, you may swap locations
of byte 0 and byte 3) so long as the resulting pin-out uses only the lanes
permitted by your HPS EMIF configuration, as shown in the above figure.
3.
You must not change placement of the address and command pins from the
default.
4.
You may place the alert# pin at any available pin location in either a data lane or
an address and command lane.
To override the default generated pin assignments, comment out the relevant
HPS_LOCATION assignments in the .qip file, and add your own location assignments
(using set_location_assignment) in the .qsf file.
3.16.2 Using the EMIF Debug Toolkit with Arria 10 HPS Interfaces
The External Memory Interface Debug Toolkit is not directly compatible with Arria 10
HPS interfaces.
To debug your Arria 10 HPS interface using the EMIF Debug Toolkit, you should create
an identically parameterized, non-HPS version of your interface, and apply the EMIF
Debug Toolkit to that interface. When you finish debugging this non-HPS interface, you
can then apply any needed changes to your HPS interface, and continue your design
development.
3.17 Arria 10 EMIF Ping Pong PHY
Ping Pong PHY allows two memory interfaces to share the address and command bus
through time multiplexing. Compared to having two independent interfaces that
allocate address and command lanes separately, Ping Pong PHY achieves the same
throughput with fewer resources, by sharing the address and command lanes.
In Arria 10 EMIF, Ping Pong PHY supports both half-rate and quarter-rate interfaces for
DDR3, and quarter-rate for DDR4.
External Memory Interface Handbook Volume 3: Reference Material
192
3 Functional Description—Intel Arria 10 EMIF IP
3.17.1 Ping Pong PHY Feature Description
Conventionally, the address and command buses of a DDR3 or DDR4 half-rate or
quarter-rate interface use 2T time—meaning that commands are issued for two fullrate clock cycles, as illustrated below.
Figure 110. 2T Command Timing
CK
CSn
Addr, ba
Extra Setup Time
2T Command Issued
Active Period
With the Ping Pong PHY, address and command signals from two independent
controllers are multiplexed onto shared buses by delaying one of the controller
outputs by one full-rate clock cycle. The result is 1T timing, with a new command
being issued on each full-rate clock cycle. The following figure shows address and
command timing for the Ping Pong PHY.
External Memory Interface Handbook Volume 3: Reference Material
193
3 Functional Description—Intel Arria 10 EMIF IP
Figure 111. 1T Command Timing Use by Ping Pong PHY
CK
CSn[0]
CSn[1]
Addr, ba
Cmd
Dev1
Cmd
Dev0
3.17.2 Ping Pong PHY Architecture
In Arria 10 EMIF, the Ping Pong PHY feature can be enabled only with the hard
memory controller, where two hard memory controllers are instantiated—one for the
primary interface and one for the secondary interface.
The hard memory controller I/O bank of the primary interface is used for address and
command and is always adjacent and above the hard memory controller bank of the
secondary interface. All four lanes of the primary hard memory controller bank are
used for address and command. The I/O bank containing the secondary hard memory
controller must have at least one lane from the secondary interface.
The following example shows a 2x16 Ping Pong PHY bank-lane configuration. The
upper bank (I/O bank N) is the address and command bank, which serves both the
primary and secondary interfaces. The primary hard memory controller is linked to the
secondary interface by the Ping Pong bus. The lower bank (I/O bank N-1) is the
secondary interface bank, which carries the data buses for both primary and
secondary interfaces. In the 2x16 case a total of four I/O banks are required for data,
hence two banks in total are sufficient for the implementation.
The data for the primary interface is routed down to the top two lanes of the
secondary I/O bank, and the data for the secondary interface is routed to the bottom
two lanes of the secondary I/O bank.
External Memory Interface Handbook Volume 3: Reference Material
194
3 Functional Description—Intel Arria 10 EMIF IP
Figure 112. 2x16 Ping Pong PHY I/O Bank-Lane Configuration
I/O Tile N
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
Primary HMC
I/O Tile N - 1
Secondary HMC
Address/
Command
Primaary
Interface
Data Bus
Secondary
Interface
Data Bus
A 2x32 interface can be implemented using three tiles, so long as the tile containing
the secondary hard memory controller has at least one secondary data lane. The order
of the lanes does not matter.
External Memory Interface Handbook Volume 3: Reference Material
195
3 Functional Description—Intel Arria 10 EMIF IP
Figure 113. 2x32 Ping Pong PHY I/O Bank-Lane Configuration.
I/O Tile N + 1
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
Control Path
I/O Tile N
Primary HMC
I/O Tile N - 1
Secondary HMC
Primary
Interface
Data Bus
Address/
Command
Secondary
Interface
Data Bus
3.17.3 Ping Pong PHY Limitations
Ping Pong PHY supports up to two ranks per memory interface. In addition, the
maximum data width is x72, which is half the maximum width of x144 for a single
interface.
Ping Pong PHY uses all lanes of the address and command I/O bank as address and
command. For information on the pin allocations of the DDR3 and DDR4 address and
command I/O bank, refer to DDR3 Scheme 1 and DDR4 Scheme 3, in External
Memory Interface Pin Information for Arria 10 Devices, on www.altera.com.
An additional limitation is that I/O lanes may be left unused when you instantiate
multiple pairs of Ping Pong PHY interfaces. The following diagram shows two pairs of
x8 Pin Pong controllers (a total of 4 interfaces). Lanes highlighted in yellow are not
driven by any memory interfaces (unused lanes and pins can still serve as general
purpose I/Os). Even with some I/O lanes left unused, the Ping Pong PHY approach is
still beneficial in terms of resource usage, compared to independent interfaces.
Memory widths of 24 bits and 40 bits have a similar situation, while 16 bit, 32 bit, and
64 bit memory widths do not suffer this limitation.
External Memory Interface Handbook Volume 3: Reference Material
196
3 Functional Description—Intel Arria 10 EMIF IP
Figure 114. Two Pairs of x8 Pin-Pong PHY Controllers
I/O Tile N
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
DBCO
Data Buffer x12
DBC1
Data Buffer x12
DBC2
Data Buffer x12
DBC3
Data Buffer x12
Control Path
I/O Tile N - 1
Primary HMC
I/O Tile N - 2
Primary
Interface
Data Bus
Address/
Command
Secondary
Interface
Data Bus
Secondary HMC
I/O Tile N - 3
Primary HMC
I/O Tile N - 4
Primary
Interface
Data Bus
Address/
Command
Secondary
Interface
Data Bus
Secondary HMC
Related Links
External Memory Interface Pin Information for Arria 10 Devices
3.17.4 Ping Pong PHY Calibration
A Ping Pong PHY interface is calibrated as a regular interface of double width.
External Memory Interface Handbook Volume 3: Reference Material
197
3 Functional Description—Intel Arria 10 EMIF IP
Calibration of a Ping Pong PHY interface incorporates two sequencers, one on the
primary hard memory controller I/O bank, and one on the secondary hard memory
controller I/O bank. To ensure that the two sequencers issue instructions on the same
memory clock cycle, the Nios II processor configures the sequencer on the primary
hard memory controller to receive a token from the secondary interface, ignoring any
commands from the Avalon bus. Additional delays are programmed on the secondary
interface to allow for the passing of the token from the sequencer on the secondary
hard memory controller tile to the sequencer on the primary hard memory controller
tile. During calibration, the Nios II processor assumes that commands are always
issued from the sequencer on the primary hard memory controller I/O bank. After
calibration, the Nios II processor adjusts the delays for use with the primary and
secondary hard memory controllers.
3.17.5 Using the Ping Pong PHY
The following steps describe how to use the Ping Pong PHY for Arria 10 EMIF.
1. Configure a single memory interface according to your requirements.
2. Select Instantiate two controllers sharing a Ping Pong PHY on the General
tab in the parameter editor.
The Quartus Prime software replicates the interface, resulting in two memory
controllers and a shared PHY. The system configures the I/O bank-lane structure,
without further input from you.
3.17.6 Ping Pong PHY Simulation Example Design
The following figure illustrates a top-level block diagram of a generated Ping Pong PHY
simulation example design, using two I/O banks.
Functionally, the IP interfaces with user traffic separately, as it would with two
independent memory interfaces. You can also generate synthesizable example
designs, where the external memory interface IP interfaces with a traffic generator.
Figure 115. Ping Pong PHY Simulation Example Design
Simulation Example Design
EMIF
Tile N
Traffic
Generator 0
Sim
Checker
Primary
HMC
Tile N - 1
Lane 3
CS, ODT, CKE
Lane 2
CAS, RAS, WE, ADDR, BA, BG, ...
Lane 1
CS, ODT, CKE
Lane 0
Lane 3
DQ, DQS, DM
Lane 2
Traffic
Generator 1
Secondary
HMC
Memory
0
Lane 1
Lane 0
External Memory Interface Handbook Volume 3: Reference Material
198
DQ, DQS, DM
Memory
1
3 Functional Description—Intel Arria 10 EMIF IP
3.18 AFI 4.0 Specification
The Altera PHY interface (AFI) 4.0 defines communication between the controller and
physical layer (PHY) in the external memory interface IP.
The AFI is a single-data-rate interface, meaning that data is transferred on the rising
edge of each clock cycle. Most memory interfaces, however, operate at double-datarate, transferring data on both the rising and falling edges of the clock signal. If the
AFI interface is to directly control a double-data-rate signal, two single-data-rate bits
must be transmitted on each clock cycle; the PHY then sends out one bit on the rising
edge of the clock and one bit on the falling edge.
The AFI convention is to send the low part of the data first and the high part second,
as shown in the following figure.
Figure 116. Single Versus Double Data Rate Transfer
clock
Single-data-rate
Double-data-rate
A High , A Low
A Low
A High
B High , B Low
B Low
B High
3.18.1 Bus Width and AFI Ratio
In cases where the AFI clock frequency is one-half or one-quarter of the memory clock
frequency, the AFI data must be twice or four times as wide, respectively, as the
corresponding memory data. The ratio between AFI clock and memory clock
frequencies is referred to as the AFI ratio. (A half-rate AFI interface has an AFI ratio of
2, while a quarter-rate interface has an AFI ratio of 4.)
In general, the width of the AFI signal depends on the following three factors:
•
The size of the equivalent signal on the memory interface. For example, if
a[15:0] is a DDR3 address input and the AFI clock runs at the same speed as
the memory interface, the equivalent afi_addr bus will be 16-bits wide.
•
The data rate of the equivalent signal on the memory interface. For example, if
d[7:0] is a double-data-rate QDR II input data bus and the AFI clock runs at the
same speed as the memory interface, the equivalent afi_write_data bus will
be 16-bits wide.
•
The AFI ratio. For example, if cs_n is a single-bit DDR3 chip select input and the
AFI clock runs at half the speed of the memory interface, the equivalent
afi_cs_n bus will be 2-bits wide.
The following formula summarizes the three factors described above:
AFI_width = memory_width * signal_rate * AFI_RATE_RATIO
Note:
The above formula is a general rule, but not all signals obey it. For definite signal-size
information, refer to the specific table.
External Memory Interface Handbook Volume 3: Reference Material
199
3 Functional Description—Intel Arria 10 EMIF IP
3.18.2 AFI Parameters
The following tables list Altera PHY interface (AFI) parameters for AFI 4.0. Not all
parameters are used for all protocols.
The parameters described in the following tables affect the width of AFI signal buses.
Parameters prefixed by MEM_IF_ refer to the signal size at the interface between the
PHY and memory device.
Table 46.
Ratio Parameters
Parameter Name
Description
AFI_RATE_RATIO
The ratio between the AFI clock frequency and the memory clock frequency.
For full-rate interfaces this value is 1, for half-rate interfaces the value is 2,
and for quarter-rate interfaces the value is 4.
DATA_RATE_RATIO
The number of data bits transmitted per clock cycle. For single-date rate
protocols this value is 1, and for double-data rate protocols this value is 2.
ADDR_RATE_RATIO
The number of address bits transmitted per clock cycle. For single-date rate
address protocols this value is 1, and for double-data rate address protocols
this value is 2.
Table 47.
Memory Interface Parameters
Parameter Name
Description
MEM_IF_ADDR_WIDTH
The width of the address bus on the memory device(s). For LPDDR3, the
width of the CA bus, which encodes commands and addresses together.
MEM_IF_BANKGROUP_WIDTH
The width of the bank group bus on the interface to the memory device(s).
Applicable to DDR4 only.
MEM_IF_BANKADDR_WIDTH
The width of the bank address bus on the interface to the memory device(s).
Typically, the log 2 of the number of banks. Not applicable to DDR4.
MEM_IF_CS_WIDTH
The number of chip selects on the interface to the memory device(s).
MEM_IF_CKE_WIDTH
Number of CKE signals on the interface to the memory device(s). This usually
equals to MEM_IF_CS_WIDTH except for certain DIMM configurations.
MEM_IF_ODT_WIDTH
Number of ODT signals on the interface to the memory device(s). This usually
equals to MEM_IF_CS_WIDTH except for certain DIMM configurations.
MEM_IF_WRITE_DQS_WIDTH
The number of DQS (or write clock) signals on the write interface. For
example, the number of DQS groups.
MEM_IF_CLK_PAIR_COUNT
The number of CK/CK# pairs.
MEM_IF_DQ_WIDTH
The number of DQ signals on the interface to the memory device(s). For
single-ended interfaces such as QDR II, this value is the number of D or Q
signals.
MEM_IF_DM_WIDTH
The number of data mask pins on the interface to the memory device(s).
Table 48.
Derived AFI Parameters
Parameter Name
Derivation Equation
AFI_ADDR_WIDTH
MEM_IF_ADDR_WIDTH * AFI_RATE_RATIO * ADDR_RATE_RATIO
AFI_BANKGROUP_WIDTH
MEM_IF_BANKGROUP_WIDTH * AFI_RATE_RATIO * ADDR_RATE_RATIO.
Applicable to DDR4 only.
AFI_BANKADDR_WIDTH
MEM_IF_BANKADDR_WIDTH * AFI_RATE_RATIO * ADDR_RATE_RATIO
continued...
External Memory Interface Handbook Volume 3: Reference Material
200
3 Functional Description—Intel Arria 10 EMIF IP
Parameter Name
Derivation Equation
AFI_CONTROL_WIDTH
AFI_RATE_RATIO * ADDR_RATE_RATIO
AFI_CS_WIDTH
MEM_IF_CS_WIDTH * AFI_RATE_RATIO
AFI_CKE_WIDTH
MEM_IF_CKE_WIDTH * AFI_RATE_RATIO
AFI_ODT_WIDTH
MEM_IF_ODT_WIDTH * AFI_RATE_RATIO
AFI_DM_WIDTH
MEM_IF_DM_WIDTH * AFI_RATE_RATIO * DATA_RATE_RATIO
AFI_DQ_WIDTH
MEM_IF_DQ_WIDTH * AFI_RATE_RATIO * DATA_RATE_RATIO
AFI_WRITE_DQS_WIDTH
MEM_IF_WRITE_DQS_WIDTH * AFI_RATE_RATIO
AFI_LAT_WIDTH
6
AFI_RLAT_WIDTH
AFI_LAT_WIDTH
AFI_WLAT_WIDTH
AFI_LAT_WIDTH * MEM_IF_WRITE_DQS_WIDTH
AFI_CLK_PAIR_COUNT
MEM_IF_CLK_PAIR_COUNT
AFI_WRANK_WIDTH
Number of ranks * MEM_IF_WRITE_DQS_WIDTH *AFI_RATE_RATIO
AFI_RRANK_WIDTH
Number of ranks * MEM_IF_READ_DQS_WIDTH *AFI_RATE_RATIO
3.18.3 AFI Signals
The following tables list Altera PHY interface (AFI) signals grouped according to their
functions.
In each table, the Direction column denotes the direction of the signal relative to the
PHY. For example, a signal defined as an output passes out of the PHY to the
controller. The AFI specification does not include any bidirectional signals.
Not all signals are used for all protocols.
3.18.3.1 AFI Clock and Reset Signals
The AFI interface provides up to two clock signals and an asynchronous reset signal.
Table 49.
Clock and Reset Signals
Signal Name
Direction
Width
Description
afi_clk
Output
1
Clock with which all data exchanged on the AFI bus
is synchronized. In general, this clock is referred to
as full-rate, half-rate, or quarter-rate, depending on
the ratio between the frequency of this clock and
the frequency of the memory device clock.
afi_half_clk
Output
1
Clock signal that runs at half the speed of the
afi_clk. The controller uses this signal when the
half-rate bridge feature is in use. This signal is
optional.
afi_reset_n
Output
1
Asynchronous reset output signal. You must
synchronize this signal to the clock domain in which
you use it.
External Memory Interface Handbook Volume 3: Reference Material
201
3 Functional Description—Intel Arria 10 EMIF IP
3.18.3.2 AFI Address and Command Signals
The address and command signals for AFI 4.0 encode read/write/configuration
commands to send to the memory device. The address and command signals are
single-data rate signals.
Table 50.
Address and Command Signals
Signal Name
Direction
Width
Description
afi_addr
Input
AFI_ADDR_WIDTH
Address or CA bus (LPDDR3 only).
ADDR_RATE_RATIO is 2 for
LPDDR3 CA bus.
afi_bg
Input
AFI_BANKGROUP_WIDTH
Bank group (DDR4 only).
afi_ba
Input
AFI_BANKADDR_WIDTH
Bank address. (Not applicable for
LPDDR3.)
afi_cke
Input
AFI_CLK_EN_WIDTH
Clock enable.
afi_cs_n
Input
AFI_CS_WIDTH
Chip select signal. (The number of
chip selects may not match the
number of ranks; for example,
RDIMMs and LRDIMMs require a
minimum of 2 chip select signals
for both single-rank and dual-rank
configurations. Consult your
memory device data sheet for
information about chip select signal
width.) (Matches the number of
ranks for LPDDR3.)
afi_ras_n
Input
AFI_CONTROL_WIDTH
RAS# (for DDR2 and DDR3
memory devices.)
afi_we_n
Input
AFI_CONTROL_WIDTH
WE# (for DDR2, DDR3, and
RLDRAM II memory devices.)
afi_rw_n
Input
AFI_CONTROL_WIDTH * 2
RWA/B# (QDR-IV).
afi_cas_n
Input
AFI_CONTROL_WIDTH
CAS# (for DDR2 and DDR3
memory devices.)
afi_act_n
Input
AFI_CONTROL_WIDTH
ACT# (DDR4).
afi_ref_n
Input
AFI_CONTROL_WIDTH
REF# (for RLDRAM II memory
devices.)
afi_rst_n
Input
AFI_CONTROL_WIDTH
RESET# (for DDR3 and DDR4
memory devices.)
afi_odt
Input
AFI_CLK_EN_WIDTH
On-die termination signal for DDR2,
DDR3, and LPDDR3 memory
devices. (Do not confuse this
memory device signal with the
FPGA’s internal on-chip termination
signal.)
afi_par
Input
AFI_CS_WIDTH
Address and command parity input.
(DDR4)
Address parity input. (QDR-IV)
afi_ainv
Input
AFI_CONTROL_WIDTH
Address inversion. (QDR-IV)
continued...
External Memory Interface Handbook Volume 3: Reference Material
202
3 Functional Description—Intel Arria 10 EMIF IP
Signal Name
Direction
Width
Description
afi_mem_clk_disable
Input
AFI_CLK_PAIR_COUNT
When this signal is asserted,
mem_clk and mem_clk_n are
disabled. This signal is used in lowpower mode.
afi_wps_n
Output
AFI_CS_WIDTH
WPS (for QDR II/II+ memory
devices.)
afi_rps_n
Output
AFI_CS_WIDTH
RPS (for QDR II/II+ memory
devices.)
3.18.3.3 AFI Write Data Signals
Write Data Signals for AFI 4.0 control the data, data mask, and strobe signals passed
to the memory device during write operations.
Table 51.
Write Data Signals
Signal Name
Direction
Width
Description
afi_dqs_burst
Input
AFI_RATE_RATIO
Controls the enable on the strobe
(DQS) pins for DDR2, DDR3,
LPDDR2, and LPDDR3 memory
devices. When this signal is
asserted, mem_dqs and mem_dqsn
are driven.
This signal must be asserted before
afi_wdata_valid to implement the
write preamble, and must be driven
for the correct duration to generate
a correctly timed mem_dqs signal.
afi_wdata_valid
Input
AFI_RATE_RATIO
Write data valid signal. This signal
controls the output enable on the
data and data mask pins.
afi_wdata
Input
AFI_DQ_WIDTH
Write data signal to send to the
memory device at double-data
rate. This signal controls the PHY’s
mem_dq output.
afi_dm
Input
AFI_DM_WIDTH
Data mask. This signal controls the
PHY’s mem_dm signal for DDR2,
DDR3, LPDDR2, LPDDR3, and
RLDRAM II memory devices.)
Also directly controls the PHY's
mem_dbi signal for DDR4.
The mem_dm and mem_dbi
features share the same port on
the memory device.
afi_bws_n
Input
AFI_DM_WIDTH
Data mask. This signal controls the
PHY’s mem_bws_n signal for
QDR II/II+ memory devices.
afi_dinv
Input
AFI_WRITE_DQS_WIDTH * 2
Data inversion. It directly controls
the PHY's mem_dinva/b signal for
QDR-IV devices.
3.18.3.4 AFI Read Data Signals
Read Data Signals for AFI 4.0 control the data sent from the memory device during
read operations.
External Memory Interface Handbook Volume 3: Reference Material
203
3 Functional Description—Intel Arria 10 EMIF IP
Table 52.
Read Data Signals
Signal Name
Direction
Width
Description
afi_rdata_en_full
Input
AFI_RATE_RATIO
Read data enable full. Indicates that the
memory controller is currently performing
a read operation. This signal is held high
for the entire read burst.If this signal is
aligned to even clock cycles, it is possible
to use 1-bit even in half-rate mode (i.e.,
AFI_RATE=2).
afi_rdata
Output
AFI_DQ_WIDTH
Read data from the memory device. This
data is considered valid only when
afi_rdata_valid is asserted by the PHY.
afi_rdata_valid
Output
AFI_RATE_RATIO
Read data valid. When asserted, this
signal indicates that the afi_rdata bus is
valid.If this signal is aligned to even clock
cycles, it is possible to use 1-bit even in
half-rate mode (i.e., AFI_RATE=2).
3.18.3.5 AFI Calibration Status Signals
The PHY instantiates a sequencer which calibrates the memory interface with the
memory device and some internal components such as read FIFOs and valid FIFOs.
The sequencer reports the results of the calibration process to the controller through
the Calibration Status Signals in the AFI interface.
Table 53.
Calibration Status Signals
Signal Name
Direction
Width
Description
afi_cal_success
Output
1
Asserted to indicate that calibration has
completed successfully.
afi_cal_fail
Output
1
Asserted to indicate that calibration has
failed.
afi_cal_req
Input
1
Effectively a synchronous reset for the
sequencer. When this signal is asserted,
the sequencer returns to the reset state;
when this signal is released, a new
calibration sequence begins.
afi_wlat
Output
AFI_WLAT_WIDTH
The required write latency in afi_clk
cycles, between address/command and
write data being issued at the PHY/
controller interface. The afi_wlat value
can be different for different groups; each
group’s write latency can range from 0 to
63. If write latency is the same for all
groups, only the lowest 6 bits are
required.
afi_rlat
Output
AFI_RLAT_WIDTH
The required read latency in afi_clk cycles
between address/command and read
data being returned to the PHY/controller
interface. Values can range from 0 to 63.
(1)
Note to Table:
1. The afi_rlat signal is not supported for PHY-only designs. Instead, you can sample the afi_rdata_valid signal to
determine when valid read data is available.
External Memory Interface Handbook Volume 3: Reference Material
204
3 Functional Description—Intel Arria 10 EMIF IP
3.18.3.6 AFI Tracking Management Signals
When tracking management is enabled, the sequencer can take control over the AFI
4.0 interface at given intervals, and issue commands to the memory device to track
the internal DQS Enable signal alignment to the DQS signal returning from the
memory device. The tracking management portion of the AFI 4.0 interface provides a
means for the sequencer and the controller to exchange handshake signals.
Table 54.
Tracking Management Signals
Signal Name
Direction
Width
Description
afi_ctl_refresh_done
Input
4
Handshaking signal from controller to
tracking manager, indicating that a
refresh has occurred and waiting for a
response.
afi_seq_busy
Output
4
Handshaking signal from sequencer to
controller, indicating when DQS tracking
is in progress.
afi_ctl_long_idle
Input
4
Handshaking signal from controller to
tracking manager, indicating that it has
exited low power state without a periodic
refresh, and waiting for response.
3.18.3.7 AFI Shadow Register Management Signals
Shadow registers are a feature that enables high-speed multi-rank support. Shadow
registers allow the sequencer to calibrate each rank separately, and save the
calibrated settings—such as deskew delay-chain configurations—of each rank in its
own set of shadow registers.
During a rank-to-rank switch, the correct set of calibrated settings is restored just in
time to optimize the data valid window. The PHY relies on additional AFI signals to
control which set of shadow registers to activate.
Table 55.
Shadow Register Management Signals
Signal Name
Direction
Width
Description
afi_wrank
Input
AFI_WRANK_WIDTH
Signal from controller
specifying which rank the
write data is going to. The
signal timing is identical to
that of afi_dqs_burst. That
is, afi_wrank must be
asserted at the same time
and must last the same
duration as the
afi_dqs_burst signal.
afi_rrank
Output
AFI_RRANK_WIDTH
Signal from controller
specifying which rank is
being read. The signal must
be asserted at the same
time as the afi_rdata_en
signal when issuing a read
command, but unlike
afi_rdata_en, afi_rrank is
stateful. That is, once
asserted, the signal value
must remain unchanged
until the controller issues a
new read command to a
different rank.
External Memory Interface Handbook Volume 3: Reference Material
205
3 Functional Description—Intel Arria 10 EMIF IP
Both the afi_wrank and afi_rrank signals encode the rank being accessed using the
one-hot scheme (e.g. in a quad-rank interface, 0001, 0010, 0100, 1000 refer to the
1st, 2nd, 3rd, 4th rank respectively). The ordering within the bus is the same as other
AFI signals. Specifically the bus is ordered by time slots, for example:
Half-rate afi_w/rrank = {T1, T0}
Quarter-rate afi_w/rrank = {T3, T2, T1, T0}
Where Tx is a number of rank-bit words that one-hot encodes the rank being accessed
at the yth full-rate cycle.
Additional Requirements for Arria 10 Shadow Register Support
To ensure that the hardware has enough time to switch from one shadow register to
another, the controller must satisfy the following minimum rank-to-rank-switch delays
(tRTRS):
•
Two read commands going to different ranks must be separated by a minimum of
3 full-rate cycles (in addition to the burst length delay needed to avoid collision of
data bursts).
•
Two write commands going to different rank must be separated by a minimum of
4 full-rate cycles (in addition to the burst length delay needed to avoid collision of
data bursts).
The Arria 10 device family supports a maximum of 4 sets of shadow registers, each
for an independent set of timings. More than 4 ranks are supported if those ranks
have four or fewer sets of independent timing. For example, the rank multiplication
mode of an LRDIMM allows more than one physical rank to share a set of timing data
as a single logical rank. Therefore Arria 10 devices can support up to 4 logical ranks,
though that means more than 4 physical ranks.
3.18.4 AFI 4.0 Timing Diagrams
3.18.4.1 AFI Address and Command Timing Diagrams
Depending on the ratio between the memory clock and the PHY clock, different
numbers of bits must be provided per PHY clock on the AFI interface. The following
figures illustrate the AFI address/command waveforms in full, half and quarter rate
respectively.
The waveforms show how the AFI command phase corresponds to the memory
command output. AFI command 0 corresponds to the first memory command slot, AFI
command 1 corresponds to the second memory command slot, and so on.
External Memory Interface Handbook Volume 3: Reference Material
206
3 Functional Description—Intel Arria 10 EMIF IP
Figure 117. AFI Address and Command Full-Rate
Memory Interface
mem_clk
mem_cs_n
mem_cke
mem_ras_n
mem_cas_n
mem_we_n
AFI Interface
afi_clk
afi_cs_n
afi_cke
afi_ras_n
afi_cas_n
afi_we_n
External Memory Interface Handbook Volume 3: Reference Material
207
3 Functional Description—Intel Arria 10 EMIF IP
Figure 118. AFI Address and Command Half-Rate
Memory Interface
mem_clk
mem_cs_n
mem_cke
mem_ras_n
mem_cas_n
mem_we_n
AFI Interface
afi_clk
afi_cs_n[1]
afi_cs_n[0]
1
0
0
1
afi_cke[1]
afi_cke[0]
1
1
1
1
afi_ras_n[1]
afi_ras_n[0]
1
0
1
1
afi_cas_n[1]
afi_cas_n[0]
1
1
0
1
afi_we_n[1]
afi_we_n[0]
1
1
0
1
External Memory Interface Handbook Volume 3: Reference Material
208
3 Functional Description—Intel Arria 10 EMIF IP
Figure 119. AFI Address and Command Quarter-Rate
Memory Interface
mem_clk
mem_cs_n
mem_cke
mem_ras_n
mem_cas_n
mem_we_n
AFI Interface
afi_clk
afi_cs_n[3]
0
1
afi_cs_n[2]
afi_cs_n[1]
afi_cs_n[0]
1
0
0
1
1
0
afi_cke[3]
1
1
afi_cke[2]
afi_cke[1]
afi_cke[0]
1
1
1
1
1
1
1
1
afi_ras_n[3]
afi_ras_n[2]
afi_ras_n[1]
1
1
0
1
afi_ras_n[0]
1
0
afi_cas_n[3]
afi_cas_n[2]
afi_cas_n[1]
afi_cas_n[0]
0
1
1
0
1
1
1
1
afi_we_n[3]
afi_we_n[2]
0
1
1
0
afi_we_n[1]
afi_we_n[0]
1
1
1
1
3.18.4.2 AFI Write Sequence Timing Diagrams
The following timing diagrams illustrate the relationships between the write command
and corresponding write data and write enable signals, in full, half, and quarter rate.
For half rate and quarter rate, when the write command is sent on the first memory
clock in a PHY clock (for example, afi_cs_n[0] = 0), that access is called aligned
access; otherwise it is called unaligned access. You may use either aligned or
unaligned access, or you may use both, but you must ensure that the distance
External Memory Interface Handbook Volume 3: Reference Material
209
3 Functional Description—Intel Arria 10 EMIF IP
between the write command and the corresponding write data are constant on the
AFI interface. For example, if a command is sent on the second memory clock in a PHY
clock, the write data must also start at the second memory clock in a PHY clock.
Write sequences with wlat=0
Figure 120. AFI Write Data Full-Rate, wlat=0
afi_clk
afi_command
WR
WR
WR
afi_wdata_valid
afi_wdata
A
B
C
D
E
F
afi_dm
M
N
O
P
Q
R
The following diagrams illustrate both aligned and unaligned access. The first three
write commands are aligned accesses where they were issued on LSB of
afi_command. The fourth write command is unaligned access where it was issued on
a different command slot. AFI signals must be shifted accordingly, based on the
command slot.
Figure 121. AFI Write Data Half-Rate, wlat=0
afi_clk
afi_command[1]
NOP
NOP
NOP
NOP
afi_command[0]
WR
WR
WR
WR
afi_wdata_valid[1]
1
1
1
1
0
afi_wdata_valid[0]
1
1
1
0
1
afi_wdata[1]
B
D
F
G
afi_wdata[0]
A
C
E
afi_dm[1]
N
P
R
afi_dm[0]
M
O
Q
External Memory Interface Handbook Volume 3: Reference Material
210
H
S
T
3 Functional Description—Intel Arria 10 EMIF IP
Figure 122. AFI Write Data Quarter-Rate, wlat=0
afi_clk
afi_command[3]
NOP
NOP
NOP
WR
afi_command[2]
NOP
NOP
NOP
NOP
afi_command[1]
NOP
NOP
NOP
NOP
afi_command[0]
WR
WR
WR
NOP
afi_wdata_valid[3]
1
1
1
1
0
afi_wdata_valid[2]
1
1
1
0
1
afi_wdata_valid[1]
1
1
1
0
1
afi_wdata_valid[0]
1
1
1
0
1
afi_wdata[3]
D
H
L
A
afi_wdata[2]
C
G
K
D
afi_wdata[1]
B
F
J
C
afi_wdata[0]
A
E
I
B
afi_dm[3]
P
T
X
afi_dm[2]
O
S
W
P
afi_dm[1]
N
R
V
O
afi_dm[0]
M
Q
U
N
M
Write sequences with wlat=non-zero
The afi_wlat is a signal from the PHY. The controller must delay afi_dqs_burst,
afi_wdata_valid, afi_wdata and afi_dm signals by a number of PHY clock cycles
equal to afi_wlat, which is a static value determined by calibration before the PHY
asserts cal_success to the controller. The following figures illustrate the cases when
wlat=1. Note that wlat is in the number of PHY clocks and therefore wlat=1 equals 1,
2, and 4 memory clocks delay, respectively, on full, half and quarter rate.
External Memory Interface Handbook Volume 3: Reference Material
211
3 Functional Description—Intel Arria 10 EMIF IP
Figure 123. AFI Write Data Full-Rate, wlat=1
afi_clk
afi_command
WR
WR
WR
afi_wdata_valid
afi_wdata
A
B
C
D
E
F
afi_dm
M
N
O
P
Q
R
Figure 124. AFI Write Data Half-Rate, wlat=1
afi_clk
afi_command[1]
NOP
NOP
NOP
NOP
afi_command[0]
WR
WR
WR
WR
afi_wdata_valid[1]
1
1
1
1
0
afi_wdata_valid[0]
1
1
1
0
1
afi_wdata[1]
B
D
F
G
afi_wdata[0]
A
C
E
afi_dm[1]
N
P
R
afi_dm[0]
M
O
Q
External Memory Interface Handbook Volume 3: Reference Material
212
H
S
T
3 Functional Description—Intel Arria 10 EMIF IP
Figure 125. AFI Write Data Quarter-Rate, wlat=1
afi_clk
afi_command[3]
NOP
NOP
NOP
WR
afi_command[2]
NOP
NOP
NOP
NOP
afi_command[1]
NOP
NOP
NOP
NOP
afi_command[0]
WR
WR
WR
NOP
afi_wdata_valid[3]
1
1
1
1
0
afi_wdata_valid[2]
1
1
1
0
1
afi_wdata_valid[1]
1
1
1
0
1
afi_wdata_valid[0]
1
1
1
0
1
afi_wdata[3]
D
H
L
A
afi_wdata[2]
C
G
K
D
afi_wdata[1]
B
F
J
C
afi_wdata[0]
A
E
I
B
afi_dm[3]
P
T
X
afi_dm[2]
O
S
W
P
afi_dm[1]
N
R
V
O
afi_dm[0]
M
Q
U
N
M
DQS burst
The afi_dqs_burst signal must be asserted one or two complete memory clock
cycles earlier to generate DQS preamble. DQS preamble is equal to one-half and onequarter AFI clock cycles in half and quarter rate, respectively.
A DQS preamble of two is required in DDR4, when the write preamble is set to two
clock cycles.
The following diagrams illustrate how afi_dqs_burst must be asserted in full, half, and
quarter-rate configurations.
External Memory Interface Handbook Volume 3: Reference Material
213
3 Functional Description—Intel Arria 10 EMIF IP
Figure 126. AFI DQS Burst Full-Rate, wlat=1
afi_clk
afi_command
WR
WR
WR
afi_dqs_burst
afi_wdata_valid
afi_wdata
A
B
C
D
E
F
afi_dm
M
N
O
P
Q
R
Figure 127. AFI DQS Burst Half-Rate, wlat=1
afi_clk
afi_command[1]
NOP
NOP
NOP
NOP
afi_command[0]
WR
WR
WR
WR
afi_dqs_burst[1]
1
1
1
1
1
1
0
afi_dqs_burst[0]
0
1
0
1
1
1
1
afi_wdata_valid[1]
1
1
1
1
0
afi_wdata_valid[0]
1
1
1
0
1
afi_wdata[1]
B
D
F
G
afi_wdata[0]
A
C
E
afi_dm[1]
N
P
R
afi_dm[0]
M
O
Q
External Memory Interface Handbook Volume 3: Reference Material
214
H
S
T
3 Functional Description—Intel Arria 10 EMIF IP
Figure 128. AFI DQS Burst Quarter-Rate, wlat=1
afi_clk
afi_command[3]
NOP
NOP
NOP
WR
afi_command[2]
NOP
NOP
NOP
NOP
afi_command[1]
NOP
NOP
NOP
NOP
afi_command[0]
WR
WR
WR
NOP
afi_dqs_burst[3]
1
1
1
1
1
1
0
afi_dqs_burst[2]
0
1
0
1
1
1
1
afi_dqs_burst[1]
0
1
0
1
1
0
1
afi_dqs_burst[0]
0
1
0
1
1
0
1
afi_wdata_valid[3]
1
1
1
1
0
afi_wdata_valid[2]
1
1
1
0
1
afi_wdata_valid[1]
1
1
1
0
1
afi_wdata_valid[0]
1
1
1
0
1
afi_wdata[3]
D
H
L
A
afi_wdata[2]
C
G
K
D
afi_wdata[1]
B
F
J
C
afi_wdata[0]
A
E
I
B
afi_dm[3]
P
T
X
afi_dm[2]
O
S
W
P
afi_dm[1]
N
R
V
O
afi_dm[0]
M
Q
U
N
M
Write data sequence with DBI (DDR4 and QDRIV only)
The DDR4 write DBI feature is supported in the PHY, and when it is enabled, the PHY
sends and receives the DBI signal without any controller involvement. The sequence is
identical to non-DBI scenarios on the AFI interface.
External Memory Interface Handbook Volume 3: Reference Material
215
3 Functional Description—Intel Arria 10 EMIF IP
Write data sequence with CRC (DDR4 only)
When the CRC feature of the PHY is enabled and used, the controller ensures at least
one memory clock cycle between write commands, during which the PHY inserts the
CRC data. Sending back to back write command would cause functional failure. The
following figures show the legal sequences in CRC mode.
Entries marked as 0 and RESERVE must be observed by the controller; no information
is allowed on those entries.
Figure 129. AFI Write Data with CRC Half-Rate, wlat=2
afi_clk
afi_command[1]
NOP
NOP
NOP
afi_command[0]
WR
WR
WR
afi_dqs_burst[1]
1
1
1
afi_dqs_burst[0]
0
1
1
afi_wdata_valid[1]
1
1
afi_wdata_valid[0]
1
1
afi_wdata[1]
B
D
afi_wdata[0]
A
C
afi_dm[1]
N
P
afi_dm[0]
M
O
External Memory Interface Handbook Volume 3: Reference Material
216
0
0
Reserve
Reserve
1
1
1
1
1
0
0
1
1
0
1
1
1
1
1
1
0
1
1
0
1
1
F
H
I
H
Reserve
E
G
Reserve
J
L
R
T
U
W
Reserve
Q
S
Reserve
V
X
3 Functional Description—Intel Arria 10 EMIF IP
Figure 130. AFI Write Data with CRC Quarter-Rate, wlat=2
afi_clk
afi_command[1]
NOP
NOP
NOP
afi_command[0]
WR
WR
WR
afi_dqs_burst[3]
1
1
1
1
1
1
1
0
afi_dqs_burst[2]
0
1
0
1
1
1
0
1
afi_dqs_burst[1]
0
1
0
1
1
0
1
1
afi_dqs_burst[0]
0
1
0
1
0
1
1
1
0
afi_wdata_valid[3]
1
1
1
1
1
0
afi_wdata_valid[2]
1
1
1
1
0
1
afi_wdata_valid[1]
1
1
1
0
1
1
afi_wdata_valid[0]
1
1
0
1
1
1
afi_wdata[3]
D
D
G
J
M
Reserve
afi_wdata[2]
C
C
F
I
Reserve
P
afi_wdata[1]
B
B
E
Reserve
L
O
afi_wdata[0]
A
A
Reserve
H
K
N
afi_dm[3]
D
D
G
J
M
Reserve
afi_dm[2]
C
C
F
I
Reserve
P
afi_dm[1]
B
B
E
Reserve
L
O
afi_dm[0]
A
A
Reserve
H
K
N
0
Reserve
Reserve
3.18.4.3 AFI Read Sequence Timing Diagrams
The following waveforms illustrate the AFI write data waveform in full, half, and
quarter-rate, respectively.
The afi_rdata_en_full signal must be asserted for the entire read burst
operation. The afi_rdata_en signal need only be asserted for the intended read
data.
Aligned and unaligned access for read commands is similar to write commands;
however, the afi_rdata_en_full signal must be sent on the same memory clock in
a PHY clock as the read command. That is, if a read command is sent on the second
memory clock in a PHY clock, afi_rdata_en_full must also be asserted, starting
from the second memory clock in a PHY clock.
External Memory Interface Handbook Volume 3: Reference Material
217
3 Functional Description—Intel Arria 10 EMIF IP
Figure 131. AFI Read Data Full-Rate
afi_clk
afi_command
RD
RD
RD
afi_rdata_en_full
afi_rdata
A
B
C
D
E
F
afi_rdata_valid
The following figure illustrates that the second and third reads require only the first
and second half of data, respectively. The first three read commands are aligned
accesses where they are issued on the LSB of afi_command. The fourth read
command is unaligned access, where it is issued on a different command slot. AFI
signals must be shifted accordingly, based on command slot.
Figure 132. AFI Read Data Half-Rate
afi_clk
afi_command[1]
NOP
NOP
NOP
RD
afi_command[0]
RD
RD
RD
NOP
afi_rdata_en_full[1]
1
1
1
1
0
afi_rdata_en_full[0]
1
1
1
0
1
afi_rdata[1]
B
D
F
afi_rdata[0]
A
C
E
afi_rdata_valid[1]
1
1
1
afi_rdata_valid[0]
1
1
1
G
H
1
1
In the following figure, the first three read commands are aligned accesses where
they are issued on the LSB of afi_command. The fourth read command is unaligned
access, where it is issued on a different command slot. AFI signals must be shifted
accordingly, based on command slot.
External Memory Interface Handbook Volume 3: Reference Material
218
3 Functional Description—Intel Arria 10 EMIF IP
Figure 133. AFI Read Data Quarter-Rate
afi_clk
afi_command[3]
NOP
NOP
NOP
RD
afi_command[2]
NOP
NOP
NOP
NOP
afi_command[1]
NOP
NOP
NOP
NOP
afi_command[0]
RD
RD
RD
NOP
afi_rdata_en_full[3]
1
1
1
1
0
afi_rdata_en_full[2]
1
1
1
0
1
afi_rdata_en_full[1]
1
1
1
0
1
afi_rdata_en_full[0]
1
1
1
0
1
afi_rdata[3]
D
H
L
afi_rdata[2]
C
G
K
P
afi_rdata[1]
B
F
J
O
afi_rdata[0]
A
E
I
N
afi_rdata_valid[3]
1
1
1
1
0
afi_rdata_valid[2]
1
1
1
0
1
afi_rdata_valid[1]
1
1
1
0
1
afi_rdata_valid[0]
1
1
1
0
1
M
3.18.4.4 AFI Calibration Status Timing Diagram
The controller interacts with the PHY during calibration at power-up and at
recalibration.
At power-up, the PHY holds afi_cal_success and afi_cal_fail 0 until calibration
is done, when it asserts afi_cal_success, indicating to controller that the PHY is
ready to use and afi_wlat and afi_rlat signals have valid values.
At recalibration, the controller asserts afi_cal_req, which triggers the same
sequence as at power-up, and forces recalibration of the PHY.
External Memory Interface Handbook Volume 3: Reference Material
219
3 Functional Description—Intel Arria 10 EMIF IP
Figure 134. Calibration
PHY Status Calibrating
Controller Working
Re-Calibrating
Controller Working
AFI Interface
afi_cal_success
afi_cal_fail
afi_cal_req
afi_wlat
9
9
afi_rlat
9
9
3.19 Resource Utilization
The following tables provide resource utilization information for external memory
interfaces on Arria 10 devices.
3.19.1 QDR-IV Resource Utilization in Arria 10 Devices
The following table shows typical resource usage of QDR-IV interfaces with soft
controller for Arria 10 devices.
Table 56.
QDR-IV Resource Utilization in Arria 10 Devices
Memory Width
(Bits)
Combinational
ALUTs
Dedicated Logic
Registers
Block Memory
Bits
M20Ks
Soft Controller
18
2123
4592
18432
8
1
36
2127
6023
36864
16
1
72
2114
8826
73728
32
1
3.20 Arria 10 EMIF Latency
The following latency data applies to all memory protocols supported by the Arria 10
EMIF IP.
Table 57.
Rate
Latency in Full-Rate Memory Clock Cycles
1
Controller
Address &
Command
PHY
Address &
Command
Memory
Read
Latency 2
PHY Read
Data
Return
Controller
Read Data
Return
Round Trip
Round Trip
Without
Memory
—
Half:Write
12
2
3-23
—
—
—
Half:Read
8
2
3-23
6
8
27-47
24
continued...
External Memory Interface Handbook Volume 3: Reference Material
220
3 Functional Description—Intel Arria 10 EMIF IP
Rate
1
Controller
Address &
Command
PHY
Address &
Command
Memory
Read
Latency 2
PHY Read
Data
Return
Controller
Read Data
Return
Round Trip
Round Trip
Without
Memory
Quarter:Writ
e
14
2
3-23
—
—
—
—
Quarter:Rea
d
10
2
3-23
6
14
35-55
32
Half:Write
(ECC)
14
2
3-23
—
—
—
—
Half:Read
(ECC)
12
2
3-23
6
8
31-51
28
Quarter:Writ
e (ECC)
14
2
3-23
—
—
—
—
Quarter:Rea
d (ECC)
12
2
3-23
6
14
37-57
34
1. User interface rate; the controller always operates in half rate.
2. Minimum and maximum read latency range for DDR3, DDR4, and LPDDR3.
3.21 Arria 10 EMIF Calibration Times
The time needed for calibration varies, depending on many factors including the
interface width, the number of ranks, frequency, board layout, and difficulty of
calibration.
The following table lists approximate typical calibration times for various protocols and
configurations.
Table 58.
Arria 10 EMIF IP Approximate Calibration Times
Protocol
DDR3, x64 UDIMM, DQS x8, DM on
DDR4, x64 UDIMM, DQS x8, DBI on
RLDRAM 3, x36
QDR II, x36, BWS on
Rank and Frequency
Typical Calibration Time
1 rank, 933 MHz
102 ms
1 rank, 800 MHz
106 ms
2 rank, 933 MHz
198 ms
2 rank, 800 MHz
206 ms
1 rank, 1067 MHz
314 ms
1 rank, 800 MHz
353 ms
2 rank 1067 MHz
625 ms
2 rank 800 MHz
727 ms
1200 MHz
2808 ms
1067 MHz
2825 ms
1200 MHz, with DM
2818 ms
1067 MHz, with DM
2833 ms
333 MHz
616 ms
continued...
External Memory Interface Handbook Volume 3: Reference Material
221
3 Functional Description—Intel Arria 10 EMIF IP
Protocol
QDR-IV, x36, BWS on
Rank and Frequency
Typical Calibration Time
633 MHz
833 ms
1067 MHz
1563 ms
1067 MHz, with DBI
1556 ms
3.22 Integrating a Custom Controller with the Hard PHY
If you want to use your own custom memory controller, you must integrate the
controller with the hard PHY to achieve a complete memory solution.
Observe the following general guidelines:
•
When you configure your external memory interface IP, ensure that you select
Configuration ➤ Hard PHY Only on the General tab in the parameter editor.
•
Consult the AFI 4.0 Specification, for detailed information on the AFI interface to
the PHY.
3.23 Memory Mapped Register (MMR) Tables
The address buses to read and write from the MMR registers are 10 bits wide, while
the read and write data buses are configured to be 32 bits. The Bits Register Link
column in the table below provides the mapping on the width of the data read within
the 32-bit bus. The reads and writes are always performed using the 32-bit-wide bus.
Register Summary
Address 32-bit Bus
Bits Register link
sbcfg8
7
16
sbcfg9
8
16
reserve2
9
16
ctrlcfg0
10
32
ctrlcfg1
11
32
ctrlcfg2
12
32
ctrlcfg3
13
32
ctrlcfg4
14
32
ctrlcfg5
15
16
ctrlcfg6
16
16
ctrlcfg7
17
16
ctrlcfg8
18
8
ctrlcfg9
19
8
dramtiming0
20
24
dramodt0
21
32
dramodt1
22
24
Register
continued...
External Memory Interface Handbook Volume 3: Reference Material
222
3 Functional Description—Intel Arria 10 EMIF IP
Register
Address 32-bit Bus
Bits Register link
sbcfg0
23
32
sbcfg1
24
32
sbcfg2
25
8
sbcfg3
26
24
sbcfg4
27
24
sbcfg5
28
8
sbcfg6
29
32
sbcfg7
30
8
caltiming0
31
32
caltiming1
32
32
caltiming2
33
32
caltiming3
34
32
caltiming4
35
32
caltiming5
36
24
caltiming6
37
32
caltiming7
38
32
caltiming8
39
32
caltiming9
40
8
caltiming10
41
8
dramaddrw
42
24
sideband0
43
8
sideband1
44
8
sideband2
45
8
sideband3
46
8
sideband4
47
8
sideband5
48
8
sideband6
49
8
sideband7
50
8
sideband8
51
8
sideband9
52
8
sideband10
53
8
sideband11
54
8
sideband12
55
8
sideband13
56
32
sideband14
57
16
sideband15
58
8
continued...
External Memory Interface Handbook Volume 3: Reference Material
223
3 Functional Description—Intel Arria 10 EMIF IP
Register
Address 32-bit Bus
Bits Register link
dramsts
59
8
dbgdone
60
8
dbgsignals
61
32
dbgreset
62
8
dbgmatch
63
32
counter0mask
64
32
counter1mask
65
32
counter0match
66
32
counter1match
67
32
niosreserve0
68
16
niosreserve1
69
16
niosreserve2
70
16
ecc1
128
10
ecc2
129
22
ecc3
130
9
ecc4
144
16
Note:
Addresses are in decimal format.
3.23.1 ctrlcfg0: Controller Configuration
address=10(32 bit)
Bit High
Bit Low
cfg_mem_type
3
0
Selects memory type. Program this
field with one of the following binary
values, "0000" for DDR3 SDRAM,
"0001" for DDR4 SDRAM, "0010" for
LPDDR3 SDRAM and "0011" for
RLDRAM3.
Read/Write
cfg_dimm_type
6
4
Selects dimm type. Program this field
with one of the following binary
values: 3' b000: for
DIMM_TYPE_COMPONENT, 3' b001:
for DIMM_TYPE_UDIMM, 3' b010:
for DIMM_TYPE_RDIMM, 3' b011:
for DIMM_TYPE_LRDIMM, 3' b100:
for DIMM_TYPE_SODIMM, and 3'
b101: for DIMM_TYPE_3DS.
Read/Write
cfg_ac_pos
8
7
Specify C/A (command/address) pin
position.
Read/Write
13
9
Configures burst length for control
path. Legal values are valid for JEDEC
allowed DRAM values for the DRAM
selected in cfg_type. For DDR3, DDR4
Read/Write
Field
cfg_ctrl_burst_length
Description
Access
continued...
External Memory Interface Handbook Volume 3: Reference Material
224
3 Functional Description—Intel Arria 10 EMIF IP
Field
Bit High
Bit Low
Description
Access
and LPDDR3, this should be
programmed with 8 (binary "01000"),
for RLDRAM III it can be programmed
with 2 or 4 or 8.
cfg_dbc0_burst_lengt
h
18
14
Configures burst length for DBC0.
Legal values are valid for JEDEC
allowed DRAM values for the DRAM
selected in cfg_type. For DDR3, DDR4
and LPDDR3, this should be
programmed with 8 (binary "01000"),
for RLDRAM III it can be programmed
with 2 or 4 or 8.
Read/Write
cfg_dbc1_burst_lengt
h
23
19
Configures burst length for DBC1.
Legal values are valid for JEDEC
allowed DRAM values for the DRAM
selected in cfg_type. For DDR3, DDR4
and LPDDR3, this should be
programmed with 8 (binary "01000"),
for RLDRAM III it can be programmed
with 2 or 4 or 8.
Read/Write
cfg_dbc2_burst_lengt
h
28
24
Configures burst length for DBC2.
Legal values are valid for JEDEC
allowed DRAM values for the DRAM
selected in cfg_type. For DDR3, DDR4
and LPDDR3, this should be
programmed with 8 (binary "01000"),
for RLDRAM III it can be programmed
with 2 or 4 or 8.
Read/Write
3.23.2 ctrlcfg1: Controller Configuration
address=11(32 bit)
Bit High
Bit Low
Description
Access
cfg_dbc3_burst_le
ngth
4
0
Configures burst length for DBC3.
Legal values are valid for JEDEC
allowed DRAM values for the DRAM
selected in cfg_type. For DDR3, DDR4
and LPDDR3, this should be
programmed with 8 (binary "01000"),
for RLDRAM III it can be programmed
with 2 or 4 or 8.
Read/Write
cfg_addr_order
6
5
Selects the order for address
interleaving. Programming this field
with different values gives different
mappings between the AXI or AvalonMM address and the SDRAM address.
Program this field with the following
binary values to select the ordering.
"00" - chip, row, bank(BG, BA),
column; "01" - chip, bank(BG, BA),
row, column; "10"-row, chip,
bank(BG, BA), column.
Read/Write
cfg_ctrl_enable_ec
c
7
7
Enable the generation and checking of
ECC.
Read/Write
cfg_dbc0_enable_e
cc
8
8
Enable the generation and checking of
ECC.
Read/Write
Field
continued...
External Memory Interface Handbook Volume 3: Reference Material
225
3 Functional Description—Intel Arria 10 EMIF IP
Field
Bit High
Bit Low
Description
Access
cfg_dbc1_enable_e
cc
9
9
Enable the generation and checking of
ECC.
Read/Write
cfg_dbc2_enable_e
cc
10
10
Enable the generation and checking of
ECC.
Read/Write
cfg_dbc3_enable_e
cc
11
11
Enable the generation and checking of
ECC.
Read/Write
cfg_reorder_data
12
12
This bit controls whether the
controller can re-order operations to
optimize SDRAM bandwidth. It should
generally be set to a one.
Read/Write
cfg_ctrl_reorder_rd
ata
13
13
This bit controls whether the
controller need to re-order the read
return data.
Read/Write
cfg_dbc0_reorder_
rdata
14
14
This bit controls whether the
controller need to re-order the read
return data.
Read/Write
cfg_dbc1_reorder_
rdata
15
15
This bit controls whether the
controller need to re-order the read
return data.
Read/Write
cfg_dbc2_reorder_
rdata
16
16
This bit controls whether the
controller need to re-order the read
return data.
Read/Write
cfg_dbc3_reorder_
rdata
17
17
This bit controls whether the
controller need to re-order the read
return data.
Read/Write
cfg_reorder_read
18
18
This bit controls whether the
controller can re-order read command
to 1.
Read/Write
cfg_starve_limit
24
19
Specifies the number of DRAM burst
transactions an individual transaction
will allow to reorder ahead of it before
its priority is raised in the memory
controller.
Read/Write
cfg_dqstrk_en
25
25
Enables DQS tracking in the PHY.
Read/Write
cfg_ctrl_enable_d
m
26
26
Set to a one to enable DRAM
operation if DM pins are connected.
Read/Write
cfg_dbc0_enable_d
m
27
27
Set to a one to enable DRAM
operation if DM pins are connected.
Read/Write
cfg_dbc1_enable_d
m
28
28
Set to a one to enable DRAM
operation if DM pins are connected.
Read/Write
cfg_dbc2_enable_d
m
29
29
Set to a one to enable DRAM
operation if DM pins are connected.
Read/Write
cfg_dbc3_enable_d
m
30
30
Set to a one to enable DRAM
operation if DM pins are connected.
Read/Write
3.23.3 ctrlcfg2: Controller Configuration
External Memory Interface Handbook Volume 3: Reference Material
226
3 Functional Description—Intel Arria 10 EMIF IP
address=12(32 bit)
Field
Bit High
Bit Low
Description
Access
cfg_ctrl_output_regd
0
0
Set to one to register the HMC
command output. Set to 0 to disable
it.
Read/Write
cfg_dbc0_output_reg
d
1
1
Set to one to register the HMC
command output. Set to 0 to disable
it.
Read/Write
cfg_dbc1_output_reg
d
2
2
Set to one to register the HMC
command output. Set to 0 to disable
it.
Read/Write
cfg_dbc2_output_reg
d
3
3
Set to one to register the HMC
command output. Set to 0 to disable
it.
Read/Write
cfg_dbc3_output_reg
d
4
4
Set to one to register the HMC
command output. Set to 0 to disable
it.
Read/Write
cfg_ctrl2dbc_switch0
6
5
Select of the MUX ctrl2dbc_switch0. 2
Read/Write
cfg_ctrl2dbc_switch1
8
7
Select of the MUX ctrl2dbc_switch1. 2
Read/Write
cfg_dbc0_ctrl_sel
9
9
DBC0 - control path select. 1
Read/Write
cfg_dbc1_ctrl_sel
10
10
DBC1 - control path select. 1
Read/Write
cfg_dbc2_ctrl_sel
11
11
DBC2 - control path select. 1
Read/Write
cfg_dbc3_ctrl_sel
12
12
DBC3 - control path select. 1
Read/Write
cfg_dbc2ctrl_sel
14
13
Specifies which DBC is driven by the
local control path. 2
Read/Write
cfg_dbc0_pipe_lat
17
15
Specifies in number of controller clock
cycles the latency of pipelining the
signals from control path to DBC0
Read/Write
cfg_dbc1_pipe_lat
20
18
Specifies in number of controller clock
cycles the latency of pipelining the
signals from control path to DBC1
Read/Write
cfg_dbc2_pipe_lat
23
21
Specifies in number of controller clock
cycles the latency of pipelining the
signals from control path to DBC2
Read/Write
cfg_dbc3_pipe_lat
26
24
Specifies in number of controller clock
cycles the latency of pipelining the
signals from control path to DBC3
Read/Write
Description
Access
3.23.4 ctrlcfg3: Controller Configuration
address=13(32 bit)
Bit High
Bit Low
cfg_ctrl_cmd_rate
2
0
3
Read/Write
cfg_dbc0_cmd_rate
5
3
3
Read/Write
cfg_dbc1_cmd_rate
8
6
3
Read/Write
cfg_dbc2_cmd_rate
11
9
3
Read/Write
Field
continued...
External Memory Interface Handbook Volume 3: Reference Material
227
3 Functional Description—Intel Arria 10 EMIF IP
Field
Bit High
Bit Low
cfg_dbc3_cmd_rate
14
12
3
Read/Write
cfg_ctrl_in_protocol
15
15
1
Read/Write
cfg_dbc0_in_protocol
16
16
1
Read/Write
cfg_dbc1_in_protocol
17
17
1
Read/Write
cfg_dbc2_in_protocol
18
18
1
Read/Write
cfg_dbc3_in_protocol
19
19
1
Read/Write
cfg_ctrl_dualport_en
20
20
Enable the second command port for
RLDRAM3 only (BL=2 or 4)
Read/Write
cfg_dbc0_dualport_e
n
21
21
Enable the second data port for
RLDRAM3 only (BL=2 or 4)
Read/Write
cfg_dbc1_dualport_e
n
22
22
Enable the second data port for
RLDRAM3 only (BL=2 or 4)
Read/Write
cfg_dbc2_dualport_e
n
23
23
Enable the second data port for
RLDRAM3 only (BL=2 or 4)
Read/Write
cfg_dbc3_dualport_e
n
24
24
Enable the second data port for
RLDRAM3 only (BL=2 or 4)
Read/Write
cfg_arbiter_type
25
25
Indicates controller arbiter operating
mode. Set this to: - 1
Read/Write
cfg_open_page_en
26
26
Set to 1 to enable the open page
policy when command reordering is
disabled (cfg_cmd_reorder = 0). This
bit does not matter when
cfg_cmd_reorder is 1.
Read/Write
cfg_rld3_multibank_
mode
30
28
Multibank setting, specific for
RLDRAM3. Set this to: - 3
Read/Write
Note:
Description
Access
DDR4 gear down mode is not supported.
3.23.5 ctrlcfg4: Controller Configuration
address=14(32 bit)
Bit High
Bit Low
cfg_tile_id
4
0
Tile ID.
Read/Write
cfg_pingpong_mode
6
5
Ping Pong mode: 2
Read/Write
cfg_ctrl_slot_rotate_
en
9
7
Cmd slot rotate enable: bit[0]
controls write, 1
Read/Write
cfg_dbc0_slot_rotate
_en
12
10
DBC0 slot rotate enable: bit[0]
controls write, 1
Read/Write
cfg_dbc1_slot_rotate
_en
15
13
DBC1 slot rotate enable: bit[0]
controls write, 1
Read/Write
cfg_dbc2_slot_rotate
_en
18
16
DBC2 slot rotate enable: bit[0]
controls write, 1
Read/Write
cfg_dbc3_slot_rotate
_en
21
19
DBC3 slot rotate enable: bit[0]
controls write, 1
Read/Write
Field
Description
Access
continued...
External Memory Interface Handbook Volume 3: Reference Material
228
3 Functional Description—Intel Arria 10 EMIF IP
Field
Bit High
Bit Low
Description
Access
cfg_ctrl_slot_offset
23
22
Enables afi information to be offset by
numbers of FR cycles. Affected afi
signal is afi_rdata_en,
afi_rdata_en_full, afi_wdata_valid,
afi_dqs_burst, afi_mrnk_write and
afi_mrnk_read. Set this to: - 2
Read/Write
cfg_dbc0_slot_offset
25
24
Enables afi information to be offset by
numbers of FR cycles. Affected afi
signal is afi_rdata_en,
afi_rdata_en_full, afi_wdata_valid,
afi_dqs_burst, afi_mrnk_write and
afi_mrnk_read. Set this to: - 2
Read/Write
cfg_dbc1_slot_offset
27
26
Enables afi information to be offset by
numbers of FR cycles. Affected afi
signal is afi_rdata_en,
afi_rdata_en_full, afi_wdata_valid,
afi_dqs_burst, afi_mrnk_write and
afi_mrnk_read. Set this to: - 2
Read/Write
cfg_dbc2_slot_offset
29
28
Enables afi information to be offset by
numbers of FR cycles. Affected afi
signal is afi_rdata_en,
afi_rdata_en_full, afi_wdata_valid,
afi_dqs_burst, afi_mrnk_write and
afi_mrnk_read. Set this to: - 2
Read/Write
cfg_dbc3_slot_offset
31
30
Enables afi information to be offset by
numbers of FR cycles. Affected afi
signal is afi_rdata_en,
afi_rdata_en_full, afi_wdata_valid,
afi_dqs_burst, afi_mrnk_write and
afi_mrnk_read. Set this to: - 2
Read/Write
Description
Access
3.23.6 ctrlcfg5: Controller Configuration
address=15(32 bit)
Bit High
Bit Low
cfg_col_cmd_slot
3
0
Specify the col cmd slot. One hot
encoding.
Read/Write
cfg_row_cmd_slot
7
4
Specify the row cmd slot. One hot
encoding.
Read/Write
cfg_ctrl_rc_en
8
8
Set to 1 to enable the rate
conversion. It converts QR input from
core to HR inside HMC.
Read/Write
cfg_dbc0_rc_en
9
9
Set to 1 to enable the rate
conversion. It converts QR input from
core to HR inside HMC.
Read/Write
cfg_dbc1_rc_en
10
10
Set to 1 to enable the rate
conversion. It converts QR input from
core to HR inside HMC.
Read/Write
cfg_dbc2_rc_en
11
11
Set to 1 to enable the rate
conversion. It converts QR input from
core to HR inside HMC.
Read/Write
cfg_dbc3_rc_en
12
12
Set to 1 to enable the rate
conversion. It converts QR input from
core to HR inside HMC.
Read/Write
Field
External Memory Interface Handbook Volume 3: Reference Material
229
3 Functional Description—Intel Arria 10 EMIF IP
3.23.7 ctrlcfg6: Controller Configuration
address=16(32 bit)
Field
cfg_cs_chip
Bit High
Bit Low
Description
Access
15
0
Chip select mapping scheme. Mapping
seperated into 4 sections: [CS3][CS2]
[CS1][CS0] Each section consists of 4
bits to indicate which CS_n signal
should be active when command goes
to current CS. Eg: if we set to 16.
Read/Write
Description
Access
3.23.8 ctrlcfg7: Controller Configuration
address=17(32 bit)
Field
Bit High
Bit Low
cfg_clkgating_en
0
0
Set to 1 to enable the clock gating.
The clock is shut off for the whole
HMC.
Read/Write
cfg_rb_reserved_entr
y
7
1
Specify how many enties are reserved
in read buffer before almost full is
asserted.
Read/Write
cfg_wb_reserved_ent
ry
14
8
Specify how many enties are reserved
in write buffer before almost full is
asserted.
Read/Write
Description
Access
3.23.9 ctrlcfg8: Controller Configuration
address=18(32 bit)
Bit High
Bit Low
cfg_3ds_en
0
0
Setting to 1 to enable #DS support
for DDR4.
Read/Write
cfg_ck_inv
1
1
Use to program CK polarity. 1
Read/Write
cfg_addr_mplx_en
2
2
Setting to 1 enables RLD3 address
mulplex mode.
Read/Write
Field
3.23.10 ctrlcfg9: Controller Configuration
address=19(32 bit)
Field
cfg_dfx_bypass_en
Bit High
Bit Low
0
0
External Memory Interface Handbook Volume 3: Reference Material
230
Description
Used for dft and timing
characterization only. 1
Access
Read/Write
3 Functional Description—Intel Arria 10 EMIF IP
3.23.11 dramtiming0: Timing Parameters
address=20(32 bit)
Field
Bit High
Bit Low
6
0
Memory read latency.
Read/Write
cfg_power_saving_ex
it_cycles
12
7
The minimum number of cycles to
stay in a low power state. This applies
to both power down and self-refresh
and should be set to the greater of
tPD and tCKESR.
Read/Write
cfg_mem_clk_disable
_entry_cycles
18
13
Set to a the number of clocks after
the execution of an self-refresh to
stop the clock. This register is
generally set based on PHY design
latency and should generally not be
changed.
Read/Write
cfg_tcl
Description
Access
3.23.12 dramodt0: On-Die Termination Parameters
address=21(32 bit)
Bit High
Bit Low
Description
Access
cfg_write_odt_chip
Field
15
0
ODT scheme setting for write
command. Setting seperated into 4
sections: [CS3][CS2][CS1][CS0] Each
section consists of 4 bits to indicate
which chip should ODT be asserted
when write occurs on current CS. Eg:
if we set to 16.
Read/Write
cfg_read_odt_chip
31
16
ODT scheme setting for read
command. Setting seperated into 4
sections: [CS3][CS2][CS1][CS0] Each
section consists of 4 bits to indicate
which chip should ODT be asserted
when write occurs on current CS. Eg:
if we set to 16.
Read/Write
3.23.13 dramodt1: On-Die Termination Parameters
External Memory Interface Handbook Volume 3: Reference Material
231
3 Functional Description—Intel Arria 10 EMIF IP
address=22(32 bit)
Field
Bit High
Bit Low
Description
Access
cfg_wr_odt_on
5
0
Indicates number of memory clock
cycle gap between write command
and ODT signal rising edge.
Read/Write
cfg_rd_odt_on
11
6
Indicates number of memory clock
cycle gap between read command
and ODT signal rising edge.
Read/Write
cfg_wr_odt_period
17
12
Indicates number of memory clock
cycle write ODT signal should stay
asserted after rising edge.
Read/Write
cfg_rd_odt_period
23
18
Indicates number of memory clock
cycle read ODT signal should stay
asserted after rising edge.
Read/Write
3.23.14 sbcfg0: Sideband Configuration
address=23(32 bit)
Field
Bit High
Bit Low
Description
Access
cfg_rld3_refresh_seq
0
15
0
Banks to Refresh for RLD3 in
sequence 0. Must not be more than 4
banks.
Read/Write
cfg_rld3_refresh_seq
1
31
16
Banks to Refresh for RLD3 in
sequence 1. Must not be more than 4
banks.
Read/Write
3.23.15 sbcfg1: Sideband Configuration
address=24(32 bit)
Field
Bit High
Bit Low
Description
Access
cfg_rld3_refresh_seq
2
15
0
Banks to Refresh for RLD3 in
sequence 2. Must not be more than 4
banks.
Read/Write
cfg_rld3_refresh_seq
3
31
16
Banks to Refresh for RLD3 in
sequence 3. Must not be more than 4
banks.
Read/Write
3.23.16 sbcfg2: Sideband Configuration
External Memory Interface Handbook Volume 3: Reference Material
232
3 Functional Description—Intel Arria 10 EMIF IP
address=25(32 bit)
Field
Bit High
Bit Low
Description
Access
cfg_srf_zqcal_disable
0
0
Set to 1 to disable ZQ Calibration
after self refresh.
Read/Write
cfg_mps_zqcal_disab
le
1
1
Set to 1 to disable ZQ Calibration
after Maximum Power Saving exit.
Read/Write
cfg_mps_dqstrk_disa
ble
2
2
Set to 1 to disable DQS Tracking after
Maximum Power Saving exit.
Read/Write
cfg_sb_cg_disable
3
3
Set to 1 to disable mem_ck gating
during self refresh and deep power
down. Clock gating is not supported
when the Ping Pong PHY feature is
enabled. Do not enable clock gating
for Ping Pong PHY interfaces.
Read/Write
cfg_user_rfsh_en
4
4
Setting to 1 to enable user refresh.
Read/Write
cfg_srf_autoexit_en
5
5
Setting to 1 to enable controller to
exit Self Refresh when new command
is detected.
Read/Write
cfg_srf_entry_exit_bl
ock
7
6
Blocking arbiter from issuing cmds for
the 4 cases, 2
Read/Write
Description
Access
3.23.17 sbcfg3: Sideband Configuration
address=26(32 bit)
Field
cfg_sb_ddr4_mr3
Bit High
Bit Low
19
0
This register stores the DDR4 MR3
Content.
Read/Write
3.23.18 sbcfg4: Sideband Configuration
address=27(32 bit)
Field
cfg_sb_ddr4_mr4
Bit High
Bit Low
19
0
Description
This register stores the DDR4 MR4
Content.
Access
Read/Write
3.23.19 sbcfg5: Sideband Configuration
External Memory Interface Handbook Volume 3: Reference Material
233
3 Functional Description—Intel Arria 10 EMIF IP
address=28(32 bit)
Field
Bit High
Bit Low
Description
Access
cfg_short_dqstrk_ctrl
_en
0
0
Set to 1 to enable controller
controlled DQS short tracking, Set to
0 to enable sequencer controlled DQS
short tracking.
Read/Write
cfg_period_dqstrk_ct
rl_en
1
1
Set to 1 to enable controller to issue
periodic DQS tracking.
Read/Write
3.23.20 sbcfg6: Sideband Configuration
address=29(32 bit)
Field
Bit High
Bit Low
cfg_period_dqstrk_in
terval
15
0
cfg_t_param_dqstrk_
to_valid_last
23
cfg_t_param_dqstrk_
to_valid
31
Description
Access
Inverval between two controller
controlled periodic DQS tracking.
Read/Write
16
DQS Tracking Rd to Valid timing for
the last Rank.
Read/Write
24
DQS Tracking Rd to Valid timing for
Ranks other than the Last.
Read/Write
3.23.21 sbcfg7: Sideband Configuration
address=30(32 bit)
Field
Bit High
Bit Low
Description
Access
cfg_rfsh_warn_thres
hold
6
o
Threshold to warn a refresh is needed
within the number of controller clock
cycles specified by the threshold.
Read/Write
Description
Access
3.23.22 sbcfg8: Sideband Configuration
address=7(32 bit)
Field
cfg_reserve
cfg_ddr4_mps_addr
mirror
Bit High
Bit Low
15
1
General purpose reserved register.
Read/Write
0
0
When asserted, indicates DDR4
Address Mirroring is enabled for MPS.
Read/Write
Description
Access
3.23.23 sbcfg9: Sideband Configuration
address=8(32 bit)
Field
cfg_sb_ddr4_mr5
Bit High
Bit Low
15
0
External Memory Interface Handbook Volume 3: Reference Material
234
DDR4 Mode Register 5.
Read/Write
3 Functional Description—Intel Arria 10 EMIF IP
3.23.24 caltiming0: Command/Address/Latency Parameters
address=31(32 bit)
Field
Bit High
Bit Low
Description
Access
cfg_t_param_act_to_
rdwr
5
0
Activate to Read/write command
timing.
Read/Write
cfg_t_param_act_to_
pch
11
6
Active to precharge.
Read/Write
cfg_t_param_act_to_
act
17
12
Active to activate timing on same
bank.
Read/Write
cfg_t_param_act_to_
act_diff_bank
23
18
Active to activate timing on different
banks, for DDR4 same bank group.
Read/Write
cfg_t_param_act_to_
act_diff_bg
29
24
Active to activate timing on different
bank groups, DDR4 only.
Read/Write
3.23.25 caltiming1: Command/Address/Latency Parameters
address=32(32 bit)
Field
Bit High
Bit Low
Description
Access
cfg_t_param_rd_to_r
d
5
0
Read to read command timing on
same bank.
Read/Write
cfg_t_param_rd_to_r
d_diff_chip
11
6
Read to read command timing on
different chips.
Read/Write
cfg_t_param_rd_to_r
d_diff_bg
17
12
Read to read command timing on
different chips.
Read/Write
cfg_t_param_rd_to_
wr
23
18
Write to read command timing on
same bank.
Read/Write
cfg_t_param_rd_to_
wr_diff_chip
29
24
Read to write command timing on
different chips
Read/Write
3.23.26 caltiming2: Command/Address/Latency Parameters
address=33(32 bit)
Field
Bit High
Bit Low
Description
Access
cfg_t_param_rd_to_
wr_diff_bg
5
0
Read to write command timing on
different bank groups.
Read/Write
cfg_t_param_rd_to_
pch
11
6
Read to precharge command timing.
Read/Write
continued...
External Memory Interface Handbook Volume 3: Reference Material
235
3 Functional Description—Intel Arria 10 EMIF IP
Field
Bit High
Bit Low
Description
Access
cfg_t_param_rd_ap_
to_valid
17
12
Read command with autoprecharge to
data valid timing.
Read/Write
cfg_t_param_wr_to_
wr
23
18
Write to write command timing on
same bank.
Read/Write
cfg_t_param_wr_to_
wr_diff_chip
29
24
Write to write command timing on
different chips.
Read/Write
3.23.27 caltiming3: Command/Address/Latency Parameters
address=34(32 bit)
Field
Bit High
Bit Low
cfg_t_param_wr_to_
wr_diff_bg
5
0
Write to write command timing on
different bank groups.
Description
Read/Write
Access
cfg_t_param_wr_to_
rd
11
6
Write to read command timing.
Read/Write
cfg_t_param_wr_to_
rd_diff_chip
17
12
Write to read command timing on
different chips.
Read/Write
cfg_t_param_wr_to_
rd_diff_bg
23
18
Write to read command timing on
different bank groups.
Read/Write
cfg_t_param_wr_to_
pch
29
24
Write to precharge command timing.
Read/Write
3.23.28 caltiming4: Command/Address/Latency Parameters
address=35(32 bit)
Field
Bit High
Bit Low
cfg_t_param_wr_ap_
to_valid
5
0
Write with autoprecharge to valid
command timing.
Description
Read/Write
cfg_t_param_pch_to
_valid
11
6
Precharge to valid command timing.
Read/Write
cfg_t_param_pch_all
_to_valid
17
12
Precharge all to banks being ready for
bank activation command.
Read/Write
cfg_t_param_arf_to_
valid
25
18
Auto Refresh to valid DRAM command
window.
Read/Write
cfg_t_param_pdn_to
_valid
31
26
Power down to valid bank command
window.
Read/Write
3.23.29 caltiming5: Command/Address/Latency Parameters
External Memory Interface Handbook Volume 3: Reference Material
236
Access
3 Functional Description—Intel Arria 10 EMIF IP
address=36(32 bit)
Field
Bit High
Bit Low
cfg_t_param_srf_to_
valid
9
0
cfg_t_param_srf_to_
zq_cal
19
10
Description
Access
Self-refresh to valid bank command
window.
Read/Write
Self refresh to ZQ calibration window.
Read/Write
3.23.30 caltiming6: Command/Address/Latency Parameters
address=37(32 bit)
Field
Bit High
Bit Low
cfg_t_param_arf_per
iod
12
0
cfg_t_param_pdn_pe
riod
28
13
Description
Access
Auto-refresh period.
Read/Write
Clock power down recovery period.
Read/Write
3.23.31 caltiming7: Command/Address/Latency Parameters
address=38(32 bit)
Field
Bit High
Bit Low
Description
Access
cfg_t_param_zqcl_to
_valid
8
0
Long ZQ calibration to valid.
Read/Write
cfg_t_param_zqcs_to
_valid
15
9
Short ZQ calibration to valid.
Read/Write
cfg_t_param_mrs_to
_valid
19
16
Mode Register Setting to valid.
Read/Write
cfg_t_param_mps_to
_valid
29
20
Timing parameter for Maximum Power
Saving to any valid command. tXMP
Read/Write
3.23.32 caltiming8: Command/Address/Latency Parameters
address=39(32 bit)
Field
Bit High
Bit Low
Description
Access
cfg_t_param_mrr_to
_valid
3
0
Timing parameter for Mode Register
Read to any valid command.
Read/Write
cfg_t_param_mpr_to
_valid
8
4
Timing parameter for Multi Purpose
Register Read to any valid command.
Read/Write
cfg_t_param_mps_e
xit_cs_to_cke
12
9
Timing parameter for exit Maximum
Power Saving. Timing requirement for
CS assertion vs CKE de-assertion.
tMPX_S
Read/Write
continued...
External Memory Interface Handbook Volume 3: Reference Material
237
3 Functional Description—Intel Arria 10 EMIF IP
Field
Bit High
Bit Low
Description
Access
cfg_t_param_mps_e
xit_cke_to_cs
16
13
Timing parameter for exit Maximum
Power Saving. Timing requirement for
CKE de-assertion vs CS de-assertion.
tMPX_LH
Read/Write
cfg_t_param_rld3_m
ultibank_ref_delay
19
17
RLD3 Refresh to Refresh Delay for all
sequences.
Read/Write
cfg_t_param_mmr_c
md_to_valid
27
20
MMR cmd to valid delay.
Read/Write
3.23.33 caltiming9: Command/Address/Latency Parameters
address=40(32 bit)
Field
Bit High
Bit Low
cfg_t_param_4_act_t
o_act
7
0
Description
The four-activate window timing
parameter.
Access
Read/Write
3.23.34 caltiming10: Command/Address/Latency Parameters
address=41(32 bit)
Field
Bit High
Bit Low
cfg_t_param_16_act
_to_act
7
0
Description
The 16-activate window timing
parameter (RLD3).
Access
Read/Write
3.23.35 dramaddrw: Row/Column/Bank Address Width Configuration
address=42(32 bit)
Bit High
Bit Low
cfg_col_addr_width
4
0
The number of column address bits
for the memory devices in your
memory interface.
Read/Write
cfg_row_addr_width
9
5
The number of row address bits for
the memory devices in your memory
interface.
Read/Write
cfg_bank_addr_width
13
10
The number of bank address bits for
the memory devices in your memory
interface.
Read/Write
cfg_bank_group_add
r_width
15
14
The number of bank group address
bits for the memory devices in your
memory interface.
Read/Write
cfg_cs_addr_width
18
16
The number of chip select address
bits for the memory devices in your
memory interface.
Read/Write
Field
External Memory Interface Handbook Volume 3: Reference Material
238
Description
Access
3 Functional Description—Intel Arria 10 EMIF IP
3.23.36 sideband0: Sideband
address=43(32 bit)
Field
mr_cmd_trigger
Bit High
Bit Low
0
0
Description
Write to 1 to trigger the execution of
the mode register command.
Access
Read/Write
3.23.37 sideband1: Sideband
address=44(32 bit)
Field
mmr_refresh_req
Bit High
Bit Low
3
0
Description
When asserted, indicates Refresh
request to the specific rank. Each bit
corresponds to each rank.
Access
Read/Write
3.23.38 sideband2: Sideband
address=45(32 bit)
Field
Bit High
Bit Low
mmr_zqcal_long_req
0
0
Description
When asserted, indicates long ZQ cal
request. This bit is write clear.
Access
Read/Write
3.23.39 sideband3: Sideband
address=46(32 bit)
Field
Bit High
Bit Low
Description
Access
mmr_zqcal_short_re
q
0
0
When asserted, indicates short ZQ cal
request. This bit is write clear.
Read/Write
Bit High
Bit Low
Description
Access
3
0
3.23.40 sideband4: Sideband
address=47(32 bit)
Field
mmr_self_rfsh_req
When asserted, indicates self refresh
request to the specific rank. Each bit
corresponds to each rank. These bits
are write clear.
Read/Write
External Memory Interface Handbook Volume 3: Reference Material
239
3 Functional Description—Intel Arria 10 EMIF IP
3.23.41 sideband5: Sideband
address=48(32 bit)
Field
mmr_dpd_mps_req
Bit High
Bit Low
Description
Access
0
0
When asserted, indicates deep power
down or max power saving request.
This bit is write clear.
Read/Write
Bit High
Bit Low
Description
Access
0
0
3.23.42 sideband6: Sideband
address=49(32 bit)
Field
mr_cmd_ack
Acknowledge to mode register
command.
Read
3.23.43 sideband7: Sideband
address=50(32 bit)
Field
mmr_refresh_ack
Bit High
Bit Low
0
0
Description
Acknowledge to indicate refresh is in
progress.
Access
Read
3.23.44 sideband8: Sideband
address=51(32 bit)
Field
mmr_zqcal_ack
Bit High
Bit Low
0
0
Description
Acknowledge to indicate ZQCAL is
progressing.
Access
Read
3.23.45 sideband9: Sideband
address=52(32 bit)
Field
mmr_self_rfsh_ack
Bit High
Bit Low
Description
Access
0
0
Acknowledge to indicate self refresh is
progressing.
Read
3.23.46 sideband10: Sideband
External Memory Interface Handbook Volume 3: Reference Material
240
3 Functional Description—Intel Arria 10 EMIF IP
address=53(32 bit)
Field
mmr_dpd_mps_ack
Bit High
Bit Low
0
0
Description
Acknowledge to indicate deep power
down/max power saving is in
progress.
Access
Read
3.23.47 sideband11: Sideband
address=54(32 bit)
Field
mmr_auto_pd_ack
Bit High
Bit Low
0
0
Description
Acknowledge to indicate auto power
down is in progress.
Access
Read
3.23.48 sideband12: Sideband
address=55(32 bit)
Field
Bit High
Bit Low
Description
Access
mr_cmd_type
2
0
Indicates the type of Mode Register
Command.
Read/Write
mr_cmd_rank
6
3
Indicates which rank the mode
register command is intended to.
Read/Write
3.23.49 sideband13: Sideband
address=56(32 bit)
Field
mr_cmd_opcode
Bit High
Bit Low
Description
Access
31
0
[31:27] reserved. Register Command
Opcode Information to be used for
Register Command LPDDR3 [26:20]
Reserved [19:10] falling edge CA.
[9:0] rising edge CA DDR4 [26:24]
C2:C0 [23] ACT [22:21] BG1:BG0
[20] Reserved [19:18] BA1:BA0 [17]
A17 [16] RAS [15] CAS [14] WE
[13:0] A13:A0 DDR3 [26:21]
Reserved [20:18] BA2:BA0 [17] A17
[16] RAS [15] CAS [14] WE [13]
Reserved [12:0] A12:A0 RLDRAM3
[26] Reserved [25:22] BA3:BA0 [21]
REF [20] WE [19:0] A19:A0
Read/Write
3.23.50 sideband14: Sideband
External Memory Interface Handbook Volume 3: Reference Material
241
3 Functional Description—Intel Arria 10 EMIF IP
address=57(32 bit)
Field
mmr_refresh_bank
Bit High
Bit Low
Description
Access
15
0
User refresh bank information, binary
representation of bank address.
Enables refresh to that bank address
when requested.
Read/Write
Bit High
Bit Low
Description
Access
3
0
Setting to 1 to stall the corresponding
rank.
Read/Write
3.23.51 sideband15: Sideband
address=58(32 bit)
Field
mmr_stall_rank
3.23.52 dramsts: Calibration Status
address=59(32 bit)
Field
Bit High
Bit Low
Description
Access
phy_cal_success
0
0
This bit will be set to 1 if the PHY was
able to successfully calibrate.
Read
phy_cal_fail
1
1
This bit will be set to 1 if the PHY was
unable to calibrate.
Read
3.23.53 ecc1: ECC General Configuration
address=128(32 bit)
Bit High
Bit Low
cfg_enable_ecc
0
0
A value of 1 enables the ECC
encoder/decoder.
Read/Write
cfg_enable_dm
1
1
A value of 1 indicate that this is a
design with DM.
Read/Write
cfg_enable_rmw
2
2
A value of 1 enables the RMW
feature, including partial/dummy
write support.
Read/Write
cfg_data_rate
6
3
Set this value to 2, 4, or 8 for full,
half or quarter rate designs.
cfg_ecc_in_protocol
7
7
Set this value to 1 for Avalon-MM or 0
for Avalon-ST input interface. Readonly register.
Read/Write
cfg_enable_auto_cor
r
8
8
A value of 1 enables the auto
correction feature, injecting a dummy
write command after a single-bit error
Read/Write
Field
Description
Access
Read
continued...
External Memory Interface Handbook Volume 3: Reference Material
242
3 Functional Description—Intel Arria 10 EMIF IP
Field
Bit High
Bit Low
Description
Access
is (SBE) detected. This feature must
be enabled together with RMW and
ECC.
cfg_enable_ecc_code
_overwrite
Reserved
9
9
31
10
A value of 1 enables the ECC code
overwrite feature. Reuse the original
read-back ECC code during RMW if a
double-bit error (DBE) is detected.
Read/Write
3.23.54 ecc2: Width Configuration
address=129(32 bit)
Field
Bit High
Bit Low
Description
Access
cfg_dram_data_widt
h
7
0
Set this value to the DRAM data
width, without taking rate conversion
into consideration. For example, for a
64+8=72 bit ECC design, set this
register to ‘d72.
Read
cfg_local_data_width
15
8
Set this value to the LOCAL data
width, without taking rate conversion
into consideration. For example, for a
64+8=72 DQ bit ECC design, set this
register to ‘d64.
Read
cfg_addr_width
21
16
Set this value to the LOCAL address
width, which corresponds to the
address field width of the cmd field.
For example, for a LOCAL address
width of 24 bits, set this value to
‘d24.
Read
Reserved
31
22
3.23.55 ecc3: ECC Error and Interrupt Configuration
address=130(32 bit)
Bit High
Bit Low
Description
Access
cfg_gen_sbe
0
0
A value of 1 enables the generate SBE
feature. Generates a single bit error
during the write process.
Read/Write
cfg_gen_dbe
1
1
A value of 1 enables the generate
DBE feature. Generates a double bit
error during the write process.
Read/Write
cfg_enable_intr
2
2
A value of 1 enables the interrupt
feature. The interrupt signal notifies if
an error condition occurs. The
condition is configurable.
Read/Write
cfg_mask_sbe_intr
3
3
A value of 1 masks the interrupt
signal when SBE occurs.
Read/Write
Field
continued...
External Memory Interface Handbook Volume 3: Reference Material
243
3 Functional Description—Intel Arria 10 EMIF IP
Field
Bit High
Bit Low
cfg_mask_dbe_intr
4
4
A value of 1 masks the interrupt
signal when DBE occurs.
Read/Write
cfg_mask_corr_drop
ped_intr
5
5
A value of 1 masks the interrupt
signal when auto correction command
can’t be scheduled, due to backpressure (FIFO full).
Read/Write
cfg_mask_hmi_intr
6
6
A value of 1 masks the interrupt
signal when the hard memory
interface asserts an interrupt signal
via hmi_interrupt port.
Read/Write
cfg_clr_intr
7
7
Writing a vale of 1 to this self-clearing
bit clears the interrupt signal, error
status, and address.
Read/Write
31
9
Reserved
Description
Access
3.23.56 ecc4: Status and Error Information
address=144(32 bit)
Field
Bit High
Bit Low
Description
Access
sts_ecc_intr
0
9
Indicates the interrupt status; a value
of 1 indicates interrupt occurred.
Read
sts_sbe_error
1
1
Indicates the SBE status; a value of 1
indicates SBE occurred.
Read
sts_dbe_error
2
2
Indicates the DBE status; a value of 1
indicates DBE occurred.
Read
sts_corr_dropped
3
3
Indicates the status of correction
command dropped; a value of 1
indicates correction command
dropped.
Read
sts_sbe_count
7
4
Indicates the number of times SBE
error has occurred. The counter will
overflow.
Read
sts_dbe_count
11
8
Indicates the number of times DBE
error has occurred. The counter will
overflow.
Read
sts_corr_dropped_co
unt
15
12
Indicates the number of times
correction command has dropped.
Counter will overflow.
Read
Reserved
31
16
3.23.57 ecc5: Address of Most Recent SBE/DBE
External Memory Interface Handbook Volume 3: Reference Material
244
3 Functional Description—Intel Arria 10 EMIF IP
address=145(32 bit)
Field
sts_err_addr*
Bit High
Bit Low
31
0
Description
Address of the most
recent single-bit error
or double-bit error.
Access
Read
3.23.58 ecc6: Address of Most Recent Correct Command Dropped
address=146(32 bit)
Bit High
Field
sts_corr_dropped_add
r
Bit Low
31
0
Description
Address of the most
recent correction
command dropped.
Access
Read
3.24 Document Revision History
Date
Version
Changes
May 2017
2017.05.08
•
•
Added Using the EMIF Debug Toolkit with Arria 10 HPS Interfaces topic.
Rebranded as Intel.
October 2016
2016.10.31
•
Added Back-to-Back User-Controlled Refresh for Hard Memory Controller
topic.
Modified first bullet point in the ECC in Arria 10 EMIF IP topic.
Added additional content about read-modify-write operations to the ECC in
Arria 10 EMIF IP topic.
•
•
•
•
May 2016
2016.05.02
•
•
•
•
•
•
•
November 2015
2015.11.02
•
•
•
•
•
•
•
Removed afi_alert_n from the AFI Address and Command Signals
table in the AFI Address and Command Signals topic.
Added ecc5: Address of Most Recent SBE/DBE and ecc6: Address of Most
recent Correct Command Dropped to Memory Mapped Register (MMR)
Tables section.
Modified text of Hard Memory Controller and Hard PHY and Soft Memory
Controller and Hard PHY.
Modified content of the Ping Pong PHY Architecture topic.
Added section on read-modify-write operations, to ECC in Arria 10 EMIF IP
topic.
Added AFI 4.0 Timing Diagrams section.
Added Arria 10 EMIF Latency section.
Added Arria 10 EMIF Calibration Times section.
Added Integrating a Custom Controller with the Hard PHY section.
Added LPDDR3 support to AFI Parameters, AFI Address and Command
Signals, and AFI Write Data Signals.
Revised rendering of address information in the Memory Mapped Register
(MMR) Tables.
Added ecc1: ECC General Configuration, ecc2: Width Configuration, ecc3:
ECC Error and Interrupt Configuration, and ecc4: Status and Error
Information in the Memory Mapped Register (MMR) Tables.
Removed RZQ Pin Sharing section.
Added bank numbers to figure in Restrictions on I/O Bank Usage for Arria
10 EMIF IP with HPS topic.
Added Arria 10 EMIF and SmartVID topic.
Changed instances of Quartus II to Quartus Prime.
continued...
External Memory Interface Handbook Volume 3: Reference Material
245
3 Functional Description—Intel Arria 10 EMIF IP
Date
Version
Changes
May 2015
2015.05.04
Maintenance release.
December 2014
2014.12.15
•
•
•
•
•
•
August 2014
2014.08.15
Added debug-related connections to the Logical Connections table in the
Logical Connections topic.
Added Arria 10 EMIF Ping Pong PHY section.
Added Arria 10 EMIF Debugging Examples section.
Added x4 mode support.
Added AFI 4.0 Specification section.
Added MMR Register Tables section.
•
Added PHY-only support for DDR3, DDR4, and RLDRAM 3, and soft
controller and hard PHY support for RLDRAM 3 and QDR II/II+/II+ Xtreme
to Supported Memory Protocols table.
•
Added afi_conduit_end, afi_clk_conduit_end,
afi_half_clk_conduit_end, and afi_reset_n_conduit_end to
•
Expanded description of ctrl_amm_avalon_slave in Logical
Connections table.
Added ECC in Arria 10 EMIF IP.
Added Configuring Your EMIF IP for Use with the Debug Toolkit.
Added Arria 10 EMIF for Hard Processor Subsystem.
Logical Connections table.
•
•
•
December 2013
2013.12.16
Initial release.
External Memory Interface Handbook Volume 3: Reference Material
246
4 Functional Description—Intel MAX® 10 EMIF IP
4 Functional Description—Intel MAX® 10 EMIF IP
Intel MAX® 10 FPGAs provide advanced processing capabilities in a low-cost, instanton, small-form-factor device featuring capabilities such as digital signal processing,
analog functionality, Nios II embedded processor support and memory controllers.
Figure 135. MAX 10 EMIF Block Diagram
External Memory Interface IP
PLL
External
Memory
Device
I/O Structure
PHY
Reference Clock
Calibration
Sequencer
DQ I/O
I/O Block
Write Path
Memory
Controller
Write Path
Address/Command
Path
4.1 MAX 10 EMIF Overview
MAX 10 FPGAs provide advanced processing capabilities in a low-cost, instant-on,
small-form-factor device featuring capabilities such as digital signal processing, analog
functionality, Nios II embedded processor support and memory controllers.
Intel Corporation. All rights reserved. Intel, the Intel logo, Altera, Arria, Cyclone, Enpirion, MAX, Nios, Quartus
and Stratix words and logos are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other
countries. Intel warrants performance of its FPGA and semiconductor products to current specifications in
accordance with Intel's standard warranty, but reserves the right to make changes to any products and services
at any time without notice. Intel assumes no responsibility or liability arising out of the application or use of any
information, product, or service described herein except as expressly agreed to in writing by Intel. Intel
customers are advised to obtain the latest version of device specifications before relying on any published
information and before placing orders for products or services.
*Other names and brands may be claimed as the property of others.
ISO
9001:2008
Registered
4 Functional Description—Intel MAX® 10 EMIF IP
Figure 136. MAX 10 EMIF Block Diagram
External Memory Interface IP
PLL
I/O Structure
PHY
Reference Clock
Calibration
Sequencer
External
Memory
Device
DQ I/O
I/O Block
Write Path
Memory
Controller
Read Path
Address/Command
Path
4.2 External Memory Protocol Support
MAX 10 FPGAs offer external memory interface support for DDR2, DDR3, DDR3L, and
LPDDR2 protocols.
Table 59.
Supported External Memory Configurations
External Memory Protocol
Maximum Frequency
Configuration
DDR3, DDR3L
303 MHz
x16 + ECC + Configuration and Status
Register (CSR)
DDR2
200 MHz
x16 + ECC + Configuration and Status
Register (CSR)
LPDDR21
200 MHz2
x16
4.3 MAX 10 Memory Controller
MAX 10 FPGAs use the HPC II external memory controller.
4.4 MAX 10 Low Power Feature
The MAX 10 low power feature is automatically activated when the self refresh or low
power down modes are activated. The low power feature sends the
afi_mem_clk_disable signal to stop the clock used by the controller.
1 MAX 10 devices support only single-die LPPDR2.
2 To achieve the specified performance, constrain the memory device I/O and core power supply
variation to within ±3%. By default, the frequency is 167 MHz.
External Memory Interface Handbook Volume 3: Reference Material
248
4 Functional Description—Intel MAX® 10 EMIF IP
To conserve power, the MAX 10 UniPHY IP core performs the following functions:
Note:
•
Tri-states the address and command signals except CKE and RESET_N signals
•
Disables the input buffer of DDR input
The MAX 10 low power feature is available from version 15.0 of the Quartus Prime
software. To enable this feature, regenerate your MAX 10 UniPHY IP core using the
Quartus Prime software version 15.0 or later.
4.5 MAX 10 Memory PHY
MAX 10 devices employ UniPHY, but without using DLLs for calibration. The physical
layer implementation for MAX 10 external memory interfaces provides a calibrated
read path and a static write path.
4.5.1 Supported Topologies
The memory PHY supports DDR2 and DDR3 protocols with up to two discrete memory
devices, and the LPDDR2 protocol with one discrete memory device.
4.5.2 Read Datapath
One PLL output is used to capture data from the memory during read operations. The
clock phase is calibrated by the sequencer before the interface is ready for use.
Figure 137. Read Datapath
Full Rate
Half Rate
Read
Burst
d0/d1/d2/d3
Double Data Rate
FPGA
FPGA
Read
Latency
LFIFO
Group 0
d2/d3
d0/d1
LFIFO
Group 1
d2/d3
d0/d1
LFIFO
Group 2
d2/d3
d0/d1
HR Register
DQ Capture
DDR to SDR
HR Register
DQ Capture
DDR to SDR
HR Register
DQ Capture
DDR to SDR
VFIFO
Group 0
d0/d1/d2/d3
VFIFO
Group 1
d0/d1/d2/d3
VFIFO
Group 2
Address/Command/Clock
I/O
d3
d2
d1
d0
d3
d2
d1
d0
MEM
Group 0
I/O
d3
d2
d1
d0
d3
d2
d1
d0
MEM
Group 1
I/O
d3
d2
d1
d0
d3
d2
d1
d0
PLL
Device 0
Device 1
MEM
Group 2
Capture Clock 0
For DDR3 interfaces, two PLL outputs capture data from the memory devices during a
read. In a 24-bit interface—whether the top 8 bits are used by ECC or not—the
supported topology is two discrete DDR3 devices of 16-bit and 8-bit DQ each. Each
discrete device has a dedicated capture clock output from the PLL.
For LPDDR2 interfaces, the supported configuration is a single memory device with
memory width of 16-bit DQ. The other PLL output is used for DQS tracking purposes,
because the tDQSCK drift might cause data capture to fail otherwise. The tracking
clock is a shifted capture clock used to sample the DQS signal. By capturing the DQS
signal, the system can compensate for DQS signal drift.
External Memory Interface Handbook Volume 3: Reference Material
249
4 Functional Description—Intel MAX® 10 EMIF IP
Figure 138. Read Datapath Timing Diagram
Soft
afi Clock
afi_rdata_en
write_enable for LFIFO (afi_rdata_valid)
Data Transferred Marked as Valid
dcba
hgfe
VFIFO Pipe
read_enable for LFIFO
Data Transferred Marked as Valid
dcba
afi_clk Captured Data (after rdata_fifo)
dcba
hgfe
hgfe
Capture Clock/2
HR Register Output (Clocked by Div/2 Clock)
dcba
Second Flopped Data
First DDIO Data Captured on Soft
ba
hgfe
ba
dc
fe
dc
fe
hg
hg
Hard
Capture Clock
Read mem_dq
a
b
DDIO Output
c
d
ba
e
dc
f
g
fe
h
hg
4.5.3 Write Datapath
The write datapath is a static path and is timing analyzed to meet timing
requirements.
Figure 139. Write Datapath
Full Rate
Half Rate
Double Data Rate
FPGA
d0/d1/d2/d3
d0/d1/d2/d3
d0/d1/d2/d3
Address/Command/Clock
HR to FR
HR to FR
HR to FR
d2/d3
d0/d1
d2/d3
d0/d1
d2/d3
d0/d1
External Memory Interface Handbook Volume 3: Reference Material
250
I/O
HR to FR
I/O
HR to FR
I/O
HR to FR
Device 0
d3
d2
d1
d0
MEM
Group 0
d3
d2
d1
d0
MEM
Group 1
d3
d2
d1
d0
Device 1
MEM
Group 2
4 Functional Description—Intel MAX® 10 EMIF IP
In the PHY write data path, for even write latency the write data valid, write
data, and dqs enable pass through one stage of fr_cycle_shifter in a flow
through path. For odd memory write latency, the output is shifted by a full-rate clock
cycle. The full-rate cycle-shifted output feeds into a simple DDIO.
Figure 140. Write Datapath Timing Diagram
soft
afi Clock
Write Data
afi_wdata
abcd
efgh
phy_ddio_dq
(after fr_cycle_shifter)
cdxx
ghab
xxef
Multiplexer Select
WR DATA Hi
WR DATA Lo
x
c
a
g
e
x
x
d
b
h
f
x
Write Data Valid
afi_wdata_valid[0]
afi_wdata_valid[1]
phy_ddio_wrdata_en[0]
(after fr_cycle_shifter)
phy_ddio_wrdata_en[1]
(after fr_cycle_shifter)
afi_dqs_burst[0]
DQS Enable
afi_dqs_burst[1]
phy_ddio_dqs_en[0]
(after fr_cycle_shifter)
phy_ddio_dqs_en[1]
(after fr_cycle_shifter)
Multiplexer Select
DQS_OE
DQ_OE
Memory Clock
hard
Transferred DQS_OE
Transferred DQ_OE
adc Clock
mem_dq
d c b a h g f e
mem_dqs
mem_dqs_n
External Memory Interface Handbook Volume 3: Reference Material
251
4 Functional Description—Intel MAX® 10 EMIF IP
4.5.4 Address and Command Datapath
Implementation of the address and command datapath differs between DDR2/DDR3
and LPDDR2, because of the double data-rate command in LPDDR2.
LPDDR2
For LPDDR2, CA is four bits wide on the AFI interface. The least significant bit of CA is
captured by a negative-edge triggered flop, and the most-significant bit of CA is
captured by a positive-edge triggered flop, before multiplexing to provide maximum
setup and hold margins for the AFI clock-to-MEM clock data transfer.
DDR2/DDR3
For DDR2/DDR3, the full-rate register and MUX architecture is similar to LPDDR2, but
both phases are driven with the same signal. The DDR3 address and command signal
is not clocked by the write/ADC clock as with LPDDR2, but by the inverted MEM clock,
for better address and command margin.
Chip Select
Because the memory controller can drive both phases of cs_n in half-rate, the signal
is fully exposed to the AFI side.
4.5.5 Sequencer
The sequencer employs an RTL-based state machine which assumes control of the
interface at reset (whether at initial startup or when the IP is reset) and maintains
control throughout the calibration process. The sequencer relinquishes control to the
memory controller only after successful calibration.
The sequencer consists of a calibration state machine, together with a read-write (RW)
manager, PHY Manager, and PLL Manager.
Figure 141. Sequencer Architecture
Memory
Controller
AFI
Multiplexer
Calibration
State Machine
Read/Write
Manager
Sequencer
External Memory Interface Handbook Volume 3: Reference Material
252
PHY
Manager
Datapath
PLL
Manager
PLL
PD
4 Functional Description—Intel MAX® 10 EMIF IP
Figure 142. Calibration State Machine Stages
Start
Protocol?
DDR2 Initialization
Sequence
DDR3 Initialization
Sequence
LPDDR2 Initialization
Sequence
Precharge and
Activate
Guaranteed Write
Read Data and
Check Data
Data Match?
no
yes
VFIFO Max?
no
Increment VFIFO
yes
VFIFO Calibrated
Increment PLL
Increment PLL
Read Data and
Check Data
Data Match?
yes
no
Decrement PLL Half
Amount of Phase
Bert Margin
Established
Precharge and
FIFO Reset
Protocol?
DDR2 User Mode
Register
DDR3 User Mode
Register
LPDDR2 User Mode
Register
Invert AFI Multiplexer
and Assert cal Success
VT Tracking
External Memory Interface Handbook Volume 3: Reference Material
253
4 Functional Description—Intel MAX® 10 EMIF IP
RW Manager
The read-write (RW) manager encapsulates the protocol to read and write to the
memory device through the Altera PHY Interface (AFI). It provides a buffer that stores
the data to be sent to and read from memory, and provides the following commands:
•
Write configuration—configures the memory for use. Sets up burst lengths, read
and write latencies, and other device specific parameters.
•
Refresh—initiates a refresh operation at the DRAM. The sequencer also provides a
register that determines whether the RW manager automatically generates refresh
signals.
•
Enable or disable multi-purpose register (MPR)—for memory devices with a special
register that contains calibration specific patterns that you can read, this
command enables or disables access to the register.
•
Activate row—for memory devices that have both rows and columns, this
command activates a specific row. Subsequent reads and writes operate on this
specific row.
•
Precharge—closes a row before you can access a new row.
•
Write or read burst—writes or reads a burst length of data.
•
Write guaranteed—writes with a special mode where the memory holds address
and data lines constant. Intel guarantees this type of write to work in the presence
of skew, but constrains to write the same data across the entire burst length.
•
Write and read back-to-back—performs back-to-back writes or reads to adjacent
banks. Most memory devices have strict timing constraints on subsequent
accesses to the same bank, thus back-to-back writes and reads have to reference
different banks.
•
Protocol-specific initialization—a protocol-specific command required by the
initialization sequence.
PHY Manager
The PHY Manager provides access to the PHY for calibration, and passes relevant
calibration results to the PHY. For example, the PHYManager sets the VFIFO and LFIFO
buffer parameters resulting from calibration, signals the PHY when the memory
initialization sequence finishes, and reports the pass/fail status of calibration.
PLL Manager
The PLL Manager controls the phase of capture clocks during calibration. The output
phases of individual PLL outputs can be dynamically adjusted relative to each other
and to the reference clock without having to load the scan chain of the PLL. The phase
is shifted by 1/8th of the period of the voltage-controlled oscillator (VCO) at a time.
The output clocks are active during this dynamic phase-shift process.
A PLL counter increments with every phase increase and decrements with every phase
reduction. The PLL Manager records the amount by which the PLL counter has shifted
since the last reset, enabling the sequencer and tracking manager to determine
whether the phase boundary has been reached.
External Memory Interface Handbook Volume 3: Reference Material
254
4 Functional Description—Intel MAX® 10 EMIF IP
4.6 Calibration
The purpose of calibration is to exercise the external memory interface to find an
optimal window for capture clock. There is no calibration for writing data or for
address and command output; these paths are analyzed by TimeQuest to meet timing
requirements.
The calibration algorithm assumes that address and command signals can reliably be
sent to the external memory device, and that write data is registered with DQS.
4.6.1 Read Calibration
Read calibration consists of two primary parts: capture clock and VFIFO buffer
calibration, and read latency tuning.
A VFIFO buffer is a FIFO buffer that is calibrated to reflect the read latency of the
interface. The calibration process selects the correct read pointer of the FIFO. The
VFIFO function delays the controller's afi_rdata_valid signal to align to data
captured internally to the PHY. LFIFO buffers are FIFO buffers which ensure that data
from different DQS groups arrives at the user side at the same time. Calibration
ensures that data arrives at the user side with minimal latency.
Capture Clock and VFIFO Calibration
Capture clock and VFIFO calibration performs the following steps:
•
Executes a guaranteed write routine in the read-write (RW) manager.
•
Sweeps VFIFO buffer values, beginning at 0, in the PHY Manager.
—
•
For each adjustment of the VFIFO buffer, sweeps the capture clock phase in
the PLL Manager to find the first phase that works. This is accomplished by
issuing a read to the RW Manager and performing a bit check. Data is
compared for all data bits.
Increments the capture clock phase in the PLL Manager until capture clock phase
values are exhausted or until the system stops working. If there are no more
capture clock phase values to try, calibration increments the VFIFO buffer value in
the PHY Manager, and phase sweep again until the system stops working.
Completion of this step establishes a working range of values.
•
As a final step, calibration centers the capture clock phase within the working
range.
Read Latency Tuning
Read latency tuning performs the following steps to achieve the optimal latency value:
•
Assigns one LFIFO buffer for each DQ group.
•
Aligns read data to the AFI clock.
•
Gradually reduces LFIFO latency until reads fail, then increases latency to find the
minimum value that yields reliable operation.
4.6.2 Write Calibration
There is no calibration routine for writing data.
External Memory Interface Handbook Volume 3: Reference Material
255
4 Functional Description—Intel MAX® 10 EMIF IP
4.7 Sequencer Debug Information
Following calibration, the sequencer loads a set of debug information onto an output
port. You can use the SignalTap II Logic Analyzer to access the debug information in
the presynthesized design.
You could also bring this port to a register, to latch the value and make it accessible to
a host processor.
The output port where the debug information is available is not normally connected to
anything, so it could be removed during synthesis.
Signal name: phy_cal_debug_info
Module: <corename>_s0.v
The signal is 32 bits wide, and is defined in the following table.
Table 60.
Sequencer Debug Data
best_comp_result [23:16]
Best data comparison result on a per-pin basis from the
Read-Write Manager. Pin 16 - 23.
Any mismatch on respective pin data comparison produces
a high bit.
If calibration fails, the result is a non-zero value. When
calibration passes, the result is zero.
best_comp_result [15:8]
Best data comparison result on a per-pin basis from the
Read-Write Manager. Pin 8 - 15.
Any mismatch on respective pin data comparison produces
a high bit.
If calibration fails, the result is a non-zero value. When
calibration passes, the result is zero.
best_comp_result [7:0]
Best data comparison result on a per-pin basis from the
Read-Write Manager. Pin 0 - 7.
Any mismatch on respective pin data comparison produces
a high bit.
If calibration fails, the result is a non-zero value. When
calibration passes, the result is zero.
margin [7:0]
Margin found by the sequencer if calibration passes.
Number represents the amount of subsequent PLL phase
where valid data is found.
If calibration fails, this number is zero. When calibration
passes, this value is non-zero.
Debug Example and Interpretation
The following table illustrates possible debug signal values and their interpretations.
best_comp_result
[23:16]
best_comp_result
[15:8]
best_comp_result
[7:0]
margin [7:0]
Passing result
0000 0000
0000 0000
0000 0000
0001 0000
Interpretation
No failing pins.
No failing pins.
No failing pins.
16 phases of margin
for valid window. Ideal
case for 300Mhz
interface.
continued...
External Memory Interface Handbook Volume 3: Reference Material
256
4 Functional Description—Intel MAX® 10 EMIF IP
best_comp_result
[23:16]
best_comp_result
[15:8]
best_comp_result
[7:0]
margin [7:0]
Failing result
0010 0000
1111 1111
0000 0000
0000 0000
Interpretation
Pin 21 failing.
All pins failing, pin
8-15.
No failing pins.
No valid window.
4.8 Register Maps
This topic provides register information for MAX 10 EMIF.
Table 61.
Controller Register Map
Address
Description
0x100 - 0x126
Reserved
0x130
ECC control register
0x131
ECC status register
0x132
ECC error address register
UniPHY does not have a configuration and status register. The HPC II controller does
have a configuration and status register, when configured for ECC.
Related Links
Soft Controller Register Map on page 371
The soft controller register map allows you to control the soft memory controller
settings.
4.9 Document Revision History
Table 62.
Document Revision History
Date
Version
Changes
May 2017
2017.05.08
Rebranded as Intel.
October 2016
2016.10.31
Maintenance release.
May 2016
2016.05.02
Maintenance release.
November 2015
2015.11.02
Changed instances of Quartus II to
Quartus Prime.
May 2015
2015.05.04
•
•
December 2014
2014.12.15
Updated the external memory
protocol support.
Added topic about the low power
feature.
Initial release.
External Memory Interface Handbook Volume 3: Reference Material
257
6 Functional Description—Hard Memory Interface
6 Functional Description—Hard Memory Interface
The hard (on-chip) memory interface components are available in the Arria V and
Cyclone V device families.
The Arria V device family includes hard memory interface components supporting
DDR2 and DDR3 SDRAM memory protocols at speeds of up to 533 MHz. For the
Quartus II software version 12.0 and later, the Cyclone V device family supports both
hard and soft interface support.
The Arria V device family supports both hard and soft interfaces for DDR3 and DDR2,
and soft interfaces for LPDDR2 SDRAM, QDR II SRAM, and RLDRAM II memory
protocols. The Cyclone V device family supports both hard and soft interfaces for
DDR3, DDR2, and LPDDR2 SDRAM memory protocols.
The hard memory interface consists of three main parts, as follows:
•
The multi-port front end (MPFE), which allows multiple independent accesses to
the hard memory controller.
•
The hard memory controller, which initializes, refreshes, manages, and
communicates with the external memory device.
•
The hard PHY, which provides the physical layer interface to the external memory
device.
Intel Corporation. All rights reserved. Intel, the Intel logo, Altera, Arria, Cyclone, Enpirion, MAX, Nios, Quartus
and Stratix words and logos are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other
countries. Intel warrants performance of its FPGA and semiconductor products to current specifications in
accordance with Intel's standard warranty, but reserves the right to make changes to any products and services
at any time without notice. Intel assumes no responsibility or liability arising out of the application or use of any
information, product, or service described herein except as expressly agreed to in writing by Intel. Intel
customers are advised to obtain the latest version of device specifications before relying on any published
information and before placing orders for products or services.
*Other names and brands may be claimed as the property of others.
ISO
9001:2008
Registered
6 Functional Description—Hard Memory Interface
Figure 143. Hard Memory Interface Architecture
Hard Memory Interface Architecture
FPGA Fabric
C ommand C ommand
Port 1
Port 2
Avalon-MM / ST
Adaptor
C ommand
Port 6
D ata
Port 1
D ata
Port 4
Write Data Write Data
Transaction Transaction
F IF O
F IF O
Transaction
F IF O
Data
F IF O
Data
F IF O
D ata
Port 4
D ata
Port 1
Read Data Read Data
Data
F IF O
Data
F IF O
Multi-Port
Front End
Register
Port
Bus Transaction
Register
Interface
Bus Transaction
DRAM Burst State Machine
Write Data
Handling
DRAM Burst
Command
Single-Port
Controller
Slot 1
Slot 2
Slot 3
Slot 8
Command
Arbitration
DRAM
Command
Read Data
Handling
Data
Data
ECC
Calc
ECC
Check
Write D a ta
Register
Control
Avalon-MM
Memory Mapped
Register
R ea d D a ta
Hard PHY
6.1 Multi-Port Front End (MPFE)
The multi-port front end and its associated fabric interface provide up to six command
ports, four read-data ports and four write-data ports, through which user logic can
access the hard memory controller. Each port can be configured as read only or write
only, or read and write ports may be combined to form bidirectional data ports. Ports
can be 32, 64, 128, or 256 data bits wide, depending on the number of ports used and
the type (unidirectional or bidirectional) of the port.
Fabric Interface
The fabric interface provides communication between the Avalon-ST-like internal
protocol of the hard memory interface and the external Avalon-MM protocol. The fabric
interface supports frequencies in the range of 10 MHz to one-half of the memory
interface frequency. For example, for an interface running at 533 MHz, the maximum
user logic frequency is 267 MHz. The MPFE handles the clock crossing between user
logic and the hard memory interface.
The multi-port front end read and write FIFO depths are 8, and the command FIFO
depth is 4. The FIFO depths are not configurable.
Operation Ordering
Requests arriving at a given port are executed in the order in which they are received.
Requests arriving at different ports have no guaranteed order of service, except when
a first transaction has completed before the second arrives.
External Memory Interface Handbook Volume 3: Reference Material
259
6 Functional Description—Hard Memory Interface
6.2 Multi-port Scheduling
Multi-port scheduling is governed by two considerations: the absolute priority of a
request and the weighting of a port. User-configurable priority and weight settings
determine the absolute and relative scheduling policy for each port.
6.2.1 Port Scheduling
The evaluation of absolute priority ensures that ports carrying higher-priority traffic
are served ahead of ports carrying lower-priority traffic. The scheduler recognizes
eight priority levels, with higher values representing higher priorities. Priority is
absolute; for example, any transaction with priority seven will always be scheduled
before transactions of priority six or lower.
When ports carry traffic of the same absolute priority, relative priority is determined
based on port weighting. Port weighting is a five-bit value, and is determined by a
weighted round robin (WRR) algorithm.
The scheduler can alter priority if the latency target for a transaction is exceeded. The
scheduler tracks latency on a per-port basis, and counts the cycles that a transaction
is pending. Each port has a priority escalation register and a pending counter
engagement register. If the number of cycles in the pending counter engagement
register elapse without a pending transaction being served, that transaction’s priority
is esclated.
To ensure that high-priority traffic is served quickly and that long and short bursts are
effectively interleaved on ports, bus transactions longer than a single DRAM burst are
scheduled as a series of DRAM bursts, with each burst arbitrated separately.
The scheduler uses a form of deficit round robin (DRR) scheduling algorithm which
corrects for past over-servicing or under-servicing of a port. Each port has an
associated weight which is updated every cycle, with a user-configured weight added
to it and the amount of traffic served subtracted from it. The port with the highest
weighting is considered the most eligible.
To ensure that lower priority ports do not build up large running weights while higher
priority ports monopolize bandwidth, the hard memory controller’s DRR weights are
updated only when a port matches the scheduled priority. Hence, if three ports have
traffic, two being priority 7 and one being priority 4, the weights for both ports at
priority 7 are updated but the port with priority 4 remains unchanged.
6.2.2 DRAM Burst Scheduling
DRAM burst scheduling recognizes addresses that access the same column/row
combination—also known as open page accesses. Such operations are always served
in the order in which they are received in the single-port controller.
External Memory Interface Handbook Volume 3: Reference Material
260
6 Functional Description—Hard Memory Interface
Selection of DRAM operations is a two-stage process; first, each pending transaction
must wait for its timers to be eligible for execution, then the transaction arbitrates
against other transactions that are also eligible for execution.
The following rules govern transaction arbitration:
•
High priority operations take precedence over lower priority operations
•
If multiple operations are in arbitration, read operations have precedence over
write operations
•
If multiple operations still exist, the oldest is served first
A high-priority transaction in the DRAM burst scheduler wins arbitration for that bank
immediately if the bank is idle and the high-priority transaction’s chip select/row/
column address does not match an address already in the single-port controller. If the
bank is not idle, other operations to that bank yield until the high-priority operation is
finished. If the address matches another chip select/row/column, the high-priority
transaction yeilds until the earlier transaction is completed.
You can force the DRAM burst scheduler to serve transactions in the order that they
are received, by setting a bit in the register set.
6.2.3 DRAM Power Saving Modes
The hard memory controller supports two DRAM power-saving modes: self-refresh,
and fast/slow all-bank precharge powerdown exit. Engagement of a DRAM power
saving mode can occur due to inactivity, or in response to a user command.
The user command to enter power-down mode forces the DRAM burst-scheduling
bank-management logic to close all banks and issue the power-down command. You
can program the controller to power down when the DRAM burst-scheduling queue is
empty for a specified number of cycles; the DRAM is reactivated when an active DRAM
command is received.
6.3 MPFE Signal Descriptions
The following table describes the signals for the multi-port front end.
Table 63.
MPFE Signals
Signal
avl_<signal_name>_#
mp_cmd_clk_#_clk
(1)
(1)
(2)
mp_rfifo_reset_n_#_reset_n
mp_wfifo_clk_#_clk
—
(1)
mp_cmd_reset_n_#_reset_n
mp_rfifo_clk_#_clk
Direction
(2)
(2)
Description
Local interface signals.
Input
Clock for the command FIFO buffer. (3) Follow Avalon-MM
master frequency. Maximum frequency is one-half of the
interface frequency, and subject to timing closure.
Input
Asynchronous reset signal for command FIFO buffer.
Input
Clock for the read data FIFO buffer. Follow Avalon-MM
master frequency. Maximum frequency is one-half of the
interface frequency, and subject to timing closure.
Input
Asynchronous reset signal for read data FIFO buffer.
Input
Clock for the write data FIFO buffer. Follow Avalon-MM
master frequency. Maximum frequency is one-half of the
interface frequency, and subject to timing closure.
continued...
External Memory Interface Handbook Volume 3: Reference Material
261
6 Functional Description—Hard Memory Interface
Signal
mp_wfifo_reset_n_#_reset_n
bonding_in_1/2/3
bonding_out_1/2/3
Direction
(2)
Description
Input
Asynchronous reset signal for write data FIFO buffer.
Input
Bonding interface input port. Connect second controller
bonding output port to this port according to the port
sequence.
Output
Bonding interface output port. Connect this port to the
second controller bonding intput port according to the
port sequence.
Notes to Table:
1. # represents the number of the slave port. Values are 0—5.
2. # represents the number of the slave port. Values are 0—3.
3. The command FIFO buffers have two stages. The first stage is 4 bus transaction deep per port. After port scheduling,
the commands are placed in the second stage FIFO buffer, which is 8 DRAM transactions deep. The second stage FIFO
buffer is used to optimize memory operations where bank look ahead and data reordering occur. The write data buffer
is 32 deep and read buffer is 64 deep.
Every input interface (command, read data, and write data) has its own clock domain.
Each command port can be connected to a different clock, but the read data and write
data ports associated with a command port must connect to the same clock as that
command port. Each input interface uses the same reset signal as its clock.
By default, the IP generates all clock signals regardless of the MPFE settings, but all
unused ports and FIFO buffers are connected to ground.
The command ports can be used only in unidirectional configurations, with either 4
write and 2 read, 3 write and 3 read, or 2 write and 4 read scenarios. For bidirectional
ports, the number of clocks is reduced from 6 to a maximum of 4.
For the scenario depicted in the following figure:
•
command port 0 is associated with read and write data FIFO 0 and 1
•
command port 1 is associated with read data FIFO 2
•
command port 2 is associated with write data FIFO 2
External Memory Interface Handbook Volume 3: Reference Material
262
6 Functional Description—Hard Memory Interface
Figure 144. Sample MPFE Configuration
Therefore, if port 0 (avl_0) is clocked by a 100 MHz clock signal, mp_cmd_clk_0,
mp_rfifo_clk_0, mp_rfifo_clk_1, mp_wfifo_clk_0, and mp_wfifo_clk_1 must all be
connected to the same 100 MHz clock, as illustrated below.
External Memory Interface Handbook Volume 3: Reference Material
263
6 Functional Description—Hard Memory Interface
Figure 145. Sample Connection Mapping
6.4 Hard Memory Controller
The hard memory controller initializes, refreshes, manages, and communicates with
the external memory device.
Note:
The hard memory controller is functionally similar to the High Performance Controller
II (HPC II). For information on signals, refer to the Functional Description—HPC II
Controller chapter.
External Memory Interface Handbook Volume 3: Reference Material
264
6 Functional Description—Hard Memory Interface
Related Links
Functional Description—HPC II Controller on page 343
The High Performance Controller II works with the UniPHY-based DDR2, DDR3, and
LPDDR2 interfaces.
6.4.1 Clocking
The ports on the MPFE can be clocked at different frequencies, and synchronization is
maintained by cross-domain clocking logic in the MPFE.
Command ports can connect to different clocks, but the data ports associated with a
given command port must be attached to the same clock as that command port. For
example, a bidirectional command port that performs a 64-bit read/write function has
its read port and write port connected to the same clock as the command port. Note
that these clocks are separate from the EMIF core generated clocks.
6.4.2 Reset
The ports of the MPFE must connect to the same reset signal.
When the reset signal is asserted, it resets the command and data FIFO buffer in the
MPFE without resetting the hard memory controller.
Note:
The global_reset_n and soft_reset_n signals are asynchronous.
For easiest management of reset signals, Intel recommends the following sequence at
power-up:
1.
Initially global_reset_n, soft_reset_n, and the MPFE reset signals are all
asserted.
2. global_reset_n is deasserted.
3. Wait for pll_locked to transition high.
4.
soft_reset_n is deasserted.
5.
(Optional) If you encounter difficulties, wait for the controller signal
local_cal_success to go high, indicating that the external memory interface
has successfully completed calibration, before deasserting the MPFE FIFO reset
signals. This will ensure that read/write activity cannot occur until the interface is
successfully calibrated.
6.4.3 DRAM Interface
The DRAM interface is 40 bits wide, and can accommodate 8-bit, 16-bit, 16-bit plus
ECC, 32-bit, or 32-bit plus ECC configurations. Any unused I/Os in the DRAM interface
can be reused as user I/Os.
The DRAM interface supports DDR2 and DDR3 memory protocols, and LPDDR2 for
Cyclone V only. Fast and medium speed grade devices are supported to 533 MHz for
Arria V and 400 MHz for Cyclone V.
External Memory Interface Handbook Volume 3: Reference Material
265
6 Functional Description—Hard Memory Interface
6.4.4 ECC
The hard controller supports both error-correcting code (ECC) calculated by the
controller and by the user. Controller ECC code employs standard Hamming logic
which can detect and correct single-bit errors and detect double-bit errors. The
controller ECC is available for 16-bit and 32-bit widths, each requiring an additional 8
bits of memory, resulting in an actual memory width of 24-bits and 40-bits,
respectively.
In user ECC mode, all bits are treated as data bits, and are written to and read from
memory. User ECC can implement nonstandard memory widths such as 24-bit or 40bit, where ECC is not required.
Controller ECC
Controller ECC provides the following features:
Byte Writes—The memory controller performs a read/modify/write operation to keep
ECC valid when a subset of the bits of a word is being written. If an entire word is
being written (but less than a full burst) and the DM pins are connected, no read is
necessary and only that word is updated. If controller ECC is disabled, byte-writes
have no performance impact.
ECC Write Backs—When a read operation detects a correctable error, the memory
location is scheduled for a read/modify/write operation to correct the single-bit error.
User ECC—User ECC is 24-bits or 40-bits wide; with user ECC, the controller performs
no ECC checking. The controller employs memory word addressing with byte enables,
and can handle arbitrary memory widths. User ECC does not disable byte writes;
hence, you must ensure that any byte writes do not result in corrupted ECC.
6.4.5 Bonding of Memory Controllers
Bonding is a feature that allows data to be split between two memory controllers,
providing the ability to service bandwidth streams similar to a single 64-bit controller.
Bonding works by dividing data buses in proportion to the memory widths, and always
sending a transaction to both controllers. When signals are returned, bonding ensures
that both sets of signals are returned identically.
Bonding can be applied to asymetric controllers, and allows controllers to have
different memory clocks. Bonding does not attempt to synchronize the controllers.
Bonding supports only one port. The Avalon port width can be varied from 64-bit to
256-bit; 32-bit port width is not supported.
The following signals require bonding circuitry:
Read data return—This bonding allows read data from the two controllers to return
with effectively one ready signal to the bus master that initiated the bus transaction.
Write ready—For Avalon-MM, this is effectively bonding on the waitrequest signal.
Write acknowledge—Synchronization on returning the write completed signal.
For each of the above implementations, data is returned in order, hence the circuitry
must match up for each valid cycle.
External Memory Interface Handbook Volume 3: Reference Material
266
6 Functional Description—Hard Memory Interface
Bonded FIFO buffers must have identical FIFO numbers; that is, read FIFO 1 on
controller 1 must be paired with Read FIFO 1 on controller 2.
Data Return Bonding
Long loop times can lead to communications problems when using bonded controllers.
The following effects are possible when using bonded controllers:
•
If one memory controller completes its transaction and receives new data before
the other controller, then the second controller can send data as soon as it arrives,
and before the first controller acknowledges that the second controller has data.
•
If the first controller has a single word in its FIFO buffer and the second controller
receives single-word transactions, the second controller must determine whether
the second word is a valid signal or not.
To accommodate the above effects, the hard controller maintains two counters for
each bonded pair of FIFO buffers and implements logic that monitors those counters
to ensure that the bonded controllers receive the same data on the same cycle, and
that they send the data out on the same cycle.
FIFO Ready
FIFO ready bonding is used for write command and write data buses. The
implementation is similar to the data return bonding.
Bonding Latency Impact
Bonding has no latency impact on ports that are not bonded.
Bonding Controller Usage
Arria V and Cyclone V devices employ three shared bonding controllers to manage the
read data return bonding, write acknowledge bonding, and command/write data ready
bonding.
The three bonding controllers require three pairs of bonding I/Os, each based on a six
port count; this means that a bonded hard memory controller requires 21 input
signals and 21 output signals for its connection to the fabric, and another 21 input
signals and 21 output signals to the paired hard memory controller.
Note:
The hard processor system (HPS) hard memory controller cannot be bonded with
another hard memory controller on the FPGA portion of the device.
Bonding Configurations and Parameter Requirements
Intel has verified hard memory controller bonding between two interfaces with the
following configuration:
•
Same clock source
•
Same memory clock frequency
•
Same memory parameters and timings (except interface width)
•
Same controller settings.
•
Same port width in MPFE settings
Bonding supports only one port. The Avalon port width can be varied from 64-bits to
256-bits; a 32-bit port width is not supported.
External Memory Interface Handbook Volume 3: Reference Material
267
6 Functional Description—Hard Memory Interface
6.5 Hard PHY
A physical layer interface (PHY) is embedded in the periphery of the Arria V device,
and can run at the same high speed as the hard controller and hard sequencer. This
hard PHY is located next to the hard controller. Differing device configurations have
different numbers and sizes of hard controller and hard PHY pairs.
The hard PHY implements logic that connects the hard controller to the I/O ports.
Because the hard controller and AFI interface support high frequencies, a portion of
the sequencer is implemented as hard logic. The Nios II processor, the instruction/
data RAM, and the Avalon fabric of the sequencer are implemented as core soft logic.
The read/write manger and PHY manager components of the sequencer, which must
operate at full rate, are implemented as hard logic in the hard PHY.
6.5.1 Interconnections
The hard PHY resides on the device between the hard controller and the I/O register
blocks. The hard PHY is instantiated or bypassed entirely, depending on the
parameterization that you specify.
The hard PHY connects to the hard memory controller and the core, enabling the use
of either the hard memory controller or a software-based controller. (You can have the
hard controller and hard PHY, or the soft controller and soft PHY; however, the
combination of soft controller with hard PHY is not supported.) The hard PHY also
connects to the I/O register blocks and the DQS logic. The path between the hard PHY
and the I/O register blocks can be bypassed, but not reconfigured—in other words, if
you use the hard PHY datapath, the pins to which it connects are predefined and
specified by the device pin table.
6.5.2 Clock Domains
The hard PHY contains circuitry that uses the following clock domains:
AFI clock domain (pll_afi_clk) —The main full-rate clock signal that synchronizes
most of the circuit logic.
Avalon clock domain (pll_avl_clk) —Synchronizes data on the internal Avalon bus,
namely the Read/Write Manager, PHY Manager, and Data Manager data. The data is
then transferred to the AFI clock domain. To ensure reliable data transfer between
clock domains, the Avalon clock period must be an integer multiple of the AFI clock
period, and the phases of the two clocks must be aligned.
Address and Command clock domain (pll_addr_cmd_clk) —Synchronizes the
global asychronous reset signal, used by the I/Os in this clock domain.
6.5.3 Hard Sequencer
The sequencer initializes the memory device and calibrates the I/Os, with the
objective of maximizing timing margins and achieving the highest possible
performance.
External Memory Interface Handbook Volume 3: Reference Material
268
6 Functional Description—Hard Memory Interface
When the hard memory controller is in use, a portion of the sequncer must run at full
rate; for this reason, the Read/Write Manager, PHY Manager, and Data Manager are
implemented as hard components within the hard PHY. The hard sequencer
communicates with the soft-logic sequencer components (including the Nios II
processor) via an Avalon bus.
6.5.4 MPFE Setup Guidelines
The following instructions provide information on configuring the multi-port front end
of the hard memory interface.
1. To enable the hard memory interface, turn on Enable Hard External Memory
Interface in the Interface Type tab in the parameter editor.
2. To export bonding interface ports to the top level, turn on Export bonding
interface in the Multiple Port Front End pulldown on the Controller Settings
tab in the parameter editor.
Note: The system exports three bonding-in ports and three bonding-out ports. You
must generate two controllers and connect the bonding ports manually.
3. To expand the interface data width from a maximum of 32 bits to a maximum of
40 bits, turn on Enable Avalon-MM data for ECC in the Multiple Port Front
End pulldown on the Controller Settings tab in the parameter editor.
Note: The controller does not perform ECC checking when this option is turned on.
4. Select the required Number of ports for the multi-port front end in the Multiple
Port Front End pulldown on the Controller Settings tab in the parameter editor.
Note: The maximum number of ports is 6, depending on the port type and width.
The maximum port width is 256 bits, which is the maximum data width of
the read data FIFO and write data FIFO buffers.
5. The table in the Multiple Port Front End pulldown on the Controller Settings
tab in the parameter editor lists the ports that are created. The columns in the
table describe each port, as follows:
•
Port: Indicates the port number.
•
Type: Indicates whether the port is read only, write only, or bidirectional.
•
Width: To achieve optimum MPFE throughput, Intel recommends setting the
MPFE data port width according to the following calculation:
2 x (frequency ratio of HMC to user logic) x (interface
data width)
For example, if the frequency of your user logic is one-half the frequency of
the hard memory controller, you should set the port width to be 4x the
interface data width. If the frequency ratio of the hard memory controller to
user logic is a fractional value, you should use a larger value; for example, if
the ratio is 1.5, you can use 2.
External Memory Interface Handbook Volume 3: Reference Material
269
6 Functional Description—Hard Memory Interface
•
Priority: The priority setting specifies the priority of the slave port, with
higher values representing higher priority. The slave port with highest priority
is served first.
•
Weight: The weight setting has a range of values of 0–31, and specifies the
relative priority of a slave port, with higher weight representing higher priority.
The weight value can determine relative bandwidth allocations for slave ports
with the same priority values.For example, if two ports have the same priority
value, and weight values of 4 and 6, respectively, the port with a weight of 4
will receive 40% of the bus bandwidth, while the port with a weight of 6 will
receive 60% of the bus bandwidth—assuming 100% total available bus
bandwidth.
6.5.5 Soft Memory Interface to Hard Memory Interface Migration
Guidelines
The following instructions provide information on mapping your soft memory interface
to a hard memory interface.
Pin Connections
1.
The hard and soft memory controllers have compatible pinouts. Assign interface
pins to the hard memory interface according to the pin table.
2.
Ensure that your soft memory interface pins can fit into the hard memory
interface. The hard memory interface can support a maximum of a 40-bit interface
with user ECC, or a maximum of 80-bits with same-side bonding. The soft
memory interface does not support bonding.
3.
Follow the recommended board layout guidelines for the hard memory interface.
Software Interface Preparation
Observe the following points in preparing your soft memory interface for migration to
a hard memory interface:
•
You cannot use the hard PHY without also using the hard memory controller.
•
The hard memory interface supports only full-rate controller mode.
•
Ensure that the MPFE data port width is set according to the soft memory interface
half-rate mode Avalon data width.
•
The hard memory interface uses a different Avalon port signal naming convention
than the software memory interface. Ensure that you change the
avl_<signal_name> signals in the soft memory interface
to .avl_<signal_name>_0 signals in the hard memory interface.
•
The hard memory controller MPFE includes an additional three clocks and three
reset ports (CMD port, RFIFO port, and WFIFO port) that do not exist with the soft
memory controller. You should connect the user logic clock signal to the MPFE
clock port, and the user logic reset signal to the MPFE reset port.
•
In the soft memory interface, the half-rate afi_clk is a user logic clock. In the
hard memory interface, afi_clk is a full-rate clock, because the core fabric
might not be able to achieve full-rate speed. When you migrate your soft memory
interface to a hard memory interface, you need to supply an additional slower rate
clock. The maximum clock rate supported by core logic is one-half of the
maximum interface frequency.
Latency
External Memory Interface Handbook Volume 3: Reference Material
270
6 Functional Description—Hard Memory Interface
Overall, you should expect to see slightly more latency when using the hard memory
controller and multi-port front end, than when using the soft memory controller.
The hard memory controller typically exhibits lower latency than the soft memory
controller; however, the multi-port front end does introduce additional latency cycles
due to FIFO buffer stages used for synchronization. The MPFE cannot be bypassed,
even if only one port is needed.
6.5.6 Bonding Interface Guidelines
Bonding allows a single data stream to be split between two memory controllers,
providing the ability to expand the interface data width similar to a single 64-bit
controller. This section provides some guidelines for setting up the bonding interface.
1.
Bonding interface ports are exported to the top level in your design. You should
connect each bonding_in* port in one hard memory controller to the
corresponding bonding_out_*port in the other hard memory controller, and vice
versa.
2. You should modify the Avalon signal connections to drive the bonding interface
with a single user logic/master, as follows:
a.
AND both avl_ready signals from both hard memory controllers before the
signals enter the user logic.
b.
AND both avl_rdata_valid signals from both hard memory controllers
before the signals enter the user logic. (The avl_rdata_valid signals
should be identical for both hard memory controllers.)
c.
Branch the following signals from the user logic to both hard memory
controllers:
d.
•
avl_burstbegin
•
avl_addr
•
avl_read_req
•
avl_write_req
•
avl_size
Split the following signals according to each multi-port front end data port
width:
•
avl_rdata
•
avl_wdata
•
avl_be
6.6 Document Revision History
Date
Version
Changes
May 2017
2017.05.08
Rebranded as Intel.
October 2016
2016.10.31
Maintenance release.
May 2016
2016.05.02
Maintenance release.
November 2015
2015.11.02
Maintenance release.
continued...
External Memory Interface Handbook Volume 3: Reference Material
271
6 Functional Description—Hard Memory Interface
Date
May 2015
December 2014
August 2014
Version
Changes
2015.05.04
Maintenance release.
21014.12.15
Maintenance release.
2014.08.15
•
Updated descriptions of mp_cmd_reset_n_#_reset_n,
mp_rfifo_reset_n_#_reset_n, and mp_wfifo_reset_n_#_reset_n
•
in the MPFE Signals table.
Added Reset section to Hard Memory Controller section.
December 2013
2013.12.16
•
•
•
•
Added footnote about command FIFOs to MPFE Signals table.
Added information about FIFO depth for the MPFE.
Added information about hard memory controller bonding.
Reworded protocol-support information for Arria V and Cyclone V devices.
November 2012
2.1
•
•
Added Hard Memory Interface Implementation Guidelines.
Moved content of EMI-Related HPS Features in SoC Devices section to
chapter 4, Functional Description—HPS Memory Controller.
June 2012
2.0
•
•
•
Added EMI-Related HPS Features in SoC Devices.
Added LPDDR2 support.
Added Feedback icon.
November 2011
1.0
Initial release.
External Memory Interface Handbook Volume 3: Reference Material
272
7 Functional Description—HPS Memory Controller
7 Functional Description—HPS Memory Controller
The hard processor system (HPS) SDRAM controller subsystem provides efficient
access to external SDRAM for the ARM* Cortex*-A9 microprocessor unit (MPU)
subsystem, the level 3 (L3) interconnect, and the FPGA fabric.
Note:
This chapter applies to the HPS architecture of Arria V and Cyclone V memory
controllers only.
The SDRAM controller provides an interface between the FPGA fabric and HPS. The
interface accepts Advanced Microcontroller Bus Architecture (AMBA®) Advanced
eXtensible Interface (AXI™) and Avalon® Memory-Mapped (Avalon-MM) transactions,
converts those commands to the correct commands for the SDRAM, and manages the
details of the SDRAM access.
7.1 Features of the SDRAM Controller Subsystem
The SDRAM controller subsystem offers programming flexibility, port and bus
configurability, error correction, and power management for external memories up to
4 GB.
•
Support for double data rate 2 (DDR2), DDR3, and low-power DDR2 (LPDDR2)
SDRAM
•
Flexible row and column addressing with the ability to support up to 4 GB of
memory in various interface configurations
•
Optional 8-bit integrated error correction code (ECC) for 16- and 32-bit data
widths3
•
User-configurable memory width of 8, 16, 16+ECC, 32, 32+ECC
•
User-configurable timing parameters
•
Two chip selects (DDR2 and DDR3)
•
Command reordering (look-ahead bank management)
•
Data reordering (out of order transactions)
•
User-controllable bank policy on a per port basis for either closed page or
conditional open page accesses
•
User-configurable priority support with both absolute and weighted round-robin
scheduling
•
Flexible FPGA fabric interface with up to 6 ports that can be combined for a data
width up to 256 bits using Avalon-MM and AXI interfaces
•
Power management supporting self refresh, partial array self refresh (PASR),
power down, and LPDDR2 deep power down
3 The level of ECC support is package dependent.
Intel Corporation. All rights reserved. Intel, the Intel logo, Altera, Arria, Cyclone, Enpirion, MAX, Nios, Quartus
and Stratix words and logos are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other
countries. Intel warrants performance of its FPGA and semiconductor products to current specifications in
accordance with Intel's standard warranty, but reserves the right to make changes to any products and services
at any time without notice. Intel assumes no responsibility or liability arising out of the application or use of any
information, product, or service described herein except as expressly agreed to in writing by Intel. Intel
customers are advised to obtain the latest version of device specifications before relying on any published
information and before placing orders for products or services.
*Other names and brands may be claimed as the property of others.
ISO
9001:2008
Registered
7 Functional Description—HPS Memory Controller
7.2 SDRAM Controller Subsystem Block Diagram
The SDRAM controller subsystem connects to the MPU subsystem, the L3
interconnect, and the FPGA fabric. The memory interface consists of the SDRAM
controller, the physical layer (PHY), control and status registers (CSRs), and their
associated interfaces.
Figure 146. SDRAM Controller Subsystem High-Level Block Diagram
SDRAM Controller Subsystem
SDRAM Controller
MPU
Subsystem
L3
Interconnect
FPGA
Fabric
64-Bit AXI
32-Bit AXI
Multi-Port
Front End
Single-Port
Controller
Altera
PHY
Interface
DDR
PHY
HPS
I/O
Pins
External
Memory
FPGA-to-HPS
SDRAM Interface
32- to 256-Bit
AXI or
Avalon-MM
Control & Status Registers
Register Slave Interface
L4 Peripheral Bus (l4_sp_clk)
SDRAM Controller
The SDRAM controller provides high performance data access and run-time
programmability. The controller reorders data to reduce row conflicts and bus turnaround time by grouping read and write transactions together, allowing for efficient
traffic patterns and reduced latency.
The SDRAM controller consists of a multiport front end (MPFE) and a single-port
controller. The MPFE provides multiple independent interfaces to the single-port
controller. The single-port controller communicates with and manages each external
memory device.
The MPFE FPGA-to-HPS SDRAM interface port has an asynchronous FIFO buffer
followed by a synchronous FIFO buffer. Both the asynchronous and synchronous FIFO
buffers have a read and write data FIFO depth of 8, and a command FIFO depth of 4.
The MPU subsystem 64-bit AXI port and L3 interconnect 32-bit AXI port have
asynchronous FIFO buffers with read and write data FIFO depth of 8, and command
FIFO depth of 4.
External Memory Interface Handbook Volume 3: Reference Material
274
7 Functional Description—HPS Memory Controller
DDR PHY
The DDR PHY provides a physical layer interface for read and write memory operations
between the memory controller and memory devices. The DDR PHY has dataflow
components, control components, and calibration logic that handle the calibration for
the SDRAM interface timing.
Related Links
Memory Controller Architecture on page 277
The SDRAM controller consists of an MPFE, a single-port controller, and an interface
to the CSRs.
7.3 SDRAM Controller Memory Options
Bank selects, and row and column address lines can be configured to work with
SDRAMs of various technology and density combinations.
Table 64.
SDRAM Controller Interface Memory Options
Memory Type
Mbits
Column
Address Bit
Width
Bank Select
Bit Width
Row Address
Bit Width
DDR2
256
10
2
13
1024
32
512
10
2
14
1024
64
1024 (1 Gb)
10
3
14
1024
128
2048 (2 Gb)
10
3
15
1024
256
4096 (4 Gb)
10
3
16
1024
512
512
10
3
13
1024
64
1024 (1 Gb)
10
3
14
1024
128
2048 (2 Gb)
10
3
15
1024
256
4096 (4 Gb)
10
3
16
1024
512
64
9
2
12
512
128
10
2
12
1024
16
256
10
2
13
1024
32
11
2
13
2048
64
11
2
14
2048
128
1024 (1 Gb) -S4
11
3
13
2048
128
2048
11
2
15
2048
4
DDR3
LPDDR2
512
1024 (1 Gb)
-S25
6
Page Size
MBytes
8
256
continued...
4 For all memory types shown in this table, the DQ width is 8.
5 S2 signifies a 2n prefetch size
6 S4 signifies a 4n prefetch size
External Memory Interface Handbook Volume 3: Reference Material
275
7 Functional Description—HPS Memory Controller
Memory Type
4
Mbits
(2 Gb) - S2
2048
(2 Gb) -S4
4096
(4 Gb)
Column
Address Bit
Width
Bank Select
Bit Width
Row Address
Bit Width
Page Size
MBytes
11
3
14
2048
256
12
3
14
4096
512
5
6
7.4 SDRAM Controller Subsystem Interfaces
7.4.1 MPU Subsystem Interface
The SDRAM controller connects to the MPU subsystem with a dedicated 64-bit AXI
interface, operating on the mpu_l2_ram_clk clock domain.
7.4.2 L3 Interconnect Interface
The SDRAM controller interfaces to the L3 interconnect with a dedicated 32-bit AXI
interface, operating on the l3_main_clk clock domain.
7.4.3 CSR Interface
The CSR interface connects to the level 4 (L4) bus and operates on the l4_sp_clk
clock domain. The MPU subsystem uses the CSR interface to configure the controller
and PHY, for example setting the memory timing parameter values or placing the
memory in a low power state. The CSR interface also provides access to the status
registers in the controller and PHY.
7.4.4 FPGA-to-HPS SDRAM Interface
The FPGA-to-HPS SDRAM interface provides masters implemented in the FPGA fabric
access to the SDRAM controller subsystem in the HPS. The interface has three port
types that are used to construct the following AXI or Avalon-MM interfaces:
•
Command ports—issue read and/or write commands, and for receive write
acknowledge responses
•
64-bit read data ports—receive data returned from a memory read
•
64-bit write data ports—transmit write data
The FPGA-to-HPS SDRAM interface supports six command ports, allowing up to six
Avalon-MM interfaces or three AXI interfaces. Each command port can be used to
implement either a read or write command port for AXI, or be used as part of an
Avalon-MM interface. The AXI and Avalon-MM interfaces can be configured to support
32-, 64-, 128-, and 256-bit data.
4 For all memory types shown in this table, the DQ width is 8.
External Memory Interface Handbook Volume 3: Reference Material
276
7 Functional Description—HPS Memory Controller
Table 65.
FPGA-to-HPS SDRAM Controller Port Types
Port Type
Available Number of Ports
Command
6
64-bit read data
4
64-bit write data
4
The FPGA-to-HPS SDRAM controller interface can be configured with the following
characteristics:
•
Avalon-MM interfaces and AXI interfaces can be mixed and matched as required
by the fabric logic, within the bounds of the number of ports provided to the
fabric.
•
Because the AXI protocol allows simultaneous read and write commands to be
issued, two SDRAM control ports are required to form an AXI interface.
•
Because the data ports are natively 64-bit, they must be combined if wider data
paths are required for the interface.
•
Each Avalon-MM or AXI interface of the FPGA-to-HPS SDRAM interface operates on
an independent clock domain.
•
The FPGA-to-HPS SDRAM interfaces are configured during FPGA configuration.
The following table shows the number of ports needed to configure different bus
protocols, based on type and data width.
Table 66.
FPGA-to-HPS SDRAM Port Utilization
Bus Protocol
Command Ports
Read Data Ports
Write Data Ports
32- or 64-bit AXI
2
1
1
128-bit AXI
2
2
2
256-bit AXI
2
4
4
32- or 64-bit Avalon-MM
1
1
1
128-bit Avalon-MM
1
2
2
256-bit Avalon-MM
1
4
4
32- or 64-bit Avalon-MM write-only
1
0
1
128-bit Avalon-MM write-only
1
0
2
256-bit Avalon-MM write-only
1
0
4
32- or 64-bit Avalon-MM read-only
1
1
0
128-bit Avalon-MM read-only
1
2
0
256-bit Avalon-MM read-only
1
4
0
7.5 Memory Controller Architecture
The SDRAM controller consists of an MPFE, a single-port controller, and an interface to
the CSRs.
External Memory Interface Handbook Volume 3: Reference Material
277
7 Functional Description—HPS Memory Controller
Figure 147. SDRAM Controller Block Diagram
SDRAM Controller
Multi-Port Front End
Single-Port Controller
Read Data
6
Data
FIFO
Buffers
Reorder
Buffer
ECC
Generation
&
Checking
Write Data
6
Data
FIFO
Buffers
FPGA
Fabric
FIFO
POP
Logic
Write Data
Buffer
Altera
PHY
Interface
Command
6 Write
WR Acknowledge
Acknowledge Queues
10
Command
FIFO
Buffers
Rank Timer
Scheduler
Command
Generator
Timer
Bank
Pool
Arbiter
Control & Status Register Interface
7.5.1 Multi-Port Front End
The Multi-Port Front End (MPFE) is responsible for scheduling pending transactions
from the configured interfaces and sending the scheduled memory transactions to the
single-port controller. The MPFE handles all functions related to individual ports.
The MPFE consists of three primary sub-blocks.
Command Block
The command block accepts read and write transactions from the FPGA fabric and the
HPS. When the command FIFO buffer is full, the command block applies backpressure
by deasserting the ready signal. For each pending transaction, the command block
calculates the next SDRAM burst needed to progress on that transaction. The
command block schedules pending SDRAM burst commands based on the
user-supplied configuration, available write data, and unallocated read data space.
Write Data Block
The write data block transmits data to the single-port controller. The write data block
maintains write data FIFO buffers and clock boundary crossing for the write data. The
write data block informs the command block of the amount of pending write data for
each transaction so that the command block can calculate eligibility for the next
SDRAM write burst.
External Memory Interface Handbook Volume 3: Reference Material
278
7 Functional Description—HPS Memory Controller
Read Data Block
The read data block receives data from the single-port controller. Depending on the
port state, the read data block either buffers the data in its internal buffer or passes
the data straight to the clock boundary crossing FIFO buffer. The read data block
reorders out-of-order data for Avalon-MM ports.
In order to prevent the read FIFO buffer from overflowing, the read data block informs
the command block of the available buffer area so the command block can pace read
transaction dispatch.
7.5.2 Single-Port Controller
The single-port logic is responsible for following actions:
•
Queuing the pending SDRAM bursts
•
Choosing the most efficient burst to send next
•
Keeping the SDRAM pipeline full
•
Ensuring all SDRAM timing parameters are met
Transactions passed to the single-port logic for a single page in SDRAM are
guaranteed to be executed in order, but transactions can be reordered between pages.
Each SDRAM burst read or write is converted to the appropriate Altera PHY interface
(AFI) command to open a bank on the correct row for the transaction (if required),
execute the read or write command, and precharge the bank (if required).
The single-port logic implements command reordering (looking ahead at the command
sequence to see which banks can be put into the correct state to allow a read or write
command to be executed) and data reordering (allowing data transactions to be
dispatched even if the data transactions are executed in an order different than they
were received from the multi-port logic).
The single-port controller consists of eight sub-modules.
7.5.2.1 Command Generator
The command generator accepts commands from the MPFE and from the internal ECC
logic, and provides those commands to the timer bank pool.
Related Links
Memory Controller Architecture on page 277
For more information, refer to the SDRAM Controller Block diagram.
7.5.2.2 Timer Bank Pool
The timer bank pool is a parallel queue that operates with the arbiter to enable data
reordering. The timer bank pool tracks incoming requests, ensures that all timing
requirements are met, and, on receiving write-data-ready notifications from the write
data buffer, passes the requests to the arbiter
Related Links
Memory Controller Architecture on page 277
For more information, refer to the SDRAM Controller Block diagram.
External Memory Interface Handbook Volume 3: Reference Material
279
7 Functional Description—HPS Memory Controller
7.5.2.3 Arbiter
The arbiter determines the order in which requests are passed to the memory device.
When the arbiter receives a single request, that request is passed immediately. When
multiple requests are received, the arbiter uses arbitration rules to determine the
order to pass requests to the memory device.
Related Links
Memory Controller Architecture on page 277
For more information, refer to the SDRAM Controller Block diagram.
7.5.2.4 Rank Timer
The rank timer performs the following functions:
•
Maintains rank-specific timing information
•
Ensures that only four activates occur within a specified timing window
•
Manages the read-to-write and write-to-read bus turnaround time
•
Manages the time-to-activate delay between different banks
Related Links
Memory Controller Architecture on page 277
For more information, refer to the SDRAM Controller Block diagram.
7.5.2.5 Write Data Buffer
The write data buffer receives write data from the MPFE and passes the data to the
PHY, on approval of the write request.
Related Links
Memory Controller Architecture on page 277
For more information, refer to the SDRAM Controller Block diagram.
7.5.2.6 ECC Block
The ECC block consists of an encoder and a decoder-corrector, which can detect and
correct single-bit errors, and detect double-bit errors. The ECC block can correct
single- bit errors and detect double-bit errors resulting from noise or other
impairments during data transmission.
Note:
The level of ECC support is package dependent.
Related Links
Memory Controller Architecture on page 277
For more information, refer to the SDRAM Controller Block diagram.
7.5.2.7 AFI Interface
The AFI interface provides communication between the controller and the PHY.
Related Links
Memory Controller Architecture on page 277
External Memory Interface Handbook Volume 3: Reference Material
280
7 Functional Description—HPS Memory Controller
For more information, refer to the SDRAM Controller Block diagram.
7.5.2.8 CSR Interface
The CSR interface is accessible from the L4 bus. The interface allows code in the HPS
MPU or soft IP cores in the FPGA fabric to configure and monitor the SDRAM controller.
Related Links
Memory Controller Architecture on page 277
For more information, refer to the SDRAM Controller Block diagram.
7.6 Functional Description of the SDRAM Controller Subsystem
7.6.1 MPFE Operation Ordering
Operation ordering is defined and enforced within a port, but not between ports. All
transactions received on a single port for overlapping addresses execute in order.
Requests arriving at different ports have no guaranteed order of service, except when
a first transaction has completed before the second arrives.
Avalon-MM does not support write acknowledgement. When a port is configured to
support Avalon-MM, you should read from the location that was previously written to
ensure that the write operation has completed. When a port is configured to support
AXI, the master accessing the port can safely issue a read operation to the same
address as a write operation as soon as the write has been acknowledged. To keep
write latency low, writes are acknowledged as soon as the transaction order is
guaranteed—meaning that any operations received on any port to the same address
as the write operation are executed after the write operation.
To reduce read latency, the single-port logic can return read data out of order to the
multi-port logic. The returned data is rearranged to its initial order on a per port basis
by the multi-port logic and no traffic reordering occurs between individual ports.
Read Data Handling
The MPFE contains a read buffer shared by all ports. If a port is capable of receiving
returned data then the read buffer is bypassed. If the size of a read transaction is
smaller than twice the memory interface width, the buffer RAM cannot be bypassed.
The lowest memory latency occurs when the port is ready to receive data and the full
width of the interface is utilized.
7.6.2 MPFE Multi-Port Arbitration
The HPS SDRAM controller multi-port front end (MPFE) contains a programmable
arbiter. The arbiter decides which MPFE port gains access to the single-port memory
controller.
The SDRAM transaction size that is arbitrated is a burst of two beats. This burst size
ensures that the arbiter does not favor one port over another when the incoming
transaction size is a large burst.
The arbiter makes decisions based on two criteria: priority and weight. The priority is
an absolute arbitration priority where the higher priority ports always win arbitration
over the lower priority ports. Because multiple ports can be set to the same priority,
the weight value refines the port choice by implementing a round-robin arbitration
External Memory Interface Handbook Volume 3: Reference Material
281
7 Functional Description—HPS Memory Controller
among ports set to the same priority. This programmable weight allows you to assign
a higher arbitration value to a port in comparison to others such that the highest
weighted port receives more transaction bandwidth than the lower weighted ports of
the same priority.
Before arbitration is performed, the MPFE buffers are checked for any incoming
transactions. The priority of each port that has buffered transactions is compared and
the highest priority wins. If multiple ports are of the same highest priority value, the
port weight is applied to determine which port wins. Because the arbiter only allows
SDRAM-sized bursts into the single-port memory controller, large transactions may
need to be serviced multiple times before the read or write command is fully accepted
to the single-port memory controller. The MPFE supports dynamic tuning of the priority
and weight settings for each port, with changes committed into the SDRAM controller
at fixed intervals of time.
Arbitration settings are applied to each port of the MPFE. The memory controller
supports a mix of Avalon-MM and AXI protocols. As defined in the "Port Mappings"
section, the Avalon-MM ports consume a single command port while the AXI ports
consume a pair of command ports to support simultaneous read and write
transactions. In total, there are ten command ports for the MPFE to arbitrate. The
following table illustrates the command port mapping within the HPS as well as the
ports exposed to the FPGA fabric.
Table 67.
HPS SDRAM MPFE Command Port Mapping
Command Port
Allowed Functions
0, 2, 4
FPGA fabric AXI read command ports
FPGA fabric Avalon-MM read or write
command ports
1, 3, 5
FPGA fabric AXI write command ports
FPGA fabric Avalon-MM read or write
command ports
Data Size
32-bit to 256-bit data
6
L3 AXI read command port
32-bit data
7
MPU AXI read command port
64-bit data
8
L3 AXI write command port
32-bit data
9
MPU AXI write command port
64-bit data
When the FPGA ports are configured for AXI, the command ports are always assigned
in groups of two starting with even number ports 0, 2, or 4 assigned to the read
command channel. For example, if you configure the first FPGA-to-SDRAM port as AXI
and the second port as Avalon-MM, you can expect the following mapping:
•
Command port 0 = AXI read
•
Command port 1 = AXI write
•
Command port 2 = Avalon-MM read and write
Setting the MPFE Priority
The priority of each of the ten command ports is configured through the
userpriority field of the mppriority register. This 30-bit register uses 3 bits per
port to configure the priority. The lowest priority is 0x0 and the highest priority is 0x7.
The bits are mapped in ascending order with bits [2:0] assigned to command port 0
and bits [29:27] assigned to command port 9.
External Memory Interface Handbook Volume 3: Reference Material
282
7 Functional Description—HPS Memory Controller
Setting the MPFE Static Weights
The static weight settings used in the round-robin command port priority scheme are
programmed in a 128-bit field distributed among four 32-bit registers:
•
mpweight_0_4
•
mpweight_1_4
•
mpweight_2_4
•
mpweight_3_4
Each port is assigned a 5-bit value within the 128-bit field, such that port 0 is
assigned to bits [4:0] of the mpweight_0_4 register, port 1 is assigned to bits [9:5]
of the mpweight_0_4 register up to port 9, which is assigned to bits[49:45]
contained in the mpweight_1_4 register. The valid weight range for each port is 0x0
to 0x1F, with larger static weights representing a larger arbitration share.
Bits[113:50] in the mpweight_1_4, mpweight_2_4 and mpweight_3_4 registers,
hold the sum of weights for each priority. This 64-bit field is divided into eight fields of
8-bits, each representing the sum of static weights. Bits[113:50] are mapped in
ascending order with bits [57:50] holding the sum of static weights for all ports with
priority setting 0x0, and bits [113:106] holding the sum of static weights for all ports
with priority setting 0x7.
Example Using MPFE Priority and Weights
In this example, the following settings apply:
•
FPGA MPFE ports 0 is assigned to AXI read commands and port 1 is assigned to
AXI write commands.
•
FPGA MPFE port 2 is assigned to Avalon-MM read and write commands.
•
L3 master ports (ports 6 and 8) and the MPU ports (port 7 and 9) are given the
lowest priority but the MPU ports are configured with more arbitration static
weight than the L3 master ports.
•
The FPGA MPFE command ports are given the highest priority; however, AXI ports
0 and 1 are given a larger static weight because they carry the highest priority
traffic in the entire system. Assigning a high priority and larger static weight
ensures ports 0 and 1 will receive the highest quality-of-service (QoS).
The table below details the port weights and sum of weights.
Table 68.
SDRAM MPFE Port Priority, Weights and Sum of Weights
Priority
Weights
Sum
of
Wei
ghts
-
Port
0
Port
1
Port
2
Port
3
Port
4
Port
5
Port
6
Port
7
Port
8
Port
9
-
1
10
10
5
0
0
0
0
0
0
0
25
0
0
0
0
0
0
0
1
4
1
4
10
If the FPGA-to-SDRAM ports are configured according to the table and if both ports are
accessed concurrently, you can expect the AXI port to receive 80% of the total
service. This value is determined by taking the sum of port 0 and 1 weights divided by
External Memory Interface Handbook Volume 3: Reference Material
283
7 Functional Description—HPS Memory Controller
the total weight for all ports of priority 1. The remaining 20% of bandwidth is allocated
to the Avalon-MM port. With these port settings, any FPGA transaction buffered by the
MPFE for either slave port blocks the MPU and L3 masters from having their buffered
transactions serviced. To avoid transaction starvation, you should assign ports the
same priority level and adjust the bandwidth allocated to each port by adjusting the
static weights.
MPFE Weight Calculation
The MPFE applies a deficit round-robin arbitration scheme to determine which port is
serviced. The larger the port weight, the more often it is serviced. Ports are serviced
only when they have buffered transactions and are set to the highest priority of all the
ports that also have buffered transactions. The arbiter determines which port to
service by examining the running weight of all the same ports at the same priority
level and the largest running weight is serviced.
Each time a port is drained of all transactions, its running weight is set to 0x80. Each
time a port is serviced, the static weight is added and the sum of weights is subtracted
from the running weight for that port. Each time a port is not serviced (same priority
as another port but has a lower running weight), the static weight for the port is
added to the running weight of the port for that particular priority level. The running
weight additions and subtractions are only applied to one priority level, so any time
ports of a different priority level are being serviced, the running weight of a lower
priority port is not modified.
MPFE Multi-Port Arbitration Considerations for Use
When using MPFE multi-port arbitration, the following considerations apply:
•
To ensure that the dynamic weight value does not roll over when a port is
serviced, the following equation should always be true:
(sum of weights - static weight) < 128
If the running weight remains less than 128, arbitration for that port remains
functional.
•
The memory controller commits the priority and weight registers into the MPFE
arbiter once every 10 SDRAM clock cycles. As a result, when the mppriority
register and mpweight_*_4 registers are configured at run time, the update
interval can occur while the registers are still being written and ports can have
different priority or weights than intended for a brief period. Because the
mppriority and mpweight_*_4 registers can be updated in a single 32-bit
transaction, Intel recommends updating first to ensure that transactions that need
to be serviced have the appropriate priority after the next update. Because the
static weights are divided among four 32-bit registers
•
In addition to the mppriority register and mpweight_*_* registers, the
remappriority register adds another level of priority to the port scheduling. By
programming bit N in the priorityremap field of the remappriority register,
any port with an absolute priority N is sent to the front of the single port
command queue and is serviced before any other transaction. Please refer to the
remappriority register for more details.
External Memory Interface Handbook Volume 3: Reference Material
284
7 Functional Description—HPS Memory Controller
The scheduler is work-conserving. Write operations can only be scheduled when
enough data for the SDRAM burst has been received. Read operations can only be
scheduled when sufficient internal memory is free and the port is not occupying too
much of the read buffer.
7.6.3 MPFE SDRAM Burst Scheduling
SDRAM burst scheduling recognizes addresses that access the same row/bank
combination, known as open page accesses. Operations to a page are served in the
order in which they are received by the single-port controller. Selection of SDRAM
operations is a two-stage process. First, each pending transaction must wait for its
timers to be eligible for execution. Next, the transaction arbitrates against other
transactions that are also eligible for execution.
The following rules govern transaction arbitration:
•
High-priority operations take precedence over lower-priority operations
•
If multiple operations are in arbitration, read operations have precedence over
write operations
•
If multiple operations still exist, the oldest is served first
A high-priority transaction in the SDRAM burst scheduler wins arbitration for that bank
immediately if the bank is idle and the high-priority transaction's chip select, row, or
column fields of the address do not match an address already in the single-port
controller. If the bank is not idle, other operations to that bank yield until the highpriority operation is finished. If the chip select, row, and column fields match an
earlier transaction, the high-priority transaction yields until the earlier transaction is
completed.
Clocking
The FPGA fabric ports of the MPFE can be clocked at different frequencies.
Synchronization is maintained by clock-domain crossing logic in the MPFE. Command
ports can operate on different clock domains, but the data ports associated with a
given command port must be attached to the same clock as that command port.
Note:
A command port paired with a read and write port to form an Avalon-MM interface
must operate at the same clock frequency as the data ports associated with it.
7.6.4 Single-Port Controller Operation
The single-port controller increases the performance of memory transactions through
command and data re-ordering, enforcing bank policies, combining write operations
and allowing burst transfers. Correction of single-bit errors and detection of double-bit
errors is handled in the ECC module of the single-port Controller.
SDRAM Interface
The SDRAM interface is up to 40 bits wide and can accommodate 8-bit, 16-bit, 16-bit
plus ECC, 32-bit, or 32-bit plus ECC configurations, depending on the device package.
The SDRAM interface supports LPDDR2, DDR2, and DDR3 memory protocols.
External Memory Interface Handbook Volume 3: Reference Material
285
7 Functional Description—HPS Memory Controller
7.6.4.1 Command and Data Reordering
The heart of the SDRAM controller is a command and data reordering engine.
Command reordering allows banks for future transactions to be opened before the
current transaction finishes.
Data reordering allows transactions to be serviced in a different order than they were
received when that new order allows for improved utilization of the SDRAM bandwidth.
Operations to the same bank and row are performed in order to ensure that
operations which impact the same address preserve the data integrity.
The following figure shows the relative timing for a write/read/write/read command
sequence performed in order and then the same command sequence performed with
data reordering. Data reordering allows the write and read operations to occur in
bursts, without bus turnaround timing delay or bank reassignment.
Figure 148. Data Reordering Effect
Data Reordering Off
Command
Address
WR
B0R0
RD
B1R0
WR
B0R0
RD
B1R0
Data Reordering On
Command
Address
WR
B0R0
WR
B0R0
RD
B1R0
RD
B1R0
The SDRAM controller schedules among all pending row and column commands every
clock cycle.
7.6.4.2 Bank Policy
The bank policy of the SDRAM controller allows users to request that a transaction's
bank remain open after an operation has finished so that future accesses do not delay
in activating the same bank and row combination. The controller supports only eight
simultaneously-opened banks, so an open bank might get closed if the bank resource
is needed for other operations.
Open bank resources are allocated dynamically as SDRAM burst transactions are
scheduled. Bank allocation is requested automatically by the controller when an
incoming transaction spans multiple SDRAM bursts or by the extended command
interface. When a bank must be reallocated, the least-recently-used open bank is used
as the replacement.
If the controller determines that the next pending command will cause the bank
request to not be honored, the bank might be held open or closed depending on the
pending operation. A request to close a bank with a pending operation in the timer
bank pool to the same row address causes the bank to remain open. A request to
leave a bank open with a pending command to the same bank but a different row
address causes a precharge operation to occur.
7.6.4.3 Write Combining
The SDRAM controller combines write operations from successive bursts on a port
where the starting address of the second burst is one greater than the ending address
of the first burst and the resulting burst length does not overflow the 11-bit
burst-length counters.
External Memory Interface Handbook Volume 3: Reference Material
286
7 Functional Description—HPS Memory Controller
Write combining does not occur if the previous bus command has finished execution
before the new command has been received.
7.6.4.4 Burst Length Support
The controller supports burst lengths of 2, 4, 8, and 16. Data widths of 8, 16, and 32
bits are supported for non-ECC operation and data widths of 24 and 40 bits are
supported for operations with ECC enabled. The following table shows the type of
SDRAM for each burst length.
Table 69.
SDRAM Burst Lengths
Burst Length
SDRAM
4
LPDDR2, DDR2
8
DDR2, DDR3, LPDDR2
16
LPDDR2
Width Matching
The SDRAM controller automatically performs data width conversion.
7.6.4.5 ECC
The single-port controller supports memory ECC calculated by the controller.
The controller ECC employs standard Hamming logic to detect and correct single-bit
errors and detect double-bit errors. The controller ECC is available for 16-bit and 32bit widths, each requiring an additional 8 bits of memory, resulting in an actual
memory width of 24-bits and 40-bits, respectively.
Note:
The level of ECC support is package dependent.
Functions the controller ECC provides are:
•
Byte Writes
•
ECC Write Backs
•
Notification of ECC Errors
7.6.4.5.1 Byte Writes
The memory controller performs a read-modify-write operation to ensure that the ECC
data remains valid when a subset of the bits of a word is being written.
Byte writes with ECC enabled are executed as a read-modify-write. Typical operations
only use a single entry in the timer bank pool. Controller ECC enabled sub-word writes
use two entries. The first operation is a read and the second operation is a write.
These two operations are transferred to the timer bank pool with an address
dependency so that the write cannot be performed until the read data has returned.
This approach ensures that any subsequent operations to the same address (from the
same port) are executed after the write operation, because they are ordered on the
row list after the write operation.
If an entire word is being written (but less than a full burst) and the DM pins are
connected, no read is necessary and only that word is updated. If controller ECC is
disabled, byte-writes have no performance impact.
External Memory Interface Handbook Volume 3: Reference Material
287
7 Functional Description—HPS Memory Controller
7.6.4.5.2 ECC Write Backs
If the controller ECC is enabled and a read operation results in a correctable ECC error,
the controller corrects the location in memory, if write backs are enabled. The
correction results in scheduling a new read-modify-write.
A new read is performed at the location to ensure that a write operation modifying the
location is not overwritten. The actual ECC correction operation is performed as a
read-modify-write operation. ECC write backs are enabled and disabled through the
cfg_enable_ecc_code_overwrites field in the ctrlcfg register.
Note:
Double-bit errors do not generate read-modify-write commands. Instead, double-bit
error address and count are reported through the erraddr and dbecount registers,
respectively. In addition, a double-bit error interrupt can be enabled through the
dramintr register.
7.6.4.5.3 User Notification of ECC Errors
When an ECC error occurs, an interrupt signal notifies the MPU subsystem, and the
ECC error information is stored in the status registers. The memory controller provides
interrupts for single-bit and double-bit errors.
The status of interrupts and errors are recorded in status registers, as follows:
•
The dramsts register records interrupt status.
•
The dramintr register records interrupt masks.
•
The sbecount register records the single-bit error count.
•
The dbecount register records the double-bit error count.
•
The erraddr register records the address of the most recent error.
For a 32-bit interface, ECC is calculated across a span of 8 bytes, meaning the error
address is a multiple of 8 bytes (4-bytes*2 burst length). To find the byte address of
the word that contains the error, you must multiply the value in the erraddr register
by 8.
7.6.4.6 Interleaving Options
The controller supports the following address-interleaving options:
•
Non-interleaved
•
Bank interleave without chip select interleave
•
Bank interleave with chip select interleave
The following interleaving examples use 512 megabits (Mb) x 16 DDR3 chips and are
documented as byte addresses. For RAMs with smaller address fields, the order of the
fields stays the same but the widths may change.
Non-interleaved
RAM mapping is non-interleaved.
External Memory Interface Handbook Volume 3: Reference Material
288
7 Functional Description—HPS Memory Controller
Figure 149. Non-interleaved Address Decoding
Address Decoding
(512 Mb x 16 DDR3 DRAM)
DDR 3
512 x 16
DDR 3
512 x 16
DDR 3
512 x 16
DDR 3
512 x 16
Controller
28
S
24
20
B (2 :0 )
16
12
8
R ( 15 :0 )
0
C ( 9 :0 )
Address Nomenclature
R =Row
B =Bank
C =Column
4
S =Chip Select
Bank Interleave Without Chip Select Interleave
Bank interleave without chip select interleave swaps row and bank from the noninterleaved address mapping. This interleaving allows smaller data structures to
spread across all banks in a chip.
Figure 150. Bank Interleave Without Chip Select Interleave Address Decoding
28
24
S
20
16
12
R ( 15 :0 )
8
B ( 2 :0)
4
0
C ( 9:0 )
Bank Interleave with Chip Select Interleave
Bank interleave with chip select interleave moves the row address to the top, followed
by chip select, then bank, and finally column address. This interleaving allows smaller
data structures to spread across multiple banks and chips (giving access to 16 total
banks for multithreaded access to blocks of memory). Memory timing is degraded
when switching between chips.
Figure 151. Bank Interleave With Chip Select Interleave Address Decoding
28
24
20
R ( 15 :0 )
16
12
S
B ( 2 :0)
8
4
0
C ( 9:0 )
External Memory Interface Handbook Volume 3: Reference Material
289
7 Functional Description—HPS Memory Controller
7.6.4.7 AXI-Exclusive Support
The single-port controller supports AXI-exclusive operations. The controller
implements a table shared across all masters, which can store up to 16 pending
writes. Table entries are allocated on an exclusive read and table entries are
deallocated on a successful write to the same address by any master.
Any exclusive write operation that is not present in the table returns an exclusive fail
as acknowledgement to the operation. If the table is full when the exclusive read is
performed, the table replaces a random entry.
Note:
When using AXI-exclusive operations, accessing the same location from Avalon-MM
interfaces can result in unpredictable results.
7.6.4.8 Memory Protection
The single-port controller has address protection to allow the software to configure
basic protection of memory from all masters in the system. If the system has been
designed exclusively with AMBA masters, TrustZone® is supported. Ports that use
Avalon-MM can be configured for port level protection.
Memory protection is based on physical addresses in memory. The single-port
controller can configure up to 20 rules to allow or prevent masters from accessing a
range of memory based on their AxIDs, level of security and the memory region being
accessed. If no rules are matched in an access, then default settings take effect.
The rules are stored in an internal protection table and can be accessed through
indirect addressing offsets in the protruledwr register in the CSR. To read a specific
rule, set the readrule bit and write the appropriate offset in the ruleoffset field of
the protruledwr register.
To write a new rule, three registers in the CSR must be configured:
1.
Table 70.
The protportdefault register is programmed to control the default behavior of
memory accesses when no rules match. When a bit is clear, all default accesses
from that port pass. When a bit is set, all default accesses from that port fails. The
bits are assigned as follows:
protportdefault register
Bits
31:10
Description
reserved
9
When this bit is set to 1, deny CPU writes during a default transaction.
When this bit is clear, allow CPU writes during a default transaction.
8
When this bit is set to 1, deny L3 writes during a default transaction.
When this bit is clear, allow L3 writes during a default transaction.
7
When this bit is set to 1, deny CPU reads during a default transaction.
When this bit is clear, allow CPU reads during a default transaction.
6
When this bit is set to 1, deny L3 reads during a default transaction.
When this bit is clear, allow L3 reads during a default transaction.
5:0
When this bit is set to 1, deny accesses from FPGA-to-SDRAM ports 0 through 5 during a
default transaction.
continued...
External Memory Interface Handbook Volume 3: Reference Material
290
7 Functional Description—HPS Memory Controller
Bits
Description
When this bit is clear, allow accesses from FPGA-to-SDRAM ports 0 through 5 during a
default transaction.
2.
The protruleid register gives the bounds of the AxID value that allows an
access
3. The protruledata register configures the specific security characteristics for a
rule.
Once the registers are configured, they can be committed to the internal protection
table by programming the ruleoffset field and setting the writerule bit in the
protruledwr register.
Secure and non-secure regions are specified by rules containing a starting address
and ending address with 1 MB boundaries for both addresses. You can override the
port defaults and allow or disallow all transactions.
The following table lists the fields that you can specify for each rule.
Table 71.
Fields for Rules in Memory Protection Table
Field
Width
Description
1
Set to 1 to activate the rule. Set to 0 to deactivate the rule.
10
Specifies the set of ports to which the rule applies, with one bit
representing each port, as follows: bits 0 to 5 correspond to FPGA
fabric ports 0 to 5, bit 6 corresponds to AXI L3 interconnect read,
bit 7 is the CPU read, bit 8 is L3 interconnect write, and bit 9 is
the CPU write.
12
Low transfer AxID of the rules to which this rule applies.
Incoming transactions match if they are greater than or equal to
this value. Ports with smaller AxIDs have the AxID shifted to the
lower bits and zero padded at the top.
12
High transfer AxID of the rules to which this rule applies.
Incoming transactions match if they are less than or equal to this
value.
Address_low
12
Points to a 1MB block and is the lower address. Incoming
addresses match if they are greater than or equal to this value.
Address_high
12
Upper limit of address. Incoming addresses match if they are less
than or equal to this value.
Protection
2
A value of 0x0 indicates that the rule applies to secure
transactions; a value of 0x1 indicates the rule applies to nonsecure transactions. Values 0x2 and 0x3 set the region to shared,
meaning both secure and non-secure accesses are valid.
Fail/allow
1
Set this value to 1 to force the operation to fail or succeed.
Valid
Port Mask
7
AxID_low7
AxID_high
7
Each port has a default access status of either allow or fail. Rules with the opposite
allow/fail value can override the default. The system evaluates each transaction
against every rule in the memory protection table. If a transaction arrives at a port
that defaults to access allowed, it fails only if a rule with the fail bit matches the
7 Although AxID and Port Mask could be redundant, including both in the table allows possible
compression of rules. If masters connected to a port do not have contiguous AxIDs, a portbased rule might be more efficient than an AxID-based rule, in terms of the number of rules
needed.
External Memory Interface Handbook Volume 3: Reference Material
291
7 Functional Description—HPS Memory Controller
transaction. Conversely, if a transaction arrives at a port that has the default rule set
to access denied, it allows access only if there is a matching rule that forces accessed
allowed. Transactions that fail the protection rules return a slave error (SLVERR).
The recommended sequence for writing a rule is:
1.
Write the protruledwr register fields as follows:
•
ruleoffset = offset selected by user that points to indirect offset in an
internal protection table..
•
writerule = 0
•
readrule = 0
2. Write the protruleaddr, protruleid, and protruledata registers so you
configure the rule you would like to enforce.
3.
Write the protruledwr register fields as follows:
•
ruleoffset = offset of the rule that needs to be written
•
writerule = 1
•
readrule = 0
Similarly, the recommended sequence for reading a rule is:
1.
2.
Write the protruledwr register fields as follows:
•
ruleoffset = offset of the rule that needs to be written
•
writerule = 0
•
readrule = 0
Write the protruledwr register fields as follows:
•
ruleoffset = offset of the rule that needs to be read
•
writerule = 0
•
readrule = 1
3. Read the values of the protruleaddr, protruleid, and protruledata
registers to determine the rule parameters.
The following figure represents an overview of how the protection rules are applied.
There is no priority among the 20 rules. All rules are always evaluated in parallel.
External Memory Interface Handbook Volume 3: Reference Material
292
7 Functional Description—HPS Memory Controller
Figure 152. SDRAM Protection Access Flow Diagram
Start
Evaluate all rules
for match
Ignore all rules that
do not match
Port default
access
Allowed
Not Allowed
Ignore all “allow”
rules that match
yes
Reject Access
Any “fail” rule
that matches?
Ignore all “fail” rules
that match
no
yes
Allow Access
Allow Access
Any “allow” rule
that matches?
no
Reject Access
End
Exclusive transactions are security checked on the read operation only. A write
operation can occur only if a valid read is marked in the internal exclusive table.
Consequently, a master performing an exclusive read followed by a write, can write to
memory only if the exclusive read was successful.
Related Links
ARM TrustZone®
For more information about TrustZone® refer to the ARM web page.
External Memory Interface Handbook Volume 3: Reference Material
293
7 Functional Description—HPS Memory Controller
7.6.4.9 Example of Configuration for TrustZone
For a TrustZone® configuration, memory is TrustZone divided into a range of memory
accessible by secure masters and a range of memory accessible by non-secure
masters. The two memory address ranges may have a range of memory that overlaps.
This example implements the following memory configuration:
•
2 GB total RAM size
•
0—512 MB dedicated secure area
•
513—576 MB shared area
•
577—2048 MB dedicated non-secure area
Figure 153. Example Memory Configuration
2048
Non-Secure
2 GB
1024
576
Shared
512
Secure
0
In this example, each port is configured by default to disallow all accesses. The
following table shows the two rules programmed into the memory protection table.
Table 72.
Rule
#
Rules in Memory Protection Table for Example Configuration
Port Mask
AxID
Low
AxID High
Address
Low
Address
High
protruledata.security
Fail/
Allow
1
0x3FF (1023)
0x000
0xFFF (4095)
0
576
0x1
Allow
2
0x3FF (1023)
0x000
0xFFF (4095)
512
2047
0x0
Allow
The port mask value, AxID Low, and AxID High, apply to all ports and all transfers
within those ports. Each access request is evaluated against the memory protection
table, and will fail unless there is a rule match allowing a transaction to complete
successfully.
External Memory Interface Handbook Volume 3: Reference Material
294
7 Functional Description—HPS Memory Controller
Table 73.
Result for a Sample Set of Transactions
Operation
Note:
Source
Address Accesses
Security
Access
Type
Result
Comments
Read
CPU
4096
secure
Allow
Matches rule 1.
Write
CPU
536, 870, 912
secure
Allow
Matches rule 1.
Write
L3 attached masters
605, 028, 350
secure
Fail
Does not match rule 1
(out of range of the
address field), does
not match rule 2
(protection bit
incorrect).
Read
L3 attached masters
4096
non-secure
Fail
Does not match rule 1
(AxPROT signal value
wrong), does not
match rule 2 (not in
address range).
Write
CPU
536, 870, 912
non-secure
Allow
Matches rule 2.
Write
L3 attached masters
605, 028, 350
non-secure
Allow
Matches rule 2.
If a master is using the Accelerator Coherency Port (ACP) to maintain cache coherency
with the Cortex-A9 MPCore processor, then the address ranges in the rules of the
memory protection table should be made mutually exclusive, such that the secure and
non-secure regions do not overlap and any area that is shared is part of the nonsecure region. This configuration prevents coherency issues from occurring.
7.7 SDRAM Power Management
The SDRAM controller subsystem supports the following power saving features in the
SDRAM:
•
Partial array self-refresh (PASR)
•
Power down
•
Deep power down for LPDDR2
To enable self-refresh for the memories of one or both chip selects, program the
selfshreq bit and the sefrfshmask bit in the lowpwreq register.
Power-saving mode initiates either due to a user command or from inactivity. The
number of idle clock cycles after which a memory can be put into power-down mode is
programmed through the autopdycycles field of the lowpwrtiming register.
Power-down mode forces the SDRAM burst-scheduling bank-management logic to
close all banks and issue the power down command. The SDRAM automatically
reactivates when an active SDRAM command is received.
To enable deep power down request for the LPDDR2 memories of one or both chip
selects, program the deeppwrdnreq bit and the deepwrdnmask field of the
lowpwreq register.
Other power-down modes are performed only under user control.
External Memory Interface Handbook Volume 3: Reference Material
295
7 Functional Description—HPS Memory Controller
7.8 DDR PHY
The DDR PHY connects the memory controller and external memory devices in the
speed critical command path.
The DDR PHY implements the following functions:
•
Calibration—the DDR PHY supports the JEDEC-specified steps to synchronize the
memory timing between the controller and the SDRAM chips. The calibration
algorithm is implemented in software.
•
Memory device initialization—the DDR PHY performs the mode register write
operations to initialize the devices. The DDR PHY handles re-initialization after a
deep power down.
•
Single-data-rate to double-data-rate conversion.
7.9 Clocks
All clocks are assumed to be asynchronous with respect to the ddr_dqs_clk memory
clock. All transactions are synchronized to memory clock domain.
Table 74.
SDRAM Controller Subsystem Clock Domains
Clock Name
Description
Clock for PHY
ddr_dq_clk
Clock for MPFE, single-port controller, CSR access, and PHY
ddr_dqs_clk
Clock for PHY that provides up to 2 times ddr_dq_clk frequency
ddr_2x_dqs_clk
Clock for CSR interface
l4_sp_clk
Clock for MPU interface
mpu_l2_ram_clk
Clock for L3 interface
l3_main_clk
Six separate clocks used for the FPGA-to-HPS SDRAM ports to the FPGA fabric
f2h_sdram_clk[5:0]
In terms of clock relationships, the FPGA fabric connects the appropriate clocks to
write data, read data, and command ports for the constructed ports.
7.10 Resets
The SDRAM controller subsystem supports a full reset (cold reset) and a warm reset.
The SDRAM controller can be configured to preserve memory contents during a warm
reset.
External Memory Interface Handbook Volume 3: Reference Material
296
7 Functional Description—HPS Memory Controller
To preserve memory contents, the reset manager can request that the single-port
controller place the SDRAM in self-refresh mode prior to issuing the warm reset. If
self-refresh mode is enabled before the warm reset to preserve memory contents, the
PHY and the memory timing logic is not reset, but the rest of the controller is reset.
7.10.1 Taking the SDRAM Controller Subsystem Out of Reset
When a cold or warm reset is issued in the HPS, the Reset Manager resets this module
and holds it in reset until software releases it.
After the MPU boots up, it can deassert the reset signal by clearing the appropriate
bits in the Reset Manager's corresponding reset trigger.
7.11 Port Mappings
The memory interface controller has a set of command, read data, and write data
ports that support AXI3, AXI4 and Avalon-MM. Tables are provided to identify port
assignments and functions.
Table 75.
Command Port Assignments
Command Port
Table 76.
Allowed Functions
0, 2, 4
FPGA fabric AXI read command ports
FPGA fabric Avalon-MM read or write command ports
1, 3, 5
FPGA fabric AXI write command ports
FPGA fabric Avalon-MM read or write command ports
6
L3 AXI read command port
7
MPU AXI read command port
8
L3 AXI write command port
9
MPU AXI write command port
Read Port Assignments
Read Port
0, 1, 2, 3
Table 77.
Allowed Functions
64-bit read data from the FPGA fabric. When 128-bit data read ports are created,
then read data ports 0 and1 get paired as well as 2 and 3.
4
32-bit L3 read data port
5
64-bit MPU read data port
Write Port Assignments
Write Port
Allowed Functions
0, 1, 2, 3
64-bit write data from the FPGA fabric. When 128-bit data write ports are created,
then write data ports 0 and 1 get paired as well as 2 and 3.
4
32-bit L3 write data port
5
64-bit MPU write data port
External Memory Interface Handbook Volume 3: Reference Material
297
7 Functional Description—HPS Memory Controller
7.12 Initialization
The SDRAM controller subsystem has control and status registers (CSRs) which control
the operation of the controller including DRAM type, DRAM timing parameters and
relative port priorities. It also has a small set of bits which depend on the FPGA fabric
to configure ports between the memory controller and the FPGA fabric; these bits are
set for you when you configure your implementation using the HPS GUI in Qsys.
The CSRs are configured using a dedicated slave interface, which provides access to
the registers. This region controls all SDRAM operation, MPFE scheduler configuration,
and PHY calibration.
The FPGA fabric interface configuration is programmed into the FPGA fabric and the
values of these register bits can be read by software. The ports can be configured
without software developers needing to know how the FPGA-to-HPS SDRAM interface
has been configured.
7.12.1 FPGA-to-SDRAM Protocol Details
The following topics summarize signals for the Avalon-MM Bidirectional port, AvalonMM Write Port, Avalon-MM Read Port, and AXI port.
Note:
If your device has multiple FPGA hardware images, then the same FPGA-to-SDRAM
port configuration should be used across all designs.
7.12.1.1 Avalon-MM Bidirectional Port
The Avalon-MM bidirectional ports are standard Avalon-MM ports used to dispatch read
and write operations.
Each configured Avalon-MM bidirectional port consists of the signals listed in the
following table.
Table 78.
Avalon-MM Bidirectional Port Signals
Name
Bit Width
Input/Output
Direction
Function
1
In
Clock for the Avalon-MM interface
1
In
Indicates read transaction
1
In
Indicates write transaction
32
In
Address of the transaction
32, 64, 128, or 256
Out
clk
8
read
8
write
address
Read data return
readdata
continued...
8 The Avalon-MM protocol does not allow read and write transactions to be posted concurrently.
External Memory Interface Handbook Volume 3: Reference Material
298
7 Functional Description—HPS Memory Controller
Name
Bit Width
Input/Output
Direction
1
Out
32, 64, 128, or 256
In
Write data for a transaction
4, 8, 16, 32
In
Byte enables for each write byte lane
1
Out
Indicates need for additional cycles to
complete a transaction
11
In
readdatavalid
Function
Indicates the readdata signal contains
valid data in response to a previous read
request.
writedata
byteenable
waitrequest
burstcount
Transaction burst length. The value of the
maximum burstcount parameter must
be a power of 2.
The read and write interfaces are configured to the same size. The byte-enable size
scales with the data bus size.
Related Links
Avalon Interface Specifications
Information about the Avalon-MM protocol
7.12.1.2 Avalon-MM Write-Only Port
The Avalon-MM write-only ports are standard Avalon-MM ports used to dispatch write
operations. Each configured Avalon-MM write port consists of the signals listed in the
following table.
Table 79.
Avalon-MM Write-Only Port Signals
Name
Bits
Direction
Function
1
In
Reset
1
In
Clock
1
In
Indicates write transaction
32
In
Address of the transaction
32, 64, 128, or 256
In
Write data for a transaction
reset
clk
write
address
writedata
continued...
External Memory Interface Handbook Volume 3: Reference Material
299
7 Functional Description—HPS Memory Controller
Name
Bits
Direction
4, 8, 16, 32
In
1
Out
11
In
Function
Byte enables for each write byte
byteenable
waitrequest
Indicates need for additional cycles to
complete a transaction
Transaction burst length
burstcount
Related Links
Avalon Interface Specifications
Information about the Avalon-MM protocol
7.12.1.3 Avalon-MM Read Port
The Avalon-MM read ports are standard Avalon-MM ports used only to dispatch read
operations. Each configured Avalon-MM read port consists of the signals listed in the
following table.
Table 80.
Avalon-MM Read Port Signals
Name
Bits
Direction
Function
1
In
Reset
1
In
Clock
1
In
Indicates read transaction
32
In
Address of the transaction
32, 64, 128, or 256
Out
Read data return
1
Out
Flags valid cycles for read data return
1
Out
Indicates the need for additional cycles to
complete a transaction. Needed for read
operations when delay is needed to
accept the read command.
11
In
reset
clk
read
address
readdata
readdatavalid
waitrequest
burstcount
Related Links
Avalon Interface Specifications
Information about the Avalon-MM protocol
External Memory Interface Handbook Volume 3: Reference Material
300
Transaction burst length
7 Functional Description—HPS Memory Controller
7.12.1.4 AXI Port
The AXI port uses an AXI-3 interface. Each configured AXI port consists of the signals
listed in the following table. Because the AXI protocol allows simultaneous read and
write commands to be issued, two SDRAM control ports are required to form an AXI
interface.
Table 81.
Name
AXI Port Signals
Bits
Direction
ARESETn
1
In
n/a
Reset
ACLK
1
In
n/a
Clock
AWID
4
In
Write address
Write identification tag
32
In
Write address
Write address
AWLEN
4
In
Write address
Write burst length
AWSIZE
3
In
Write address
Width of the transfer size
AWBURST
2
In
Write address
Burst type
AWLOCK
2
In
Write address
Lock type signal which indicates if the access is
exclusive; valid values are 0x0 (normal access)
and 0x1 (exclusive access)
AWCACHE
4
In
Write address
Cache policy type
AWPROT
3
In
Write address
Protection-type signal used to indicate whether a
transaction is secure or non-secure
AWREADY
1
Out
Write address
Indicates ready for a write command
AWVALID
1
In
Write address
Indicates valid write command.
WID
4
In
Write data
Write data transfer ID
WDATA
32, 64,
128 or
256
In
Write data
Write data
WSTRB
4, 8, 16,
32
In
Write data
Byte-based write data strobe. Each bit width
corresponds to 8 bit wide transfer for 32-bit wide
to 256-bit wide transfer.
WLAST
1
In
Write data
Last transfer in a burst
WVALID
1
In
Write data
Indicates write data and strobes are valid
WREADY
1
Out
Write data
Indicates ready for write data and strobes
BID
4
Out
Write response
Write response transfer ID
BRESP
2
Out
Write response
Write response status
BVALID
1
Out
Write response
Write response valid signal
BREADY
1
In
Write response
Write response ready signal
ARID
4
In
Read address
Read identification tag
32
In
Read address
Read address
ARLEN
4
In
Read address
Read burst length
ARSIZE
3
In
Read address
Width of the transfer size
AWADDR
ARADDR
Channel
Function
continued...
External Memory Interface Handbook Volume 3: Reference Material
301
7 Functional Description—HPS Memory Controller
Name
Bits
Direction
Channel
Function
ARBURST
2
In
Read address
Burst type
ARLOCK
2
In
Read address
Lock type signal which indicates if the access is
exclusive; valid values are 0x0 (normal access)
and 0x1 (exclusive access)
ARCACHE
4
In
Read address
Lock type signal which indicates if the access is
exclusive; valid values are 0x0 (normal access)
and 0x1 (exclusive access)
ARPROT
3
In
Read address
Protection-type signal used to indicate whether a
transaction is secure or non-secure
ARREADY
1
Out
Read address
Indicates ready for a read command
ARVALID
1
In
Read address
Indicates valid read command
RID
4
Out
Read data
Read data transfer ID
RDATA
32, 64,
128 or
256
Out
Read data
Read data
RRESP
2
Out
Read data
Read response status
RLAST
1
Out
Read data
Last transfer in a burst
RVALID
1
Out
Read data
Indicates read data is valid
RREADY
1
In
Read data
Read data channel ready signal
Related Links
ARM AMBA Open Specification
AMBA Open Specifications, including information about the AXI-3 interface
7.13 SDRAM Controller Subsystem Programming Model
SDRAM controller configuration occurs through software programming of the
configuration registers using the CSR interface.
7.13.1 HPS Memory Interface Architecture
The configuration and initialization of the memory interface by the ARM processor is a
significant difference compared to the FPGA memory interfaces, and results in several
key differences in the way the HPS memory interface is defined and configured.
Boot-up configuration of the HPS memory interface is handled by the initial software
boot code, not by the FPGA programmer, as is the case for the FPGA memory
interfaces. The Quartus Prime software is involved in defining the configuration of I/O
ports which is used by the boot-up code, as well as timing analysis of the memory
interface. Therefore, the memory interface must be configured with the correct PHYlevel timing information. Although configuration of the memory interface in Qsys is
still necessary, it is limited to PHY- and board-level settings.
7.13.2 HPS Memory Interface Configuration
To configure the external memory interface components of the HPS, open the HPS
interface by selecting the Arria V/Cyclone V Hard Processor System component in
Qsys. Within the HPS interface, select the EMIF tab to open the EMIF parameter editor.
External Memory Interface Handbook Volume 3: Reference Material
302
7 Functional Description—HPS Memory Controller
The EMIF parameter editor contains four additional tabs: PHY Settings, Memory
Parameters, Memory Timing, and Board Settings. The parameters available on these
tabs are similar to those available in the parameter editors for non-SoC device
families.
There are significant differences between the EMIF parameter editor for the Hard
Processor System and the parameter editors for non-SoC devices, as follows:
Note:
•
Because the HPS memory controller is not configurable through the Quartus Prime
software, the Controller and Diagnostic tabs, which exist for non-SoC devices, are
not present in the EMIF parameter editor for the hard processor system.
•
Unlike the protocol-specific parameter editors for non-SoC devices, the EMIF
parameter editor for the Hard Processor System supports multiple protocols,
therefore there is an SDRAM Protocol parameter, where you can specify your
external memory interface protocol. By default, the EMIF parameter editor
assumes the DDR3 protocol, and other parameters are automatically populated
with DDR3-appropriate values. If you select a protocol other than DDR3, change
other associated parameter values appropriately.
•
Unlike the memory interface clocks in the FPGA, the memory interface clocks for
the HPS are initialized by the boot-up code using values provided by the
configuration process. You may accept the values provided by UniPHY, or you may
use your own PLL settings. If you choose to specify your own PLL settings, you
must indicate that the clock frequency that UniPHY should use is the requested
clock frequency, and not the achieved clock frequency calculated by UniPHY.
The HPS does not support EMIF synthesis generation, compilation, or timing analysis.
The HPS hard memory controller cannot be bonded with another hard memory
controller on the FPGA portion of the device.
7.13.3 HPS Memory Interface Simulation
Qsys provides a complete simulation model of the HPS memory interface controller
and PHY, providing cycle-level accuracy, comparable to the simulation models for the
FPGA memory interface.
The simulation model supports only the skip-cal simulation mode; quick-cal and fullcal are not supported. An example design is not provided. However, you can create a
test design by adding the traffic generator component to your design using Qsys. Also,
the HPS simulation model does not use external memory pins to connect to the DDR
memory model; instead, the memory model is incorporated directly into the HPS
SDRAM interface simulation modules. The memory instance incorporated into the HPS
model is in the simulation model hierarchy at: hps_0/fpga_interfaces/f2sdram/
hps_sdram_inst/mem/
Simulation of the FPGA-to-SDRAM interfaces requires that you first bring the
interfaces out of reset, otherwise transactions cannot occur. Connect the H2F reset to
the F2S port resets and add a stage to your testbench to assert and deassert the H2F
reset in the HPS. Appropriate Verilog code is shown below:
initial
begin
// Assert reset
<base name>.hps.fpga_interfaces.h2f_reset_inst.reset_assert();
// Delay
#1
External Memory Interface Handbook Volume 3: Reference Material
303
7 Functional Description—HPS Memory Controller
// Deassert reset
<base name>.hps.fpga_interfaces.h2f_reset_inst.reset_deassert();
end
7.13.4 Generating a Preloader Image for HPS with EMIF
To generate a Preloader image for an HPS-based external memory interface, you must
complete the following tasks:
•
Create a Qsys project.
•
Create a top-level file and add constraints.
•
Create a Preloader BSP file.
•
Create a Preloader image.
7.13.4.1 Creating a Qsys Project
Before you can generate a preloader image, you must create a Qsys project, as
follows:
1. On the Tools menu in the Quartus Prime software, click Qsys.
2.
Under Component library, expand Embedded Processor System, select Hard
Processor System and click Add.
3.
Specify parameters for the FPGA Interfaces, Peripheral Pin Multiplexing, and
HPS Clocks, based on your design requirements.
4.
On the SDRAM tab, select the SDRAM protocol for your interface.
5.
Populate the necessary parameter fields on the PHY Settings, Memory
Parameters, Memory Timing, and Board Settings tabs.
6.
Add other Qsys components in your Qsys design and make the appropriate bus
connections.
7. Save the Qsys project.
8.
Click Generate on the Generation tab, to generate the Qsys design.
7.13.4.2 Creating a Top-Level File and Adding Constraints
This topic describes adding your Qsys system to your top-level design and adding
constraints to your design.
1. Add your Qsys system to your top-level design.
2.
Add the Quartus Prime IP files (.qip) generated in step 2, to your Quartus Prime
project.
3.
Perform analysis and synthesis on your design.
4. Constrain your EMIF design by running the
<variation_name>_p0_pin_assignments.tcl pin constraints script.
5.
Add other necessary constraints—such as timing constraints, location
assignments, and pin I/O standard assignments—for your design.
6.
Compile your design to generate an SRAM object file (.sof) and the hardware
handoff files necessary for creating a preloader image.
External Memory Interface Handbook Volume 3: Reference Material
304
7 Functional Description—HPS Memory Controller
Note: You must regenerate the hardware handoff files whenever the HPS
configuration changes; for example, due to changes in Peripheral Pin
Multiplexing or I/O standard for HPS pins.
Related Links
Intel SoC FPGA Embedded Design Suite User's Guide
For more information on how to create a preloader BSP file and image.
7.14 Debugging HPS SDRAM in the Preloader
To assist in debugging your design, tools are available at the preloader stage.
•
UART or semihosting printout
•
Simple memory test
•
Debug report
•
Predefined data patterns
The following topics provide procedures for implementing each of the above tools.
7.14.1 Enabling UART or Semihosting Printout
UART printout is enabled by default. If UART is not available on your system, you can
use semihosting together with the debugger tool. To enable semihosting in the
Preloader, follow these steps:
1. When you create the .bsp file in the BSP Editor, select SEMIHOSTING in the
spl.debug window.
2.
Enable semihosting in the debugger, by typing set semihosting enabled
true at the command line in the debugger.
External Memory Interface Handbook Volume 3: Reference Material
305
7 Functional Description—HPS Memory Controller
7.14.2 Enabling Simple Memory Test
After the SDRAM is successfully calibrated, a simple memory test may be performed
using the debugger.
1. When you create the .bsp file in the BSP Editor, select
HARDWARE_DIAGNOSTIC in the spl.debug window..
2.
The simple memory test assumes SDRAM with a memory size of 1 GB. If your
board contains a different SDRAM memory size, open the file <design folder>
\spl_bsp\uboot-socfpga\include\configs\socfpga_cyclone5.h in a
text editor, and change the PHYS_SDRAM_1_SIZE parameter at line 292 to specify
your actual memory size in bytes.
External Memory Interface Handbook Volume 3: Reference Material
306
7 Functional Description—HPS Memory Controller
3.
The simple memory test assumes SDRAM with a memory size of 1 GB. If your
board contains a different SDRAM memory size, open the file <design folder>
\spl_bsp\uboot-socfpga\include\configs\socfpga_arria5.h in a text
editor, and change the PHYS_SDRAM_1_SIZE parameter at line 292 to specify
your actual memory size in bytes.
7.14.3 Enabling the Debug Report
You can enable the SDRAM calibration sequencer to produce a debug report on the
UART printout or semihosting output. To enable the debug report, follow these steps:
External Memory Interface Handbook Volume 3: Reference Material
307
7 Functional Description—HPS Memory Controller
1.
After you have enabled the UART or semihosting, open the file <project
directory>\hps_isw_handoff\sequencer_defines.hin a text editor.
2.
Locate the line #define RUNTIME_CAL_REPORT 0 and change it to #define
RUNTIME_CAL_REPORT 1.
Figure 154. Semihosting Printout With Debug Support Enabled
7.14.3.1 Analysis of Debug Report
The following analysis will help you interpret the debug report.
External Memory Interface Handbook Volume 3: Reference Material
308
7 Functional Description—HPS Memory Controller
•
The Read Deskew and Write Deskew results shown in the debug report are before
calibration. (Before calibration results are actually from the window seen during
calibration, and are most useful for debugging.)
•
For each DQ group, the Write Deskew, Read Deskew, DM Deskew, and Read after
Write results map to the before-calibration margins reported in the EMIF Debug
Toolkit.
Note: The Write Deskew, Read Deskew, DM Deskew, and Read after Write results
are reported in delay steps (nominally 25ps, in Arria V and Cyclone V
devices), not in picoseconds.
•
DQS Enable calibration is reported as a VFIFO setting (in one clock period steps),
a phase tap (in one-eighth clock period steps), and a delay chain step (in 25ps
steps).
SEQ.C: DQS Enable ; Group 0 ; Rank 0 ; Start VFIFO
SEQ.C: DQS Enable ; Group 0 ; Rank 0 ; End
VFIFO
SEQ.C: DQS Enable ; Group 0 ; Rank 0 ; Center VFIFO
5 ; Phase 6 ; Delay
6 ; Phase 5 ; Delay
6 ; Phase 2 ; Delay
4
9
1
Analysis of DQS Enable results: A VFIFO tap is 1 clock period, a phase is 1/8 clock
period (45 degrees) and delay is nominally 25ps per tap. The DQSen window is
the difference between the start and end—for the above example, assuming a
frequency of 400 MHz (2500ps), that calculates as follows: start is 5*2500
+ 6*2500/8 +4*25 = 14475ps. By the same calculation, the end is 16788ps.
Consequently, the DQSen window is 2313ps.
•
The size of a read window or write window is equal to (left edge + right
edge) * delay chain step size. Both the left edge and the right edge can
be negative or positive.:
SEQ.C:
delay
SEQ.C:
delay
Read Deskew ;
0 ; DQS delay
Write Deskew ;
6 ; DQS delay
DQ
8
DQ
4
0 ; Rank 0 ; Left edge
18 ; Right edge
27 ; DQ
0 ; Rank 0 ; Left edge
30 ; Right edge
17 ; DQ
Analysis of DQ and DQS delay results: The DQ and DQS output delay (write) is the
D5 delay chain. The DQ input delay (read) is the D1 delay chain, the DQS input
delay (read) is the D4 delay chain.
External Memory Interface Handbook Volume 3: Reference Material
309
7 Functional Description—HPS Memory Controller
•
Consider the following example of latency results:
SEQ.C: LFIFO Calibration ; Latency 10
Analysis of latency results: This is the calibrated PHY read latency. The EMIF
Debug Toolkit does not report this figure. This latency is reported in clock cycles.
•
Consider the following example of FOM results:
SEQ.C: FOM IN = 83
SEQ.C: FOM OUT = 91
Analysis of FOM results: The FOM IN value is a measure of the health of the read
interface; it is calculated as the sum over all groups of the minimum margin on
DQ plus the margin on DQS, divided by 2. The FOM OUT is a measure of the
health of the write interface; it is calculated as the sum over all groups of the
minimum margin on DQ plus the margin on DQS, divided by 2. You may refer to
these values as indicators of improvement when you are experimenting with
various termination schemes, assuming there are no individual misbehaving DQ
pins.
•
The debug report does not provide delay chain step size values. The delay chain
step size varies with device speed grade. Refer to your device data sheet for exact
incremental delay values for delay chains.
Related Links
Functional Description–UniPHY
For more information about calibration, refer to the Calibration Stages section in
the Functional Description-UniPHY chapter of the External Memory Interface
Handbook .
7.14.4 Writing a Predefined Data Pattern to SDRAM in the Preloader
You can include your own code to write a predefined data pattern to the SDRAM in the
preloader for debugging purposes.
1.
Include your code in the file: <project_folder>\software\spl_bsp\ubootsocfpga\arch\arm\cpu\armv7\socfpga\spl.c .
Adding the following code to the spl.c file causes the controller to write walking
1s and walking 0s, repeated five times, to the SDRAM.
/*added for demo, place after the last #define statement in spl.c */
#define ROTATE_RIGHT(X) ( (X>>1) | (X&1?0X80000000:0) )
/*added for demo, place after the calibration code */
test_data_walk0((long *)0x100000,PHYS_SDRAM_1_SIZE);
int test_data_walk0(long *base, long maxsize)
{
volatile long *addr;
long
cnt;
ulong
data_temp[3];
ulong
expected_data[3];
ulong
read_data;
int
i = 0; //counter to loop different data pattern
int
num_address;
num_address=50;
data_temp[0]=0XFFFFFFFE; //initial data for walking 0 pattern
data_temp[1]=0X00000001; //initial data for walking 1 pattern
data_temp[2]=0XAAAAAAAA; //initial data for A->5 switching
External Memory Interface Handbook Volume 3: Reference Material
310
7 Functional Description—HPS Memory Controller
expected_data[0]=0XFFFFFFFE; //initial data for walking 0 pattern
expected_data[1]=0X00000001; //initial data for walking 1 pattern
expected_data[2]=0XAAAAAAAA; //initial data for A->5 switching
for (i=0;i<3;i++) {
printf("\nSTARTED %08X DATA PATTERN !!!!\n",data_temp[i]);
/*write*/
for (cnt = (0+i*num_address); cnt < ((i+1)*num_address) ; cnt++ ) {
addr = base + cnt;
/* pointer arith! */
sync ();
*addr = data_temp[i];
data_temp[i]=ROTATE_RIGHT(data_temp[i]);
}
/*read*/
for (cnt = (0+i*num_address); cnt < ((i+1)*num_address) ; cnt = cnt++ ) {
addr = base + cnt;
/* pointer arith! */
sync ();
read_data=*addr;
printf("Address:%X Expected: %08X
Read:%08X \n",addr,
expected_data[i],read_data);
if (expected_data[i] !=read_data) {
puts("!!!!!!FAILED!!!!!!\n\n");
hang();
}
expected_data[i]=ROTATE_RIGHT(expected_data[i]);
}
}
}
====//End Of Code//=====
Figure 155. Memory Contents After Executing Example Code
7.15 SDRAM Controller Address Map and Register Definitions
This section lists the SDRAM register address map and describes the registers.
7.15.1 SDRAM Controller Address Map
Address map for the SDRAM Interface registers
Base Address: 0xFFC20000
External Memory Interface Handbook Volume 3: Reference Material
311
7 Functional Description—HPS Memory Controller
SDRAM Controller Module
Register
Offset
Widt
h
Acce
ss
Reset Value
ctrlcfg on page 315
0x5000
32
RW
0x0
Controller Configuration Register
dramtiming1 on page 317
0x5004
32
RW
0x0
DRAM Timings 1 Register
dramtiming2 on page 317
0x5008
32
RW
0x0
DRAM Timings 2 Register
dramtiming3 on page 318
0x500C
32
RW
0x0
DRAM Timings 3 Register
dramtiming4 on page 319
0x5010
32
RW
0x0
DRAM Timings 4 Register
lowpwrtiming on page 320
0x5014
32
RW
0x0
Lower Power Timing Register
dramodt on page 320
0x5018
32
RW
0x0
ODT Control Register
dramaddrw on page 321
0x502C
32
RW
0x0
DRAM Address Widths Register
dramifwidth on page 322
0x5030
32
RW
0x0
DRAM Interface Data Width Register
dramsts on page 323
0x5038
32
RW
0x0
DRAM Status Register
dramintr on page 323
0x503C
32
RW
0x0
ECC Interrupt Register
sbecount on page 324
0x5040
32
RW
0x0
ECC Single Bit Error Count Register
dbecount on page 325
0x5044
32
RW
0x0
ECC Double Bit Error Count Register
erraddr on page 325
0x5048
32
RW
0x0
ECC Error Address Register
dropcount on page 326
0x504C
32
RW
0x0
ECC Auto-correction Dropped Count
Register
dropaddr on page 326
0x5050
32
RW
0x0
ECC Auto-correction Dropped Address
Register
lowpwreq on page 327
0x5054
32
RW
0x0
Low Power Control Register
lowpwrack on page 328
0x5058
32
RW
0x0
Low Power Acknowledge Register
staticcfg on page 329
0x505C
32
RW
0x0
Static Configuration Register
ctrlwidth on page 329
0x5060
32
RW
0x0
Memory Controller Width Register
portcfg on page 330
0x507C
32
RW
0x0
Port Configuration Register
fpgaportrst on page 332
0x5080
32
RW
0x0
FPGA Ports Reset Control Register
protportdefault on page
333
0x508C
32
RW
0x0
Memory Protection Port Default Register
protruleaddr on page 334
0x5090
32
RW
0x0
Memory Protection Address Register
protruleid on page 334
0x5094
32
RW
0x0
Memory Protection ID Register
protruledata on page 335
0x5098
32
RW
0x0
Memory Protection Rule Data Register
protrulerdwr on page 336
0x509C
32
RW
0x0
Memory Protection Rule Read-Write
Register
mppriority on page 337
0x50AC
32
RW
0x0
Scheduler priority Register
remappriority on page
338
0x50E0
32
RW
0x0
Controller Command Pool Priority Remap
Register
External Memory Interface Handbook Volume 3: Reference Material
312
Description
7 Functional Description—HPS Memory Controller
Port Sum of Weight Register
Register
Offset
Widt
h
Acce
ss
Reset Value
Description
mpweight_0_4 on page 339
0x50B0
32
RW
0x0
Port Sum of Weight Register[1/4]
mpweight_1_4 on page 339
0x50B4
32
RW
0x0
Port Sum of Weight Register[2/4]
mpweight_2_4 on page 340
0x50B8
32
RW
0x0
Port Sum of Weight Register[3/4]
mpweight_3_4 on page 340
0x50BC
32
RW
0x0
Port Sum of Weight Register[4/4]
7.15.1.1 SDRAM Controller Module Register Descriptions
Address map for the SDRAM controller and multi-port front-end. All registers in this
group reset to zero.
Offset: 0x5000
ctrlcfg on page 315
The Controller Configuration Register determines the behavior of the controller.
dramtiming1 on page 317
This register implements JEDEC standardized timing parameters. It should be
programmed in clock cycles, for the value specified by the memory vendor.
dramtiming2 on page 317
This register implements JEDEC standardized timing parameters. It should be
programmed in clock cycles, for the value specified by the memory vendor.
dramtiming3 on page 318
This register implements JEDEC standardized timing parameters. It should be
programmed in clock cycles, for the value specified by the memory vendor.
dramtiming4 on page 319
This register implements JEDEC standardized timing parameters. It should be
programmed in clock cycles, for the value specified by the memory vendor.
lowpwrtiming on page 320
This register controls the behavior of the low power logic in the controller.
dramodt on page 320
This register controls which ODT pin asserts with chip select 0 (CS0) assertion and
which ODT pin asserts with chip select 1 (CS1) assertion.
dramaddrw on page 321
This register configures the width of the various address fields of the DRAM. The
values specified in this register must match the memory devices being used.
dramifwidth on page 322
This register controls the interface width of the SDRAM controller.
dramsts on page 323
This register provides the status of the calibration and ECC logic.
dramintr on page 323
This register can enable, disable and clear the SDRAM error interrupts.
sbecount on page 324
This register tracks the single-bit error count.
dbecount on page 325
This register tracks the double-bit error count.
External Memory Interface Handbook Volume 3: Reference Material
313
7 Functional Description—HPS Memory Controller
erraddr on page 325
This register holds the address of the most recent ECC error.
dropcount on page 326
This register holds the address of the most recent ECC error.
dropaddr on page 326
This register holds the last dropped address.
lowpwreq on page 327
This register instructs the controller to put the DRAM into a power down state.
Note that some commands are only valid for certain memory types.
lowpwrack on page 328
This register gives the status of the power down commands requested by the Low
Power Control register.
staticcfg on page 329
This register controls configuration values which cannot be updated during active
transfers. First configure the membl and eccn fields and then re-write these fields
while setting the applycfg bit. The applycfg bit is write only.
ctrlwidth on page 329
This register controls the width of the physical DRAM interface.
portcfg on page 330
Each bit of the autopchen field maps to one of the control ports. If a port
executes mostly sequential memory accesses, the corresponding autopchen bit
should be 0. If the port has highly random accesses, then its autopchen bit
should be set to 1.
fpgaportrst on page 332
This register implements functionality to allow the CPU to control when the MPFE
will enable the ports to the FPGA fabric.
protportdefault on page 333
This register controls the default protection assignment for a port. Ports which
have explicit rules which define regions which are illegal to access should set the
bits to pass by default. Ports which have explicit rules which define legal areas
should set the bit to force all transactions to fail. Leaving this register to all zeros
should be used for systems which do not desire any protection from the memory
controller.
protruleaddr on page 334
This register is used to control the memory protection for port 0 transactions.
Address ranges can either be used to allow access to memory regions or disallow
access to memory regions. If TrustZone is being used, access can be enabled for
protected transactions or disabled for unprotected transactions. The default state
of this register is to allow all access. Address values used for protection are only
physical addresses.
protruleid on page 334
This register configures the AxID for a given protection rule.
protruledata on page 335
This register configures the protection memory characteristics of each protection
rule.
protrulerdwr on page 336
This register is used to perform read and write operations to the internal protection
table.
mppriority on page 337
External Memory Interface Handbook Volume 3: Reference Material
314
7 Functional Description—HPS Memory Controller
This register is used to configure the DRAM burst operation scheduling.
remappriority on page 338
This register applies another level of port priority after a transaction is placed in
the single port queue.
Port Sum of Weight Register Register Descriptions on page 339
This register is used to configure the DRAM burst operation scheduling.
7.15.1.1.1 ctrlcfg
The Controller Configuration Register determines the behavior of the controller.
Module Instance
Base Address
sdr
Register Address
0xFFC20000
0xFFC25000
Offset: 0x5000
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
Reserved
25
24
23
22
burst
terme
n
burst
intre
n
nodmp
ins
dqstr
ken
RW
0x0
RW
0x0
RW
0x0
RW
0x0
9
8
7
6
21
20
19
18
17
16
1
0
starvelimit
RW 0x0
15
14
13
12
11
10
reord
eren
gendb
e
gensb
e
eccco
rren
eccen
addrorder
membl
memtype
RW 0x0
RW 0x0
RW
0x0
RW
0x0
RW
0x0
RW 0x0
RW
0x0
cfg_e
nable
_ecc_
code_
overw
rites
RW
0x0
5
4
3
2
RW
0x0
ctrlcfg Fields
Name
Bit
Description
Access
Rese
t
25
bursttermen
Set to a one to enable the controller to issue burst terminate
commands. This must only be set when the DRAM memory
type is LPDDR2.
RW
0x0
24
burstintren
Set to a one to enable the controller to issue burst interrupt
commands. This must only be set when the DRAM memory
type is LPDDR2.
RW
0x0
23
nodmpins
Set to a one to enable DRAM operation if no DM pins are
connected.
RW
0x0
22
dqstrken
Enables DQS tracking in the PHY.
RW
0x0
continued...
External Memory Interface Handbook Volume 3: Reference Material
315
7 Functional Description—HPS Memory Controller
Bit
Description
Access
Rese
t
starvelimit
Specifies the number of DRAM burst transactions an individual
transaction will allow to reorder ahead of it before its priority
is raised in the memory controller.
RW
0x0
15
reorderen
This bit controls whether the controller can re-order
operations to optimize SDRAM bandwidth. It should generally
be set to a one.
RW
0x0
14
gendbe
Enable the deliberate insertion of double bit errors in data
written to memory. This should only be used for testing
purposes.
RW
0x0
13
gensbe
Enable the deliberate insertion of single bit errors in data
written to memory. This should only be used for testing
purposes.
RW
0x0
12
cfg_enable_ecc_code_ov
erwrites
Set to a one to enable ECC overwrites. ECC overwrites occur
when a correctable ECC error is seen and cause a new read/
modify/write to be scheduled for that location to clear the ECC
error.
RW
0x0
11
ecccorren
Enable auto correction of the read data returned when single
bit error is detected.
RW
0x0
10
eccen
Enable the generation and checking of ECC. This bit must only
be set if the memory connected to the SDRAM interface is 24
or 40 bits wide. If you set this, you must clear the
useeccasdata field in the staticcfg register.
RW
0x0
9:8
addrorder
This bit field selects the order for address interleaving.
Programming this field with different values gives different
mappings between the AXI or Avalon-MM address and the
SDRAM address. Program this field with the following binary
values to select the ordering.
RW
0x0
21:1
6
Name
Value
Description
Address Interleaving
0x0
chip, row, bank,
column
Bank interleaved with no
rank (chip select)
interleaving
0x1
chip, bank, row,
column
No interleaving
0x2
row, chip, bank,
column
Bank interleaved with rank
(chip select) interleaving
0x3
reserved
N/A
Intel recommends programming addrorder to 0x0 or 0x2.
7:3
membl
Configures burst length as a static decimal value. Legal values
are valid for JEDEC allowed DRAM values for the DRAM
selected in cfg_type. For DDR3, this should be programmed
with 8 (binary "01000"), for DDR2 it can be either 4 or 8
depending on the exact DRAM chip. LPDDR2 can be
programmed with 4, 8, or 16 and LPDDR can be programmed
with 2, 4, or 8. You must also program the membl field in the
staticcfg register.
RW
0x0
2:0
memtype
This bit field selects the memory type. This field can be
programmed with the following binary values:
RW
0x0
Value
0x0
Description
Reserved
External Memory Interface Handbook Volume 3: Reference Material
316
7 Functional Description—HPS Memory Controller
Bit
Name
Description
Value
Access
Rese
t
Description
0x1
Memory type is DDR2 SDRAM
0x2
Memory type is DDR3 SDRAM
0x3
reserved
0x4
Memory type is LPDDR2 SDRAM
0x5-0x7
Reserved
7.15.1.1.2 dramtiming1
This register implements JEDEC standardized timing parameters. It should be
programmed in clock cycles, for the value specified by the memory vendor.
Module Instance
Base Address
sdr
Register Address
0xFFC20000
0xFFC25004
Offset: 0x5004
Access: RW
Bit Fields
31
30
15
29
14
13
28
27
26
25
24
23
22
21
20
19
18
17
16
trfc
tfaw
trrd
RW 0x0
RW 0x0
RW 0x0
12
11
10
9
8
7
6
5
4
3
2
1
trrd
tcl
tal
tcwl
RW 0x0
RW 0x0
RW 0x0
RW 0x0
0
dramtiming1 Fields
Name
Bit
Description
Access
Rese
t
31:2
4
trfc
The refresh cycle timing parameter.
RW
0x0
23:1
8
tfaw
The four-activate window timing parameter.
RW
0x0
17:1
4
trrd
The activate to activate, different banks timing parameter.
RW
0x0
13:9
tcl
Memory read latency.
RW
0x0
8:4
tal
Memory additive latency.
RW
0x0
3:0
tcwl
Memory write latency.
RW
0x0
7.15.1.1.3 dramtiming2
This register implements JEDEC standardized timing parameters. It should be
programmed in clock cycles, for the value specified by the memory vendor.
External Memory Interface Handbook Volume 3: Reference Material
317
7 Functional Description—HPS Memory Controller
Module Instance
Base Address
sdr
Register Address
0xFFC20000
0xFFC25008
Offset: 0x5008
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
Reserved
15
14
13
12
26
25
24
23
22
21
20
19
18
17
16
twtr
twr
trp
trcd
RW 0x0
RW 0x0
RW 0x0
RW
0x0
11
10
9
8
7
6
trcd
trefi
RW 0x0
RW 0x0
5
4
3
2
1
0
dramtiming2 Fields
Bit
Name
Description
Access
Rese
t
28:2
5
twtr
The write to read timing parameter.
RW
0x0
24:2
1
twr
The write recovery timing.
RW
0x0
20:1
7
trp
The precharge to activate timing parameter.
RW
0x0
16:1
3
trcd
The activate to read/write timing parameter.
RW
0x0
12:0
trefi
The refresh interval timing parameter.
RW
0x0
7.15.1.1.4 dramtiming3
This register implements JEDEC standardized timing parameters. It should be
programmed in clock cycles, for the value specified by the memory vendor.
Module Instance
sdr
Base Address
0xFFC20000
Register Address
0xFFC2500C
Offset: 0x500C
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
External Memory Interface Handbook Volume 3: Reference Material
318
7 Functional Description—HPS Memory Controller
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
Reserved
15
14
13
12
11
10
9
8
7
6
20
19
18
17
tccd
tmrd
RW 0x0
RW 0x0
5
4
3
2
1
tmrd
trc
tras
trtp
RW
0x0
RW 0x0
RW 0x0
RW 0x0
16
0
dramtiming3 Fields
Name
Bit
Description
Access
Rese
t
22:1
9
tccd
The CAS to CAS delay time.
RW
0x0
18:1
5
tmrd
Mode register timing parameter.
RW
0x0
14:9
trc
The activate to activate timing parameter.
RW
0x0
8:4
tras
The activate to precharge timing parameter.
RW
0x0
3:0
trtp
The read to precharge timing parameter.
RW
0x0
7.15.1.1.5 dramtiming4
This register implements JEDEC standardized timing parameters. It should be
programmed in clock cycles, for the value specified by the memory vendor.
Base Address
Module Instance
sdr
Register Address
0xFFC20000
0xFFC25010
Offset: 0x5010
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
23
Reserved
15
14
13
12
11
10
9
8
7
22
21
20
19
18
17
minpwrsavecycles
pwrdownexit
RW 0x0
RW 0x0
6
5
4
pwrdownexit
selfrfshexit
RW 0x0
RW 0x0
3
2
1
16
0
External Memory Interface Handbook Volume 3: Reference Material
319
7 Functional Description—HPS Memory Controller
dramtiming4 Fields
Bit
Name
Description
Access
Rese
t
23:2
0
minpwrsavecycles
The minimum number of cycles to stay in a low power state.
This applies to both power down and self-refresh and should
be set to the greater of tPD and tCKESR.
RW
0x0
19:1
0
pwrdownexit
The power down exit cycles, tXPDLL.
RW
0x0
selfrfshexit
The self refresh exit cycles, tXS.
RW
0x0
9:0
7.15.1.1.6 lowpwrtiming
This register controls the behavior of the low power logic in the controller.
Module Instance
Base Address
sdr
Register Address
0xFFC20000
0xFFC25014
Offset: 0x5014
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
Reserved
18
17
16
clkdisablecycles
RW 0x0
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1
0
autopdcycles
RW 0x0
lowpwrtiming Fields
Bit
Name
Description
Access
Rese
t
19:1
6
clkdisablecycles
Set to a the number of clocks after the execution of an selfrefresh to stop the clock. This register is generally set based
on PHY design latency and should generally not be changed.
RW
0x0
15:0
autopdcycles
The number of idle clock cycles after which the controller
should place the memory into power-down mode.
RW
0x0
7.15.1.1.7 dramodt
This register controls which ODT pin asserts with chip select 0 (CS0) assertion and
which ODT pin asserts with chip select 1 (CS1) assertion.
Module Instance
sdr
Base Address
0xFFC20000
External Memory Interface Handbook Volume 3: Reference Material
320
Register Address
0xFFC25018
7 Functional Description—HPS Memory Controller
Offset: 0x5018
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
6
5
4
3
18
17
16
2
1
0
Reserved
15
14
13
12
11
10
9
8
Reserved
7
cfg_read_odt_chip
cfg_write_odt_chip
RW 0x0
RW 0x0
dramodt Fields
Bit
Name
Description
Access
Rese
t
7:4
cfg_read_odt_chip
This register controls which ODT pin is asserted during reads.
Bits[5:4] select the ODT pin that asserts with CS0 and
bits[7:6] select the ODT pin that asserts with CS1. For
example, a value of 0x9 asserts ODT[0] for accesses CS0 and
ODT[1] for accesses with CS1. This field can be set to 0x1 is
there is only one chip select available.
RW
0x0
3:0
cfg_write_odt_chip
This register controls which ODT pin is asserted during writes.
Bits[1:0] select the ODT pin that asserts with CS0 and
bits[3:2] select the ODT pin that asserts with CS1. For
example, a value of 0x9 asserts ODT[0] for accesses CS0 and
ODT[1] for accesses with CS1. This field can be set to 0x1 is
there is only one chip select available.
RW
0x0
7.15.1.1.8 dramaddrw
This register configures the width of the various address fields of the DRAM. The
values specified in this register must match the memory devices being used.
Module Instance
sdr
Base Address
0xFFC20000
Register Address
0xFFC2502C
Offset: 0x502C
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
External Memory Interface Handbook Volume 3: Reference Material
321
7 Functional Description—HPS Memory Controller
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
6
5
4
3
2
1
0
Reserved
15
14
13
12
11
10
9
8
7
csbits
bankbits
rowbits
colbits
RW 0x0
RW 0x0
RW 0x0
RW 0x0
dramaddrw Fields
Name
Bit
Description
Access
Rese
t
15:1
3
csbits
This field defines the number of chip select address bits. Set
this field to 0x0 for single chip select and to 0x1 for two chip
selects.
When this field is set to 0x1, you may use rank interleaved
mode by programming the ctrlcfg.addrorder field to 0x2. If
you are using a single rank memory interface (csbits=0x0),
you may not enable the rank interleaved mode
(ctrlcfg.addrorder must be set less than 0x2).
When this field is set to 0x1 to enable dual ranks, the chip
select (cs) bit of the incoming address is used to determine
which chip select is active. When the chip select bit of the
incoming address is 0, chip select 0 becomes active. When the
chip select bit of the incoming address is 1, chip select 1
becomes active.
RW
0x0
12:1
0
bankbits
The number of bank address bits for the memory devices in
your memory interface.
RW
0x0
9:5
rowbits
The number of row address bits for the memory devices in
your memory interface.
RW
0x0
4:0
colbits
The number of column address bits for the memory devices in
your memory interface.
RW
0x0
7.15.1.1.9 dramifwidth
This register controls the interface width of the SDRAM controller.
Base Address
Module Instance
sdr
Register Address
0xFFC20000
0xFFC25030
Offset: 0x5030
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
6
5
4
3
2
1
0
Reserved
15
14
13
12
11
10
9
8
Reserved
7
ifwidth
RW 0x0
External Memory Interface Handbook Volume 3: Reference Material
322
7 Functional Description—HPS Memory Controller
dramifwidth Fields
Bit
7:0
Name
Description
Access
Rese
t
RW
0x0
This register controls the width of the SDRAM interface,
including any bits used for ECC. For example, for a 32-bit
interface with ECC, program this register to 0x28. The
ctrlwidth register must also be programmed.
ifwidth
7.15.1.1.10 dramsts
This register provides the status of the calibration and ECC logic.
Base Address
Module Instance
sdr
Register Address
0xFFC20000
0xFFC25038
Offset: 0x5038
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
15
14
13
12
11
10
9
8
23
22
21
20
19
18
17
16
6
5
4
3
2
1
0
corrd
rop
dbeer
r
sbeer
r
calfa
il
calsu
ccess
RW
0x0
RW
0x0
RW
0x0
RW
0x0
RW
0x0
Reserved
7
Reserved
dramsts Fields
Name
Bit
Description
Access
Rese
t
4
corrdrop
This bit is set to 1 when any auto-corrections have been
dropped.
RW
0x0
3
dbeerr
This bit is set to 1 when any ECC double bit errors are
detected.
RW
0x0
2
sbeerr
This bit is set to 1 when any ECC single bit errors are
detected.
RW
0x0
1
calfail
This bit is set to 1 when the PHY is unable to calibrate.
RW
0x0
0
calsuccess
This bit will be set to 1 if the PHY was able to successfully
calibrate.
RW
0x0
7.15.1.1.11 dramintr
This register can enable, disable and clear the SDRAM error interrupts.
Module Instance
sdr
Base Address
0xFFC20000
Register Address
0xFFC2503C
External Memory Interface Handbook Volume 3: Reference Material
323
7 Functional Description—HPS Memory Controller
Offset: 0x503C
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
6
5
20
19
18
17
16
Reserved
15
14
13
12
11
10
9
8
7
Reserved
4
3
2
1
0
intrc
lr
corrd
ropma
sk
dbema
sk
sbema
sk
intre
n
RW
0x0
RW
0x0
RW
0x0
RW
0x0
RW
0x0
dramintr Fields
Bit
Name
Description
Access
Rese
t
4
intrclr
Writing to this self-clearing bit clears the interrupt signal.
Writing to this bit also clears the error count and error
address registers: sbecount, dbecount, dropcount,
erraddr, and dropaddr.
RW
0x0
3
corrdropmask
Set this bit to a one to mask interrupts for an ECC correction
write back needing to be dropped. This indicates a burst of
memory errors in a short period of time.
RW
0x0
2
dbemask
Mask the double bit error interrupt.
RW
0x0
1
sbemask
Mask the single bit error interrupt.
RW
0x0
0
intren
Enable the interrupt output.
RW
0x0
7.15.1.1.12 sbecount
This register tracks the single-bit error count.
Module Instance
sdr
Base Address
0xFFC20000
Register Address
0xFFC25040
Offset: 0x5040
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
External Memory Interface Handbook Volume 3: Reference Material
324
7 Functional Description—HPS Memory Controller
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
6
5
4
3
2
1
0
Reserved
15
14
13
12
11
10
9
8
7
Reserved
count
RW 0x0
sbecount Fields
Name
Bit
7:0
Description
Reports the number of single bit errors that have occurred
since the status register counters were last cleared.
count
Access
Rese
t
RW
0x0
7.15.1.1.13 dbecount
This register tracks the double-bit error count.
Module Instance
Base Address
sdr
Register Address
0xFFC20000
0xFFC25044
Offset: 0x5044
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
6
5
4
19
18
17
16
3
2
1
0
Reserved
15
14
13
12
11
10
9
8
7
Reserved
count
RW 0x0
dbecount Fields
Name
Bit
7:0
count
Description
Reports the number of double bit errors that have occurred
since the status register counters were last cleared.
Access
Rese
t
RW
0x0
7.15.1.1.14 erraddr
This register holds the address of the most recent ECC error.
Module Instance
sdr
Base Address
0xFFC20000
Register Address
0xFFC25048
Offset: 0x5048
External Memory Interface Handbook Volume 3: Reference Material
325
7 Functional Description—HPS Memory Controller
Access: RW
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
6
5
4
3
2
1
0
addr
RW 0x0
15
14
13
12
11
10
9
8
7
addr
RW 0x0
erraddr Fields
Name
Bit
31:0
Description
The address of the most recent ECC error.
addr
Access
Rese
t
RW
0x0
Note: For a 32-bit interface, ECC is calculated across a span
of 8 bytes, meaning the error address is a multiple of
8 bytes (4 bytes*2 burst length). To find the byte
address of the word that contains the error, you must
multiply the value in the erraddr register by 8.
7.15.1.1.15 dropcount
This register holds the address of the most recent ECC error.
Module Instance
Base Address
sdr
Register Address
0xFFC20000
0xFFC2504C
Offset: 0x504C
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
6
5
4
3
2
1
0
Reserved
15
14
13
12
11
10
9
8
7
Reserved
corrdropcount
RW 0x0
dropcount Fields
Bit
7:0
Name
corrdropcount
Description
This gives the count of the number of ECC write back
transactions dropped due to the internal FIFO overflowing.
7.15.1.1.16 dropaddr
This register holds the last dropped address.
External Memory Interface Handbook Volume 3: Reference Material
326
Access
Rese
t
RW
0x0
7 Functional Description—HPS Memory Controller
Module Instance
Base Address
sdr
Register Address
0xFFC20000
0xFFC25050
Offset: 0x5050
Access: RW
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
6
5
4
3
2
1
0
corrdropaddr
RW 0x0
15
14
13
12
11
10
9
8
7
corrdropaddr
RW 0x0
dropaddr Fields
Bit
31:0
Name
Description
This register gives the last address which was dropped.
corrdropaddr
Access
Rese
t
RW
0x0
7.15.1.1.17 lowpwreq
This register instructs the controller to put the DRAM into a power down state. Note
that some commands are only valid for certain memory types.
Base Address
Module Instance
sdr
Register Address
0xFFC20000
0xFFC25054
Offset: 0x5054
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
5
4
19
18
17
2
1
16
Reserved
15
14
13
12
11
10
Reserved
9
8
7
6
selfrfshmask
RW 0x0
3
selfr
shreq
deeppwrdnmas
k
RW
0x0
RW 0x0
0
deepp
wrdnr
eq
RW
0x0
External Memory Interface Handbook Volume 3: Reference Material
327
7 Functional Description—HPS Memory Controller
lowpwreq Fields
Bit
5:4
3
2:1
0
Name
Description
Access
Rese
t
selfrfshmask
Write a one to each bit of this field to have a self refresh
request apply to both chips.
RW
0x0
selfrshreq
Write a one to this bit to request the RAM be put into a self
refresh state. This bit is treated as a static value so the RAM
will remain in self-refresh as long as this register bit is set to
a one. This power down mode can be selected for all DRAMs
supported by the controller.
RW
0x0
deeppwrdnmask
Write ones to this register to select which DRAM chip selects
will be powered down. Typical usage is to set both of these
bits when deeppwrdnreq is set but the controller does support
putting a single chip into deep power down and keeping the
other chip running.
RW
0x0
deeppwrdnreq
Write a one to this bit to request a deep power down. This bit
should only be written with LPDDR2 DRAMs, DDR3 DRAMs do
not support deep power down.
RW
0x0
7.15.1.1.18 lowpwrack
This register gives the status of the power down commands requested by the Low
Power Control register.
Module Instance
Base Address
sdr
Register Address
0xFFC20000
0xFFC25058
Offset: 0x5058
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
18
6
5
4
3
2
17
16
Reserved
15
14
13
12
11
10
9
8
7
Reserved
1
0
selfr
fshac
k
deepp
wrdna
ck
RW
0x0
RW
0x0
lowpwrack Fields
Bit
Name
Description
Access
Rese
t
1
selfrfshack
This bit is a one to indicate that the controller is in a selfrefresh state.
RW
0x0
0
deeppwrdnack
This bit is set to a one after a deep power down has been
executed
RW
0x0
External Memory Interface Handbook Volume 3: Reference Material
328
7 Functional Description—HPS Memory Controller
7.15.1.1.19 staticcfg
This register controls configuration values which cannot be updated during active
transfers. First configure the membl and eccn fields and then re-write these fields
while setting the applycfg bit. The applycfg bit is write only.
Module Instance
Base Address
sdr
Register Address
0xFFC20000
0xFFC2505C
Offset: 0x505C
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
6
5
4
3
2
1
0
apply
cfg
useec
casda
ta
Reserved
15
14
13
12
11
10
9
8
7
Reserved
RW
0x0
membl
RW 0x0
RW
0x0
staticcfg Fields
Name
Bit
Description
Access
Rese
t
3
applycfg
Write with this bit set to apply all the settings loaded in SDR
registers to the memory interface. This bit is write-only and
always returns 0 if read.
RW
0x0
2
useeccasdata
This field allows the FPGA ports to directly access the extra
data bits that are normally used to hold the ECC code. The
interface width must be set to 24 or 40 in the dramifwidth
register. If you set this, you must clear the eccen field in the
ctrlcfg register.
RW
0x0
membl
This field specifies the DRAM burst length. The following
encodings set the burst length:
• 0x0= Burst length of 2 clocks
• 0x1= Burst length of 4 clocks
• 0x2= Burst length of 8 clocks
• 0x3= Burst length of 16 clocks
If you program the this field, you must also set the membl
field in the ctrlcfg register.
RW
0x0
1:0
7.15.1.1.20 ctrlwidth
This register controls the width of the physical DRAM interface.
Module Instance
sdr
Base Address
0xFFC20000
Register Address
0xFFC25060
External Memory Interface Handbook Volume 3: Reference Material
329
7 Functional Description—HPS Memory Controller
Offset: 0x5060
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
18
6
5
4
3
2
17
16
1
0
Reserved
15
14
13
12
11
10
9
8
7
Reserved
ctrlwidth
RW 0x0
ctrlwidth Fields
Bit
1:0
Name
ctrlwidth
Description
This field specifies the SDRAM controller interface width:
Value
Access
Rese
t
RW
0x0
Description
0x0
8-bit interface width
0x1
16-bit (no ECC) or 24-bit (ECC enabled) interface
width
0x2
32-bit (no ECC) or 40-bit (ECC enabled) interface
width
Additionally, you must program the dramifwidth register.
7.15.1.1.21 portcfg
Each bit of the autopchen field maps to one of the control ports. If a port executes
mostly sequential memory accesses, the corresponding autopchen bit should be 0. If
the port has highly random accesses, then its autopchen bit should be set to 1.
Module Instance
sdr
Base Address
0xFFC20000
Register Address
0xFFC2507C
Offset: 0x507C
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
External Memory Interface Handbook Volume 3: Reference Material
330
7 Functional Description—HPS Memory Controller
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
Reserved
18
17
16
autopchen
RW 0x0
15
14
13
12
11
10
9
autopchen
8
7
6
5
Reserved
4
3
2
1
0
portprotocol
RW 0x0
RO 0x0
portcfg Fields
Bit
19:10
Name
autopchen
Description
Access
Reset
Auto-Precharge Enable: One bit is assigned to each control
port. For each bit, the encodings are as follows:
RW
0x0
RO
0x0
Value
Description
0x0
The controller requests an automatic precharge
following a bus command completion (close the
row automatically)
0x1
The controller attempts to keep a row open. All
active ports with random dominated operations
should set the autopchen bit to 1.
The bits in this field correspond to the control ports as
follows:
5:0
portprotocol
Bit 9
CPU write
Bit 8
L3 write
Bit 7
CPU read
Bit 6
L3 read
Bit 5
FPGA-to-SDRAM port 5
Bit 4
FPGA-to-SDRAM port 4
Bit 3
FPGA-to-SDRAM port 3
Bit 2
FPGA-to-SDRAM port 2
Bit 1
FPGA-to-SDRAM port 1
Bit 0
FPGA-to-SDRAM port 0
Port Protocol: You can read this field to determine the
protocol configuration of each of the FPGA-to-SDRAM ports.
The bits in this field correspond to the control ports as
follows:
Bit 5
FPGA-to-SDRAM port 5
Bit 4
FPGA-to-SDRAM port 4
Bit 3
FPGA-to-SDRAM port 3
Bit 2
FPGA-to-SDRAM port 2
Bit 1
FPGA-to-SDRAM port 1
continued...
External Memory Interface Handbook Volume 3: Reference Material
331
7 Functional Description—HPS Memory Controller
Bit
Name
Description
Bit 0
Access
Reset
FPGA-to-SDRAM port 0
When you read the corresponding port bit after the FPGA
has been configured, it will have one of the following
values:
Value
Protocol Configuration Type
0x1
AXI (reset value for all ports)
0x0
Avalon-MM
Note: The value in this field is only valid after the fabric
has been configured.
Note: The AXI protocol requires both a read and a write
port. Therefore, you must ensure that AXI ports are
allocated in the pairs (port 0, port 1), (port 2, port
3), and (port 4, port 5).
Aside from the requirement noted above, AXI and AvalonMM ports can be mixed freely.
7.15.1.1.22 fpgaportrst
This register implements functionality to allow the CPU to control when the MPFE will
enable the ports to the FPGA fabric.
Base Address
Module Instance
sdr
Register Address
0xFFC20000
0xFFC25080
Offset: 0x5080
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
6
5
4
3
2
1
0
Reserved
15
14
13
12
11
10
9
8
Reserved
7
portrstn
RW 0x0
fpgaportrst Fields
Name
Bit
13:0
portrstn
Description
Access
Rese
t
This register should be written to with a 1 to enable the
selected FPGA port to exit reset. Writing a bit to a zero will
stretch the port reset until the register is written. Read data
ports are connected to bits 3:0, with read data port 0 at bit 0
to read data port 3 at bit 3. Write data ports 0 to 3 are
mapped to 4 to 7, with write data port 0 connected to bit 4 to
RW
0x0
External Memory Interface Handbook Volume 3: Reference Material
332
7 Functional Description—HPS Memory Controller
Bit
Name
Description
Access
Rese
t
write data port 3 at bit 7. Command ports are connected to
bits 8 to 13, with command port 0 at bit 8 to command port 5
at bit 13. Expected usage would be to set all the bits at the
same time but setting some bits to a zero and others to a one
is supported.
7.15.1.1.23 protportdefault
This register controls the default protection assignment for a port. Ports which have
explicit rules which define regions which are illegal to access should set the bits to
pass by default. Ports which have explicit rules which define legal areas should set the
bit to force all transactions to fail. Leaving this register to all zeros should be used for
systems which do not desire any protection from the memory controller.
Module Instance
Base Address
sdr
Register Address
0xFFC20000
0xFFC2508C
Offset: 0x508C
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
15
14
13
12
11
10
9
8
23
22
21
20
19
18
17
16
6
5
4
3
2
1
0
Reserved
7
Reserved
portdefault
RW 0x0
protportdefault Fields
Bit
9:0
Name
portdefault
Description
Access
Rese
t
Determines the default action for specified transactions. When
a bit is zero, the specified access is allowed by default. When
a bit is one, the specified access is denied by default.
RW
0x0
Bit 9
CPU write
Bit 8
L3 write
Bit 7
CPU read
Bit 6
L3 read
Bit 5
Access to FPGA-to-SDRAM port 5
Bit 4
Access to FPGA-to-SDRAM port 4
Bit 3
Access to FPGA-to-SDRAM port 3
Bit 2
Access to FPGA-to-SDRAM port 2
External Memory Interface Handbook Volume 3: Reference Material
333
7 Functional Description—HPS Memory Controller
Bit
Name
Description
Access
Bit 1
Access to FPGA-to-SDRAM port 1
Bit 0
Access to FPGA-to-SDRAM port 0
Rese
t
7.15.1.1.24 protruleaddr
This register is used to control the memory protection for port 0 transactions. Address
ranges can either be used to allow access to memory regions or disallow access to
memory regions. If TrustZone is being used, access can be enabled for protected
transactions or disabled for unprotected transactions. The default state of this register
is to allow all access. Address values used for protection are only physical addresses.
Module Instance
Base Address
sdr
Register Address
0xFFC20000
0xFFC25090
Offset: 0x5090
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
Reserved
20
19
18
17
16
2
1
0
highaddr
RW 0x0
15
14
13
12
11
10
9
8
7
6
5
highaddr
lowaddr
RW 0x0
RW 0x0
4
3
protruleaddr Fields
Name
Bit
Description
Access
Rese
t
23:1
2
highaddr
Upper 12 bits of the address for a check. Address is compared
to be greater than or equal to the address of a transaction.
Note that since AXI transactions cannot cross a 4K byte
boundary, the transaction start and transaction end address
must also fall within the same 1MByte block pointed to by this
address pointer.
RW
0x0
11:0
lowaddr
Lower 12 bits of the address for a check. Address is compared
to be less than or equal to the address of a transaction. Note
that since AXI transactions cannot cross a 4K byte boundary,
the transaction start and transaction end address must also
fall within the same 1MByte block pointed to by this address
pointer.
RW
0x0
7.15.1.1.25 protruleid
This register configures the AxID for a given protection rule.
External Memory Interface Handbook Volume 3: Reference Material
334
7 Functional Description—HPS Memory Controller
Module Instance
Base Address
sdr
Register Address
0xFFC20000
0xFFC25094
Offset: 0x5094
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
Reserved
19
18
17
16
2
1
0
highid
RW 0x0
15
14
13
12
11
10
9
8
7
6
5
highid
lowid
RW 0x0
RW 0x0
4
3
protruleid Fields
Bit
Name
Description
Access
Rese
t
23:1
2
highid
AxID for the protection rule. Incoming AxID needs to be less
than or equal to this value. For all AxIDs from a port, AxID
high should be programmed to all ones.
RW
0x0
11:0
lowid
AxID for the protection rule. Incoming AxID needs to be
greater than or equal to this value. For all AxIDs from a port,
AxID high should be programmed to all ones.
RW
0x0
7.15.1.1.26 protruledata
This register configures the protection memory characteristics of each protection rule.
Module Instance
sdr
Base Address
0xFFC20000
Register Address
0xFFC25098
Offset: 0x5098
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
External Memory Interface Handbook Volume 3: Reference Material
335
7 Functional Description—HPS Memory Controller
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
6
5
4
3
2
1
0
Reserved
15
14
Reserved
13
12
11
10
9
8
ruler
esult
7
portmask
valid
rule
RW 0x0
RW
0x0
security
RW 0x0
RW
0x0
protruledata Fields
Name
Description
Access
Rese
t
ruleresult
Set this bit to a one to force a protection failure, zero to allow
the access the succeed
RW
0x0
portmask
The bits in this field determine which ports the rule applies to.
If a port's bit is set, the rule applies to that port; if the bit is
clear, the rule does not apply. The bits in this field correspond
to the control ports as follows:
RW
0x0
Bit
13
12:3
Bit 9
CPU write
Bit 8
L3 write
Bit 7
CPU read
Bit 6
L3 read
Bit 5
FPGA-to-SDRAM port 5
Bit 4
FPGA-to-SDRAM port 4
Bit 3
FPGA-to-SDRAM port 3
Bit 2
FPGA-to-SDRAM port 2
Bit 1
FPGA-to-SDRAM port 1
Bit 0
FPGA-to-SDRAM port 0
&
2
1:0
validrule
Set to bit to a one to make a rule valid, set to a zero to
invalidate a rule.
RW
0x0
security
Valid security field encodings are:
RW
0x0
Value
Description
0x0
Rule applies to secure transactions
0x1
Rule applies to non-secure transactions
0x2 or 0x3
Rule applies to secure and non-secure
transactions
7.15.1.1.27 protrulerdwr
This register is used to perform read and write operations to the internal protection
table.
External Memory Interface Handbook Volume 3: Reference Material
336
7 Functional Description—HPS Memory Controller
Module Instance
Base Address
sdr
Register Address
0xFFC20000
0xFFC2509C
Offset: 0x509C
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
6
5
4
3
2
1
0
readr
ule
write
rule
RW
0x0
RW
0x0
Reserved
15
14
13
12
11
10
9
8
7
Reserved
ruleoffset
RW 0x0
protrulerdwr Fields
Bit
Name
Description
Access
Rese
t
6
readrule
Write to this bit to have the memory_prot_data register
loaded with the value from the internal protection table at
offset. Table value will be loaded before a rdy is returned so
read data from the register will be correct for any follow-on
reads to the memory_prot_data register.
RW
0x0
5
writerule
Write to this bit to have the memory_prot_data register to the
table at the offset specified by port_offset. Bit automatically
clears after a single cycle and the write operation is complete.
RW
0x0
ruleoffset
This field defines which of the 20 rules in the protection table
you want to read or write.
RW
0x0
4:0
7.15.1.1.28 mppriority
This register is used to configure the DRAM burst operation scheduling.
Module Instance
sdr
Base Address
0xFFC20000
Register Address
0xFFC250AC
Offset: 0x50AC
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
External Memory Interface Handbook Volume 3: Reference Material
337
7 Functional Description—HPS Memory Controller
Bit Fields
31
30
29
28
27
26
25
24
Reserved
23
22
21
20
19
18
17
16
5
4
3
2
1
0
userpriority
RW 0x0
15
14
13
12
11
10
9
8
7
6
userpriority
RW 0x0
mppriority Fields
Bit
29:0
Name
userpriority
Description
Access
Rese
t
User Priority: This field sets the absolute user priority of each
port, which is represented as a 3-bit value. 0x0 is the lowest
priority and 0x7 is the highest priority. Port 0 is configured by
programming userpriority[2:0], port 1 is configured by
programming userpriority[5:3], port 2 is configured by
programming userpriority[8:6], and so on.
RW
0x0
7.15.1.1.29 remappriority
This register applies another level of port priority after a transaction is placed in the
single port queue.
Module Instance
Base Address
sdr
Register Address
0xFFC20000
0xFFC250E0
Offset: 0x50E0
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
6
5
20
19
18
17
16
4
3
2
1
0
Reserved
15
14
13
12
11
10
9
8
7
Reserved
priorityremap
RW 0x0
remappriority Fields
Bit
7:0
Name
priorityremap
Description
Each bit of this field represents a priority level. If bit N in the
priorityremap field is set, then any port transaction with
absolute user priority of N jumps to the front of the single
port queue and is serviced ahead of any tranactions in the
queue. For example, if bit 5 is set in the priorityremap
field of the remappriority register, then any port
External Memory Interface Handbook Volume 3: Reference Material
338
Access
Rese
t
RW
0x0
7 Functional Description—HPS Memory Controller
Bit
Name
Description
Access
Rese
t
transaction with a userpriority value of 0x5 in the
mppriority register is serviced ahead of any other
transaction already in the single port queue.
7.15.1.1.30 Port Sum of Weight Register Register Descriptions
This register is used to configure the DRAM burst operation scheduling.
Offset: 0xb0
mpweight_0_4
This register is used to configure the DRAM burst operation scheduling.
Module Instance
Base Address
sdr
Register Address
0xFFC20000
0xFFC250B0
Offset: 0x50B0
Access: RW
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
6
5
4
3
2
1
0
staticweight_31_0
RW 0x0
15
14
13
12
11
10
9
8
7
staticweight_31_0
RW 0x0
mpweight_0_4 Fields
Name
Bit
31:0
staticweight_31_0
Description
Set static weight of the port. Each port is programmed with a
5 bit value. Port 0 is bits 4:0, port 1 is bits 9:5, up to port 9
being bits 49:45
Access
Rese
t
RW
0x0
mpweight_1_4
This register is used to configure the DRAM burst operation scheduling.
Module Instance
sdr
Base Address
0xFFC20000
Register Address
0xFFC250B4
Offset: 0x50B4
Access: RW
External Memory Interface Handbook Volume 3: Reference Material
339
7 Functional Description—HPS Memory Controller
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
18
sumofweights_13_0
17
16
staticweight
_49_32
RW 0x0
RW 0x0
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1
0
staticweight_49_32
RW 0x0
mpweight_1_4 Fields
Bit
Name
Description
Access
Rese
t
31:1
8
sumofweights_13_0
Set the sum of static weights for particular user priority. This
register is used as part of the deficit round robin
implementation. It should be set to the sum of the weights for
the ports
RW
0x0
17:0
staticweight_49_32
Set static weight of the port. Each port is programmed with a
5 bit value. Port 0 is bits 4:0, port 1 is bits 9:5, up to port 9
being bits 49:45
RW
0x0
mpweight_2_4
This register is used to configure the DRAM burst operation scheduling.
Module Instance
Base Address
sdr
Register Address
0xFFC20000
0xFFC250B8
Offset: 0x50B8
Access: RW
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
6
5
4
3
2
1
0
sumofweights_45_14
RW 0x0
15
14
13
12
11
10
9
8
7
sumofweights_45_14
RW 0x0
mpweight_2_4 Fields
Name
Bit
31:0
sumofweights_45_14
Description
Access
Rese
t
Set the sum of static weights for particular user priority. This
register is used as part of the deficit round robin
implementation. It should be set to the sum of the weights for
the ports
RW
0x0
mpweight_3_4
This register is used to configure the DRAM burst operation scheduling.
External Memory Interface Handbook Volume 3: Reference Material
340
7 Functional Description—HPS Memory Controller
Module Instance
Base Address
sdr
Register Address
0xFFC20000
0xFFC250BC
Offset: 0x50BC
Access: RW
Important:
To prevent indeterminate system behavior, reserved areas of memory must not be
accessed by software or hardware. Any area of the memory map that is not explicitly
defined as a register space or accessible memory is considered reserved.
Bit Fields
31
30
29
28
27
26
25
24
23
22
21
20
19
18
Reserved
17
16
sumofweights
_63_46
RW 0x0
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1
0
sumofweights_63_46
RW 0x0
mpweight_3_4 Fields
Bit
17:0
Name
sumofweights_63_46
Description
Access
Rese
t
Set the sum of static weights for particular user priority. This
register is used as part of the deficit round robin
implementation. It should be set to the sum of the weights for
the ports
RW
0x0
7.16 Document Revision History
Date
Version
Changes
May 2017
2017.05.08
•
•
October 2016
2016.10.28
Maintenance release
May 2016
2016.05.03
Maintenance release
November 2015
2015.11.02
•
•
•
•
•
•
Added note to first topic.
Rebranded as Intel.
Added information regarding calculation of ECC error byte address
location from erraddr register in "User Notification of ECC Errors"
sectoin
Added information regarding bus response to memory protection
transaction failure in "Memory Protection" section
Clarified "Protection" row in "Fields for Rules in Memory Protection"
table in the "Memory Protection" section
Clarified protruledata.security column in "Rules in Memory Protection
Table for Example Configuration" table in the "Example of Configuration
for TrustZone" section
Added note about double-bit error functionality in "ECC Write Backs"
subsection of "ECC" section
Added the "DDR Calibration" subsection under "DDR PHY" section
continued...
External Memory Interface Handbook Volume 3: Reference Material
341
7 Functional Description—HPS Memory Controller
Date
Version
Changes
May 2015
2015.05.04
•
Added the recommended sequence for writing or reading a rule in the
"Memory Protection" section.
December 2014
2014.12.15
•
•
Added SDRAM Protection Access Flow Diagram to "Memory Protection"
subsection in the "Single-Port Controller Operation" section.
Changed the "SDRAM Multi-Port Scheduling" section to "SDRAM MultiPort Arbitration" and added detailed information on how to use and
program the priority and weighted arbitration scheme.
2014.6.30
•
•
•
•
Added Port Mappings section.
Added SDRAM Controller Memory Options section.
Enhanced Example of Configuration for TrustZone section.
Added SDRAM Controller address map and registers.
December 2013
2013.12.30
•
•
•
Added Generating a Preloader Image for HPS with EMIF section.
Added Debugging HPS SDRAM in the Preloader section.
Enhanced Simulation section.
November 2012
1.1
Added address map and register definitions section.
January 2012
1.0
Initial release.
June 2014
External Memory Interface Handbook Volume 3: Reference Material
342
8 Functional Description—HPC II Controller
8 Functional Description—HPC II Controller
The High Performance Controller II works with the UniPHY-based DDR2, DDR3, and
LPDDR2 interfaces. The controller provides high memory bandwidth, high clock rate
performance, and run-time programmability. The controller can reorder data to reduce
row conflicts and bus turn-around time by grouping reads and writes together,
allowing for efficient traffic patterns and reduced latency.
Note:
The controller described here is the High Performance Controller II (HPC II) with
advanced features for designs generated in the Quartus II software version 11.0 and
later, and the Quartus Prime software. Designs created in earlier versions and
regenerated in version 11.0 and later do not inherit the new advanced features; for
information on HPC II without the version 11.0 and later advanced features, refer to
the External Memory Interface Handbook for Quartus II version 10.1, available in the
External Memory Interfaces literature section on www.altera.com.
Related Links
External Memory Interface Handbook, v10.1
8.1 HPC II Memory Interface Architecture
The memory interface consists of the memory controller logic block, the physical logic
layer (PHY), and their associated interfaces. The following figure shows a high-level
block diagram of the overall external memory interface architecture.
Intel Corporation. All rights reserved. Intel, the Intel logo, Altera, Arria, Cyclone, Enpirion, MAX, Nios, Quartus
and Stratix words and logos are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other
countries. Intel warrants performance of its FPGA and semiconductor products to current specifications in
accordance with Intel's standard warranty, but reserves the right to make changes to any products and services
at any time without notice. Intel assumes no responsibility or liability arising out of the application or use of any
information, product, or service described herein except as expressly agreed to in writing by Intel. Intel
customers are advised to obtain the latest version of device specifications before relying on any published
information and before placing orders for products or services.
*Other names and brands may be claimed as the property of others.
ISO
9001:2008
Registered
8 Functional Description—HPC II Controller
Figure 156. High-Level Diagram of Memory Interface Architecture
Memory Interface IP
External Memory
AFI Interface
Avalon-ST Interface
Avalon-MM or AXI Converter
Data Master
AFI Interface
PHY
Memory Controller
CSR Interface
CSR Master
8.2 HPC II Memory Controller Architecture
The memory controller logic block uses an Avalon Streaming (Avalon-ST) interface as
its native interface, and communicates with the PHY layer by the Altera PHY Interface
(AFI).
The following figure shows a block diagram of the memory controller architecture.
External Memory Interface Handbook Volume 3: Reference Material
344
8 Functional Description—HPC II Controller
Figure 157. Memory Controller Architecture Block Diagram
Memory Controller
Command
Generator
Timing Bank
Pool
Arbiter
Write Data Buffer
AFI Interface to PHY
Avalon-ST Input Interface
Rank Timer
ECC
Read Data Buffer
CSR Interface
Avalon-ST Input Interface
The Avalon-ST interface serves as the entry point to the memory controller, and
provides communication with the requesting data masters.
For information about the Avalon interface, refer to Avalon Interface Specifications.
AXI to Avalon-ST Converter
The HPC II memory controller includes an AXI to Avalon-ST converter for
communication with the AXI protocol. The AXI to Avalon-ST converter provides write
address, write data, write response, read address, and read data channels on the AXI
interface side, and command, write data, and read data channels on the Avalon-ST
interface side.
External Memory Interface Handbook Volume 3: Reference Material
345
8 Functional Description—HPC II Controller
Handshaking
The AXI protocol employs a handshaking process similar to the Avalon-ST protocol,
based on ready and valid signals.
Command Channel Implementation
The AXI interface includes separate read and write channels, while the Avalon-ST
interface has only one command channel. Arbitration of the read and write channels is
based on these policies:
•
Round robin
•
Write priority—write channel has priority over read channel
•
Read priority—read channel has priority over write channel
You can choose an arbitration policy by setting the COMMAND_ARB_TYPE parameter to
one of ROUND_ROBIN, WRITE_PRIORITY, or READ_PRIORITY in the
alt_mem_ddrx_axi_st_converter.v file.
Data Ordering
The AXI specification requires that write data IDs must arrive in the same order as
write address IDs are received. Similarly, read data must be returned in the same
order as its associated read address is received.
Consequently, the AXI to Avalon-ST converter does not support interleaving of write
data; all data must arrive in the same order as its associated write address IDs. On
the read side, the controller returns read data based on the read addresses received.
Burst Types
The AXI to Avalon-ST converter supports the following burst types:
•
Incrementing burst—the address for each transfer is an increment of the previous
transfer address; the increment value depends on the size of the transfer.
•
Wrapping burst—similar to the incrementing burst, but wraps to the lower address
when the burst boundary is reached. The starting address must be aligned to the
size of the transfer. Burst length must be 2, 4, 8, or 16. The burst wrap boundary
= burst size * burst length.
Related Links
Avalon Interface Specifications
8.2.1 Backpressure Support
The write response and read data channels do not support data transfer with
backpressure; consequently, you must assert the ready signal for the write response
and read data channels to 1 to ensure acceptance of data at any time.
External Memory Interface Handbook Volume 3: Reference Material
346
8 Functional Description—HPC II Controller
Figure 158. Data Transfer Without Backpressure
clk
rready
rvalid
rid
rresp
rdata
D0
D1
D2
D3
For information about data transfer with and without backpressure, refer to the Avalon
Interface Specifications.
Related Links
Avalon Interface Specifications
8.2.2 Command Generator
The command generator accepts commands from the front-end Avalon-ST interface
and from local ECC internal logic, and provides those commands to the timing bank
pool.
8.2.3 Timing Bank Pool
The timing bank pool is a parallel queue that works with the arbiter to enable data
reordering. The timing bank pool tracks incoming requests, ensures that all timing
requirements are met and, upon receiving write-data-ready notification from the write
data buffer, passes the requests to the arbiter in an ordered and efficient manner.
8.2.4 Arbiter
The arbiter determines the order in which requests are passed to the memory device.
When the arbiter receives a single request, that request is passed immediately;
however, when multiple requests are received, the arbiter uses arbitration rules to
determine the order in which to pass requests to the memory device.
Arbitration Rules
The arbiter uses the following arbitration rules:
•
If only one master is issuing a request, grant that request immediately.
•
If there are outstanding requests from two or more masters, the arbiter applies
the following tests, in order:
•
Is there a read request? If so, the arbiter grants the read request ahead of any
write requests.
•
If neither of the above conditions apply, the arbiter grants the oldest request first.
External Memory Interface Handbook Volume 3: Reference Material
347
8 Functional Description—HPC II Controller
8.2.5 Rank Timer
The rank timer maintains rank-specific timing information, and performs the following
functions:
•
Ensures that only four activates occur within a specified timing window.
•
Manages the read-to-write and write-to-read bus turnaround time.
•
Manages the time-to-activate delay between different banks.
8.2.6 Read Data Buffer and Write Data Buffer
The read data buffer receives data from the PHY and passes that data through the
input interface to the master. The write data buffer receives write data from the input
interface and passes that data to the PHY, upon approval of the write request.
8.2.7 ECC Block
The error-correcting code (ECC) block comprises an encoder and a decoder-corrector,
which can detect and correct single-bit errors, and detect double-bit errors. The ECC
block can remedy errors resulting from noise or other impairments during data
transmission.
8.2.8 AFI and CSR Interfaces
The AFI interface provides communication between the controller and the physical
layer logic (PHY). The CSR interface provides communication with your system’s
internal control status registers.
For more information about AFI signals, refer to AFI 4.0 Specification in the Functional
Description - UniPHY chapter.
Note:
Unaligned reads and writes on the AFI interface are not supported.
Related Links
•
Avalon Interface Specifications
•
Functional Description—UniPHY on page 13
8.3 HPC II Controller Features
The HPC II memory controller offers a variety of features.
8.3.1 Data Reordering
The controller implements data reordering to maximize efficiency for read and write
commands. The controller can reorder read and write commands as necessary to
mitigate bus turn-around time and reduce conflict between rows.
External Memory Interface Handbook Volume 3: Reference Material
348
8 Functional Description—HPC II Controller
Inter-bank data reordering reorders commands going to different bank addresses.
Commands going to the same bank address are not reordered. This reordering
method implements simple hazard detection on the bank address level.
The controller implements logic to limit the length of time that a command can go
unserved. This logic is known as starvation control. In starvation control, a counter is
incremented for every command served. You can set a starvation limit, to ensure that
a waiting command is served immediately, when the starvation counter reaches the
specified limit.
8.3.2 Pre-emptive Bank Management
Data reordering allows the controller to issue bank-management commands preemptively, based on the patterns of incoming commands. The desired page in memory
can be already open when a command reaches the AFI interface.
8.3.3 Quasi-1T and Quasi-2T
One controller clock cycle equals two memory clock cycles in a half-rate interface, and
to four memory clock cycles in a quarter-rate interface. To fully utilize the command
bandwidth, the controller can operate in Quasi-1T half-rate and Quasi-2T quarter-rate
modes.
In Quasi-1T and Quasi-2T modes, the controller issues two commands on every
controller clock cycle. The controller is constrained to issue a row command on the
first clock phase and a column command on the second clock phase, or vice versa.
Row commands include activate and precharge commands; column commands include
read and write commands.
The controller operates in Quasi-1T in half-rate mode, and in Quasi-2T in quarter-rate
mode; this operation is transparent and has no user settings.
8.3.4 User Autoprecharge Commands
The autoprecharge read and autoprecharge write commands allow you to indicate to
the memory device that this read or write command is the last access to the currently
open row. The memory device automatically closes or autoprecharges the page it is
currently accessing so that the next access to the same bank is quicker.
This command is useful for applications that require fast random accesses.
Since the HPC II controller can reorder transactions for best efficiency, when you
assert the local_autopch_req signal, the controller evaluates the current command
and buffered commands to determine the best autoprecharge operation.
8.3.5 Address and Command Decoding Logic
When the main state machine issues a command to the memory, it asserts a set of
internal signals. The address and command decoding logic turns these signals into
AFI-specific commands and address.
External Memory Interface Handbook Volume 3: Reference Material
349
8 Functional Description—HPC II Controller
The following signals are generated:
•
Clock enable and reset signals: afi_cke, afi_rst_n
•
Command and address signals: afi_cs_n, afi_ba, afi_addr, afi_ras_n,
afi_cas_n, afi_we_n
8.3.6 Low-Power Logic
There are two types of low-power logic: the user-controlled self-refresh logic and
automatic power-down with programmable time-out logic.
User-Controlled Self-Refresh
When you assert the local_self_rfsh_req signal, the controller completes any
currently executing reads and writes, and then interrupts the command queue and
immediately places the memory into self-refresh mode. When the controller places the
memory into self-refresh mode, it responds by asserting an acknowledge signal,
local_self_rfsh_ack. You can leave the memory in self-refresh mode for as long
as you choose.
To bring the memory out of self-refresh mode, you must deassert the request signal,
and the controller responds by deasserting the acknowledge signal when the memory
is no longer in self-refresh mode.
Note:
If a user-controlled refresh request and a system-generated refresh request occur at
the same time, the user-controlled refresh takes priority; the system-generated
refresh is processed only after the user-controlled refresh request is completed.
Automatic Power-Down with Programmable Time-Out
The controller automatically places the memory in power-down mode to save power if
the requested number of idle controller clock cycles is observed in the controller. The
Auto Power Down Cycles parameter on the Controller Settings tab allows you to
specify a range between 1 to 65,535 idle controller clock cycles. The counter for the
programmable time-out starts when there are no user read or write requests in the
command queue. Once the controller places the memory in power-down mode, it
responds by asserting the acknowledge signal, local_power_down_ack.
8.3.7 ODT Generation Logic
The on-die termination (ODT) generation logic generates the necessary ODT signals
for the controller, based on the scheme that Intel recommends.
DDR2 SDRAM
Note:
There is no ODT for reads.
Table 82.
ODT—DDR2 SDRAM Single Slot Single Chip-select Per DIMM (Write)
Write On
ODT Enabled
mem_odt [0]
mem_cs[0]
Note:
There is no ODT for reads.
External Memory Interface Handbook Volume 3: Reference Material
350
8 Functional Description—HPC II Controller
Table 83.
ODT—DDR2 SDRAM Single Slot Dual Chip-select Per DIMM (Write)
Write On
ODT Enabled
mem_odt [0]
mem_cs[0]
mem_cs[1]
Table 84.
mem_odt[1]
ODT—DDR2 SDRAM Dual Slot Single Chip-select Per DIMM (Write)
Write On
ODT Enabled
mem_odt [1]
mem_cs[0]
mem_cs[1]
Table 85.
mem_odt[0]
ODT—DDR2 SDRAM Dual Slot Dual Chip-select Per DIMM (Write)
Write On
ODT Enabled
mem_cs[0]
mem_odt[2]
mem_cs[1]
mem_odt[3]
mem_cs[2]
mem_odt[0]
mem_cs[3]
mem_odt[1]
DDR3 SDRAM
Note:
There is no ODT for reads.
Table 86.
ODT—DDR3 SDRAM Single Slot Single Chip-select Per DIMM (Write)
Write On
ODT Enabled
mem_odt [0]
mem_cs[0]
Note:
There is no ODT for reads.
External Memory Interface Handbook Volume 3: Reference Material
351
8 Functional Description—HPC II Controller
Table 87.
ODT—DDR3 SDRAM Single Slot Dual Chip-select Per DIMM (Write)
Write On
ODT Enabled
mem_odt [0]
mem_cs[0]
mem_cs[1]
Table 88.
mem_odt[1]
ODT—DDR3 SDRAM Dual Slot Single Chip-select Per DIMM (Write)
Write On
ODT Enabled
mem_odt [0] and mem_odt [1]
mem_cs[0]
mem_odt [0] and mem_odt [1]
mem_cs[1]
Table 89.
ODT—DDR3 SDRAM Dual Slot Single Chip-select Per DIMM (Read)
Read On
ODT Enabled
mem_cs[0]
mem_odt[1]
mem_cs[1]
mem_odt[0]
Table 90.
ODT—DDR3 SDRAM Dual Slot Dual Chip-select Per DIMM (Write)
Write On
ODT Enabled
mem_odt [0] and mem_odt [2]
mem_cs[0]
mem_odt [1]and mem_odt [3]
mem_cs[1]
mem_odt [0]and mem_odt [2]
mem_cs[2]
mem_odt [1]and mem_odt [3]
mem_cs[3]
External Memory Interface Handbook Volume 3: Reference Material
352
8 Functional Description—HPC II Controller
Table 91.
ODT—DDR3 SDRAM Dual Slot Dual Rank Per DIMM (Read)
Read On
ODT Enabled
mem_cs[0]
mem_odt[2]
mem_cs[1]
mem_odt[3]
mem_cs[2]
mem_odt[0]
mem_cs[3]
mem_odt[1]
8.3.8 Burst Merging
The burst merging feature improves controller efficiency by merging two burst chop
commands of sequential addresses into one burst length command.
Burst merging is opportunistic and happens when the controller receives commands
faster than it can process (for example, Avalon commands of multiple burst length) or
when the controller temporarily stops processing commands due to Refresh.
The burst merging feature is turned off by default when you generate a controller. If
your traffic exercises patterns that you can merge, you should turn on burst merging,
as follows:
1.
In a text editor, open the <variation_name>_c0.v top-level file for your design.
2. Search for the ENABLE_BURST_MERGE parameter in the .v file.
3. Change the ENABLE_BURST_MERGE value from 0 to 1.
8.3.9 ECC
The ECC logic comprises an encoder and a decoder-corrector, which can detect and
correct single-bit errors, and detect double-bit errors. The ECC logic is available in
multiples of 16, 24, 40, and 72 bits.
Note:
For the hard memory controller with multiport front end available in Arria V and
Cyclone V devices, ECC logic is limited to widths of 24 and 40.
•
The ECC logic has the following features:
•
Has Hamming code ECC logic that encodes every 64, 32, 16, or 8 bits of data into
72, 40, 24, or 16 bits of codeword.
•
Has a latency increase of one clock for both writes and reads.
•
For a 128-bit interface, ECC is generated as one 64-bit data path with 8-bits of
ECC path, plus a second 64-bit data path with 8-bits of ECC path.
•
Detects and corrects all single-bit errors.
•
Detects all double-bit errors.
•
Counts the number of single-bit and double-bit errors.
External Memory Interface Handbook Volume 3: Reference Material
353
8 Functional Description—HPC II Controller
Note:
•
Accepts partial writes, which trigger a read-modify-write cycle, for memory
devices with DM pins.
•
Can inject single-bit and double-bit errors to trigger ECC correction for testing and
debugging purposes.
•
Generates an interrupt signal when an error occurs.
When using ECC, you must initialize your entire memory content to zero before
beginning to write to the memory. If you do not initialize the content to zero, and if
you read from uninitialized memory locations without having first written to them, you
will see junk data which will trigger an ECC interrupt.
When a single-bit or double-bit error occurs, the ECC logic triggers the
ecc_interrupt signal to inform you that an ECC error has occurred. When a singlebit error occurs, the ECC logic reads the error address, and writes back the corrected
data. When a double-bit error occurs, the ECC logic does not do any error correction
but it asserts the avl_rdata_error signal to indicate that the data is incorrect. The
avl_rdata_error signal follows the same timing as the avl_rdata_valid signal.
Enabling autocorrection allows the ECC logic to delay all controller pending activities
until the correction completes. You can disable autocorrection and schedule the
correction manually when the controller is idle to ensure better system efficiency. To
manually correct ECC errors, follow these steps:
1. When an interrupt occurs, read out the SBE_ERROR register. When a single-bit
error occurs, the SBE_ERROR register is equal to one.
2. Read out the ERR_ADDR register.
3.
Correct the single-bit error by issuing a dummy write to the memory address
stored in the ERR_ADDR register. A dummy write is a write request with the
local_be signal zero, that triggers a partial write which is effectively a readmodify-write event. The partial write corrects the data at that address and writes
it back.
8.3.9.1 Partial Writes
The ECC logic supports partial writes.
Along with the address, data, and burst signals, the Avalon-MM interface also supports
a signal vector, local_be, that is responsible for byte-enable. Every bit of this signal
vector represents a byte on the data-bus. Thus, a logic low on any of these bits
instructs the controller not to write to that particular byte, resulting in a partial write.
The ECC code is calculated on all bytes of the data-bus. If any bytes are changed, the
IP core must recalculate the ECC code and write the new code back to the memory.
For partial writes, the ECC logic performs the following steps:
1.
The ECC logic sends a read command to the partial write address.
2.
Upon receiving a return data from the memory for the particular address, the ECC
logic decodes the data, checks for errors, and then merges the corrected or
correct dataword with the incoming information.
3.
The ECC logic issues a write to write back the updated data and the new ECC
code.
External Memory Interface Handbook Volume 3: Reference Material
354
8 Functional Description—HPC II Controller
The following corner cases can occur:
•
A single-bit error during the read phase of the read-modify-write process. In this
case, the IP core corrects the single-bit error first, increments the single-bit error
counter and then performs a partial write to this corrected decoded data word.
•
A double-bit error during the read phase of the read-modify-write process. In this
case, the IP core increments the double-bit error counter and issues an interrupt.
The IP core writes a new write word to the location of the error. The ECC status
register keeps track of the error information.
The following figures show partial write operations for the controller, for full and half
rate configurations, respectively.
Figure 159. Partial Write for the Controller--Full Rate
avl_address
0
1
avl_size
avl_be
avl_wdata
2
X1
XF
01234567 89ABCDEF
mem_dm
mem_dq
67
R
R
R
EF
CD
AB
89
Figure 160. Partial Write for the Controller--Half Rate
avl_address
0
avl_size
1
avl_be
X1
avl_wdata
01234567
mem_dm
mem_dq
67
R
R
R
8.3.9.2 Partial Bursts
DIMMs that do not have the DM pins do not support partial bursts. You must write a
minimum (or multiples) of memory-burst-length-equivalent words to the memory at
the same time.
The following figure shows a partial burst operation for the controller.
External Memory Interface Handbook Volume 3: Reference Material
355
8 Functional Description—HPC II Controller
Figure 161. Partial Burst for Controller
avl_address
0
avl_size
1
avl_be
X1
01234567
avl_wdata
mem_dm
mem_dq
67
45
23
01
8.4 External Interfaces
This section discusses the interfaces between the controller and other external
memory interface components.
8.4.1 Clock and Reset Interface
The clock and reset interface is part of the AFI interface.
The controller can have up to two clock domains, which are synchronous to each
other. The controller operates with a single clock domain when there is no integrated
half-rate bridge, and with two-clock domains when there is an integrated half-rate
bridge. The clocks are provided by UniPHY.
The main controller clock is afi_clk, and the optional half-rate controller clock is
afi_half_clk. The main and half-rate clocks must be synchronous and have a 2:1
frequency ratio. The optional quarter-rate controller clock is afi_quarter_clk,
which must also be synchronous and have a 4:1 frequency ratio.
8.4.2 Avalon-ST Data Slave Interface
The Avalon-ST data slave interface consists of the following Avalon-ST channels, which
together form a single data slave:
•
The command channel, which serves as command and address for both read and
write operations.
•
The write data channel, which carries write data.
•
The read data channel, which carries read data.
Related Links
Avalon Interface Specifications
8.4.3 AXI Data Slave Interface
The AXI data interface consists of the following channels, which communicate with the
Avalon-ST interface through the AXI to Avalon-ST converter:
External Memory Interface Handbook Volume 3: Reference Material
356
8 Functional Description—HPC II Controller
•
The write address channel, which carries address information for write operations.
•
The write data channel, which carries write data.
•
The write response channel, which carries write response data.
•
The read address channel, which carries address information for read operations.
•
The read data channel, which carries read data.
8.4.3.1 Enabling the AXI Interface
This section provides guidance for enabling the AXI interface.
1. To enable the AXI interface, first open in an editor the file appropriate for the
required flow, as indicated below:
•
For synthesis flow: <working_dir>/<variation_name>/
<variation_name>_c0.v
•
For simulation flow: <working_dir>/<variation_name>_sim/
<variation_name>/<variation_name>_c0.v
•
Example design fileset for synthesis: <working_dir>/
<variation_name>_example_design/example_project/
<variation_name>_example/submodules/
<variation_name>_example_if0_c0.v
•
Example design fileset for simulation: <working_dir>/
<variation_name>_example_design/simulation/verilog/submodules/
<variation_name>_example_sim_e0_if0_c0.v
2. Locate and remove the alt_mem_ddrx_mm_st_converter instantiation from
the .v file opened in the preceding step.
3. Instantiate the alt_mem_ddrx_axi_st_converter module into the open .v file.
Refer to the following code fragment as a guide:
module ? # ( parameter
// AXI parameters
AXI_ID_WIDTH = <replace parameter value>,
AXI_ADDR_WIDTH = <replace parameter value>,
AXI_LEN_WIDTH = <replace parameter value>,
AXI_SIZE_WIDTH = <replace parameter value>,
AXI_BURST_WIDTH = <replace parameter value>,
AXI_LOCK_WIDTH = <replace parameter value>,
AXI_CACHE_WIDTH = <replace parameter value>,
AXI_PROT_WIDTH = <replace parameter value>,
AXI_DATA_WIDTH = <replace parameter value>,
AXI_RESP_WIDTH = <replace parameter value>
)
(
// Existing ports
...
// AXI Interface ports
// Write address channel
input wire [AXI_ID_WIDTH - 1 : 0] awid,
input wire [AXI_ADDR_WIDTH - 1 : 0] awaddr,
input wire [AXI_LEN_WIDTH - 1 : 0] awlen,
input wire [AXI_SIZE_WIDTH - 1 : 0] awsize,
input wire [AXI_BURST_WIDTH - 1 : 0] awburst,
input wire [AXI_LOCK_WIDTH - 1 : 0] awlock,
input wire [AXI_CACHE_WIDTH - 1 : 0] awcache,
input wire [AXI_PROT_WIDTH - 1 : 0] awprot,
input wire awvalid,
output wire awready,
// Write data channel
External Memory Interface Handbook Volume 3: Reference Material
357
8 Functional Description—HPC II Controller
input wire [AXI_ID_WIDTH - 1 : 0] wid,
input wire [AXI_DATA_WIDTH - 1 : 0] wdata,
input wire [AXI_DATA_WIDTH / 8 - 1 : 0] wstrb,
input wire wlast,
input wire wvalid,
output wire wready,
// Write response channel
output wire [AXI_ID_WIDTH - 1 : 0] bid,
output wire [AXI_RESP_WIDTH - 1 : 0] bresp,
output wire bvalid,
input wire bready,
// Read address channel
input wire [AXI_ID_WIDTH - 1 : 0] arid,
input wire [AXI_ADDR_WIDTH - 1 : 0] araddr,
input wire [AXI_LEN_WIDTH - 1 : 0] arlen,
input wire [AXI_SIZE_WIDTH - 1 : 0] arsize,
input wire [AXI_BURST_WIDTH - 1 : 0] arburst,
input wire [AXI_LOCK_WIDTH - 1 : 0] arlock,
input wire [AXI_CACHE_WIDTH - 1 : 0] arcache,
input wire [AXI_PROT_WIDTH - 1 : 0] arprot,
input wire arvalid,
output wire arready,
// Read data channel
output wire [AXI_ID_WIDTH - 1 : 0] rid,
output wire [AXI_DATA_WIDTH - 1 : 0] rdata,
output wire [AXI_RESP_WIDTH - 1 : 0] rresp,
output wire rlast,
output wire rvalid,
input wire rready
);
// Existing wire, register declaration and instantiation
...
// AXI interface instantiation
alt_mem_ddrx_axi_st_converter #
(
.AXI_ID_WIDTH (AXI_ID_WIDTH ),
.AXI_ADDR_WIDTH (AXI_ADDR_WIDTH ),
.AXI_LEN_WIDTH (AXI_LEN_WIDTH ),
.AXI_SIZE_WIDTH (AXI_SIZE_WIDTH ),
.AXI_BURST_WIDTH (AXI_BURST_WIDTH ),
.AXI_LOCK_WIDTH (AXI_LOCK_WIDTH ),
.AXI_CACHE_WIDTH (AXI_CACHE_WIDTH ),
.AXI_PROT_WIDTH (AXI_PROT_WIDTH ),
.AXI_DATA_WIDTH (AXI_DATA_WIDTH ),
.AXI_RESP_WIDTH (AXI_RESP_WIDTH ),
.ST_ADDR_WIDTH (ST_ADDR_WIDTH ),
.ST_SIZE_WIDTH (ST_SIZE_WIDTH ),
.ST_ID_WIDTH (ST_ID_WIDTH ),
.ST_DATA_WIDTH (ST_DATA_WIDTH ),
.COMMAND_ARB_TYPE (COMMAND_ARB_TYPE)
)
a0
(
.ctl_clk (afi_clk),
.ctl_reset_n (afi_reset_n),
.awid (awid),
.awaddr (awaddr),
.awlen (awlen),
.awsize (awsize),
.awburst (awburst),
.awlock (awlock),
.awcache (awcache),
.awprot (awprot),
.awvalid (awvalid),
.awready (awready),
.wid (wid),
.wdata (wdata),
.wstrb (wstrb),
.wlast (wlast),
.wvalid (wvalid),
.wready (wready),
External Memory Interface Handbook Volume 3: Reference Material
358
8 Functional Description—HPC II Controller
.bid (bid),
.bresp (bresp),
.bvalid (bvalid),
.bready (bready),
.arid (arid),
.araddr (araddr),
.arlen (arlen),
.arsize (arsize),
.arburst (arburst),
.arlock (arlock),
.arcache (arcache),
.arprot (arprot),
.arvalid (arvalid),
.arready (arready),
.rid (rid),
.rdata (rdata),
.rresp (rresp),
.rlast (rlast),
.rvalid (rvalid),
.rready (rready),
.itf_cmd_ready (ng0_native_st_itf_cmd_ready),
.itf_cmd_valid (a0_native_st_itf_cmd_valid),
.itf_cmd (a0_native_st_itf_cmd),
.itf_cmd_address (a0_native_st_itf_cmd_address),
.itf_cmd_burstlen (a0_native_st_itf_cmd_burstlen),
.itf_cmd_id (a0_native_st_itf_cmd_id),
.itf_cmd_priority (a0_native_st_itf_cmd_priority),
.itf_cmd_autoprecharge (a0_native_st_itf_cmd_autopercharge),
.itf_cmd_multicast (a0_native_st_itf_cmd_multicast),
.itf_wr_data_ready (ng0_native_st_itf_wr_data_ready),
.itf_wr_data_valid (a0_native_st_itf_wr_data_valid),
.itf_wr_data (a0_native_st_itf_wr_data),
.itf_wr_data_byte_en (a0_native_st_itf_wr_data_byte_en),
.itf_wr_data_begin (a0_native_st_itf_wr_data_begin),
.itf_wr_data_last (a0_native_st_itf_wr_data_last),
.itf_wr_data_id (a0_native_st_itf_wr_data_id),
.itf_rd_data_ready (a0_native_st_itf_rd_data_ready),
.itf_rd_data_valid (ng0_native_st_itf_rd_data_valid),
.itf_rd_data (ng0_native_st_itf_rd_data),
.itf_rd_data_error (ng0_native_st_itf_rd_data_error),
.itf_rd_data_begin (ng0_native_st_itf_rd_data_begin),
.itf_rd_data_last (ng0_native_st_itf_rd_data_last),
.itf_rd_data_id (ng0_native_st_itf_rd_data_id)
);
4. Set the required parameters for the AXI interface. The following table lists the
available parameters.
5.
Export the AXI interface to the top-level wrapper, making it accessible to the AXI
master.
6.
To add the AXI interface to the Quartus Prime project:
•
On the Assignments > Settings menu in the Quartus Prime software, open the
File tab.
•
Add the alt_mem_ddrx_axi_st_converter.v file to the project.
External Memory Interface Handbook Volume 3: Reference Material
359
8 Functional Description—HPC II Controller
8.4.3.2 AXI Interface Parameters
Table 92.
AXI Interface Parameters
Parameter Name
Description / Value
AXI_ID_WIDTH
Width of the AXI ID bus. Default value is 4.
AXI_ADDR_WIDTH
Width of the AXI address bus. Must be set according to the Avalon interface address and data bus
width as shown below:AXI_ADDR_WIDTH = LOCAL_ADDR_WIDTH + log2(LOCAL_DATA_WIDTH/
8)LOCAL_ADDR_WIDTH is the memory controller Avalon interface address
width.LOCAL_DATA_WIDTH is the memory controller Avalon data interface width.
AXI_LEN_WIDTH
Width of the AXI length bus. Default value is 8. Should be set to LOCAL_SIZE_WIDTH - 1, where
LOCAL_SIZE_WIDTH is the memory controller Avalon interface burst size width
AXI_SIZE_WIDTH
Width of the AXI size bus. Default value is 3.
AXI_BURST_WIDTH
Width of the AXI burst bus. Default value is 2.
AXI_LOCK_WIDTH
Width of the AXI lock bus. Default value is 2.
AXI_CACHE_WIDTH
Width of the AXI cache bus. Default value is 4.
AXI_PROT_WIDTH
Width of the AXI protection bus. Default value is 3.
AXI_DATA_WIDTH
Width of the AXI data bus. Should be set to match the Avalon interface data bus
width.AXI_DATA_WIDTH = LOCAL_DATA_WIDTH, where LOCAL_DATA_WIDTH is the memory
controller Avalon interface input data width.
AXI_RESP_WIDTH
Width of the AXI response bus. Default value is 2.
ST_ADDR_WIDTH
Width of the Avalon interface address. Must be set to match the Avalon interface address bus
width.ST_ADDR_WIDTH = LOCAL_ADDR_WIDTH, where LOCAL_ADDR_WIDTH is the memory
controller Avalon interface address width.
ST_SIZE_WIDTH
Width of the Avalon interface burst size.ST_SIZE_WIDTH = AXI_LEN_WIDTH + 1
ST_ID_WIDTH
Width of the Avalon interface ID. Default value is 4.ST_ID_WIDTH = AXI_ID_WIDTH
ST_DATA_WIDTH
Width of the Avalon interface data.ST_DATA_WIDTH = AXI_DATA_WIDTH.
COMMAND_ARB_TYPE
Specifies the AXI command arbitration type, as shown:ROUND_ROBIN: arbitrates between read
and write address channel in round robin fashion. Default option.WRITE_PRIORITY: write address
channel has priority if both channels send request simultaneously.READ_PRIORITY: read address
channel has priority if both channels send request simultaneously.
REGISTERED
Setting this parameter to 1 adds an extra register stage in the AXI interface and incurs one extra
clock cycle of latency. Default value is 1.
8.4.3.3 AXI Interface Ports
Table 93.
Name
AXI Interface Ports
Direction
Description
awid
Input
AXI write address channel ID bus.
awaddr
Input
AXI write address channel address bus.
awlen
Input
AXI write address channel length bus.
awsize
Input
AXI write address channel size bus.
awburst
Input
AXI write address channel burst bus.(Interface supports only INCR and WRAP burst types.)
awlock
Input
AXI write address channel lock bus.(Interface does not support this feature.)
awcache
Input
AXI write address channel cache bus.(Interface does not support this feature.)
continued...
External Memory Interface Handbook Volume 3: Reference Material
360
8 Functional Description—HPC II Controller
Name
Direction
Description
awprot
Input
AXI write address channel protection bus.(Interface does not support this feature.)
awvalid
Input
AXI write address channel valid signal.
awready
Output
AXI write address channel ready signal.
wid
Input
AXI write address channel ID bus.
wdata
Input
AXI write address channel data bus.
wstrb
Input
AXI write data channel strobe bus.
wlast
Input
AXI write data channel last burst signal.
wvalid
Input
AXI write data channel valid signal.
wready
Output
AXI write data channel ready signal.
bid
Output
AXI write response channel ID bus.
bresp
Output
AXI write response channel response bus.Response encoding information:‘b00 - OKAY‘b01 Reserved‘b10 - Reserved‘b11 - Reserved
bvalid
Output
AXI write response channel valid signal.
bready
Input
AXI write response channel ready signal.Must be set to 1. Interface does not support back pressure
for write response channel.
arid
Input
AXI read address channel ID bus.
araddr
Input
AXI read address channel address bus.
arlen
Input
AXI read address channel length bus.
arsize
Input
AXI read address channel size bus.
arburst
Input
AXI read address channel burst bus.(Interface supports only INCR and WRAP burst types.)
arlock
Input
AXI read address channel lock bus.(Interface does not support this feature.)
arcache
Input
AXI read address channel cache bus.(Interface does not support this feature.)
arprot
Input
AXI read address channel protection bus.(Interface does not support this feature.)
arvalid
Input
AXI read address channel valid signal.
arready
Output
AXI read address channel ready signal.
rid
Output
AXI read data channel ID bus.
rdata
Output
AXI read data channel data bus.
rresp
Output
AXI read data channel response bus.Response encoding information:‘b00 - OKAY‘b01 Reserved‘b10 - Data error‘b11 - Reserved
rlast
Output
AXI read data channel last burst signal.
rvalid
Output
AXI read data channel valid signal.
rready
Input
AXI read data channel ready signal.Must be set to 1. Interface does not support back pressure for
write response channel.
For information about the AXI specification, refer to the ARM website.
Related Links
www.arm.com
External Memory Interface Handbook Volume 3: Reference Material
361
8 Functional Description—HPC II Controller
8.4.4 Controller-PHY Interface
The interface between the controller and the PHY is part of the AFI interface. The
controller assumes that the PHY performs all necessary calibration processes without
any interaction with the controller.
For more information about AFI signals, refer to AFI 4.0 Specification.
8.4.5 Memory Side-Band Signals
The HPC II controller supports several optional side-band signals.
Self-Refresh (Low Power) Interface
The optional low power self-refresh interface consists of a request signal and an
acknowledgement signal, which you can use to instruct the controller to place the
memory device into self-refresh mode. This interface is clocked by afi_clk.
When you assert the request signal, the controller places the memory device into selfrefresh mode and asserts the acknowledge signal. To bring the memory device out of
self-refresh mode, you deassert the request signal; the controller then deasserts the
acknowledge signal when the memory device is no longer in self-refresh mode.
Note:
For multi-rank designs using the HPC II memory controller, a self-refresh and a userrefresh cannot be made to the same memory chip simultaneously. Also, the selfrefresh ack signal indicates that at least one device has entered self-refresh, but does
not necessarily mean that all devices have entered self-refresh.
User-Controlled Refresh Interface
The optional user-controlled refresh interface consists of a request signal, a chip select
signal, and an acknowledgement signal. This interface provides increased control over
worst-case read latency and enables you to issue refresh bursts during idle periods.
This interface is clocked by afi_clk.
When you assert a refresh request signal to instruct the controller to perform a refresh
operation, that request takes priority over any outstanding read or write requests that
might be in the command queue. In addition to the request signal, you must also
choose the chip to be refreshed by asserting the refresh chip select signal along with
the request signal. If you do not assert the chip select signal with the request signal,
unexpected behavior may result.
The controller attempts to perform a refresh as long as the refresh request signal is
asserted; if you require only one refresh, you should deassert the refresh request
signal after the acknowledgement signal is received. If you maintain the request signal
high after the acknowledgement is sent, it would indicate that further refresh is
required. You should deassert the request signal after the required number of
acknowledgement/refresh is received from the controller. You can issue up to a
maximum of nine consecutive refresh commands.
Note:
For multi-rank designs using the HPC II memory controller, a self-refresh and a userrefresh cannot be made to the same memory chip simultaneously.
External Memory Interface Handbook Volume 3: Reference Material
362
8 Functional Description—HPC II Controller
Configuration and Status Register (CSR) Interface
The controller has a configuration and status register (CSR) interface that allows you
to configure timing parameters, address widths, and the behavior of the controller.
The CSR interface is a 32-bit Avalon-MM slave of fixed address width; if you do not
need this feature, you can disable it to save area.
This interface is clocked by csr_clk, which is the same as afi_clk, and is always
synchronous relative to the main data slave interface.
8.4.6 Controller External Interfaces
The following table lists the controller’s external interfaces.
Table 94.
Summary of Controller External Interfaces
Interface Name
Display Name
Type
Description
Clock and Reset Interface
Clock and Reset
Interface
Clock and Reset Interface
AFI
(1)
Clock and reset generated by
UniPHY to the controller.
Avalon-ST Data Slave Interface
Command Channel
Avalon-ST Data Slave
Interface
Avalon-ST
(2)
Address and command channel for
read and write, single command
single data (SCSD).
Write Data Channel
Avalon-ST Data Slave
Interface
Avalon-ST
(2)
Write Data Channel, single
command multiple data (SCMD).
Read Data Channel
Avalon-ST Data Slave
Interface
Avalon-ST
(2)
Read data channel, SCMD with read
data error response.
Controller-PHY Interface
AFI 4.0
AFI Interface
AFI
(1)
Interface between controller and
PHY.
Memory Side-Band Signals
Self Refresh (Low
Power) Interface
Self Refresh (Low Power)
Interface
Avalon Control & Status
Interface (2)
SDRAM-specific signals to place
memory into low-power mode.
User-Controller Refresh
Interface
User-Controller Refresh
Interface
Avalon Control & Status
Interface (2)
SDRAM-specific signals to request
memory refresh.
Configuration and Status Register (CSR) Interface
CSR
Configuration and Status
Register Interface
Avalon-MM
(2)
Enables on-the-fly configuration of
memory timing parameters,
address widths, and controller
behaviour.
Notes:
1. For information about AFI signals, refer to AFI 4.0 Specification in the Functional Description - UniPHY chapter.
2. For information about Avalon signals, refer to Avalon Interface Specifications.
Related Links
•
Avalon Interface Specifications
•
Functional Description—UniPHY on page 13
External Memory Interface Handbook Volume 3: Reference Material
363
8 Functional Description—HPC II Controller
8.5 Top-Level Signals Description
The top-level signals include clock and reset signals, local interface signals, controller
interface signals, and CSR interface signals.
8.5.1 Clock and Reset Signals
The following table lists the clock and reset signals.
Note:
The suffix _n denotes active low signals.
Table 95.
Clock and Reset Signals
Name
Direction
Description
Input
The asynchronous reset input to the controller. The IP core derives all
other reset signals from resynchronized versions of this signal. This
signal holds the PHY, including the PLL, in reset while low.
Input
The reference clock input to PLL.
global_reset_n
pll_ref_clk
Output
phy_clk
Output
The system clock that the PHY provides to the user. All user inputs to
and outputs from the controller must be synchronous to this clock.
The reset signal that the PHY provides to the user. The IP core asserts
reset_phy_clk_n asynchronously and deasserts synchronously to
phy_clk clock domain.
reset_phy_clk_n
Output
An alternative clock that the PHY provides to the user. This clock always
runs at the same frequency as the external memory interface. In halfrate designs, this clock is twice the frequency of the phy_clk and you
can use it whenever you require a 2x clock. In full-rate designs, the
same PLL output as the phy_clk signal drives this clock.
Output
An alternative clock that the PHY provides to the user. This clock always
runs at half the frequency as the external memory interface. In full-rate
designs, this clock is half the frequency of the phy_clk and you can use
it, for example to clock the user side of a half-rate bridge. In half-rate
designs, or if the Enable Half Rate Bridge option is turned on. The
same PLL output that drives the phy_clk signal drives this clock.
Output
Reference clock to feed to an externally instantiated DLL.
Output
Reset request output that indicates when the PLL outputs are not locked.
Use this signal as a reset request input to any system-level reset
controller you may have. This signal is always low when the PLL is trying
to lock, and so any reset logic using Intel advises you detect a reset
request on a falling edge rather than by level detection.
Input
Edge detect reset input for control by other system reset logic. Assert to
cause a complete reset to the PHY, but not to the PLL that the PHY uses.
aux_full_rate_clk
aux_half_rate_clk
dll_reference_clk
reset_request_n
soft_reset_n
seriesterminationcontrol
Input (for OCT
slave)
Output (for OCT
master)
Required signal for PHY to provide series termination calibration value.
Must be connected to a user-instantiated OCT control block (alt_oct) or
another UniPHY instance that is set to OCT master mode.
Unconnected PHY signal, available for sharing with another PHY.
continued...
External Memory Interface Handbook Volume 3: Reference Material
364
8 Functional Description—HPC II Controller
Name
Direction
parallelterminationcontrol
Input (for OCT
slave)
Output (for OCT
master)
Description
Required signal for PHY to provide series termination calibration value.
Must be connected to a user-instantiated OCT control block (alt_oct) or
another UniPHY instance that is set to OCT master mode.
Unconnected PHY signal, available for sharing with another PHY.
oct_rdn
Input (for OCT
master)
Must connect to calibration resistor tied to GND on the appropriate RDN
pin on the device. (Refer to appropriate device handbook.)
oct_rup
Input (for OCT
master)
Must connect to calibration resistor tied to Vccio on the appropriate RUP
pin on the device. (See appropriate device handbook.)
Input
Allows the use of DLL in another PHY instance in this PHY instance.
Connect the export port on the PHY instance with a DLL to the import
port on the other PHY instance.
Output
Clock for the configuration and status register (CSR) interface, which is
the same as afi_clk and is always synchronous relative to the main
data slave interface.
dqs_delay_ctrl_import
csr_clk
Note:
1. Applies only to the hard memory controller with multiport front end available in Arria V and Cyclone V devices.
8.5.2 Local Interface Signals
The following table lists the controller local interface signals.
Table 96.
Local Interface Signals
Signal Name
avl_addr[]
(1)
Direction
Input
Description
Memory address at which the burst should start. By default, the IP core
maps local address to the bank interleaving scheme. You can change
the ordering via the Local-to-Memory Address Mapping option in the
Controller Settings page.
This signal must remain stable only during the first transaction of a
burst. The constantBurstBehavior property is always false for
UniPHY controllers.
The IP core sizes the width of this bus according to the following
equations:
• Full rate controllers:
For one chip select: width = row bits + bank bits + column bits – 1
For multiple chip selects: width = chip bits* + row bits + bank bits +
column bits – 1
If the bank address is 2 bits wide, row is 13 bits wide and column is 10
bits wide, the local address is 24 bits wide. To map local_address to
bank, row and column address:
avl_addr is 24 bits wide
avl_addr[23:11]= row address[12:0]
avl_addr[10:9] = bank address [1:0]
avl_addr[8:0] = column address[9:1]
The IP core ignores the least significant bit (LSB) of the column address
(multiples of two) on the memory side, because the local data width is
twice that of the memory data bus width.
• Half rate controllers:
continued...
External Memory Interface Handbook Volume 3: Reference Material
365
8 Functional Description—HPC II Controller
Signal Name
Direction
Description
For one chip select: width = row bits + bank bits + column bits – 2
For multiple chip selects: width = chip bits* + row bits + bank bits +
column bits – 2
If the bank address is 2 bits wide, row is 13 bits wide and column is 10
bits wide, the local address is 23 bits wide. To map local_address to
bank, row and column address:
avl_addr is 23 bits wide
avl_addr[22:10] = row address[12:0]
avl_addr[9:8] = bank address [1:0]
avl_addr[7:0] = column address[9:2]
The IP core ignores two LSBs of the column address (multiples of four)
on the memory side, because the local data width is four times that of
the memory data bus width.
• Quarter rate controllers:
For one chip select: width = row bits + bank bits + column bits – 3
For multiple chip selects: width = chip bits* + row bits + bank bits +
column bits – 3
If the bank address is 2 bits wide, row is 13 bits wide and column is 10
bits wide, the local address is 22 bits wide.
(* chip bits is a derived value indicating the number of address bits
necessary to uniquely address every memory rank in the system; this
value is not user configurable.)
• Full-rate hard memory controllers (Arria V and Cyclone V):
For one chip select: width = row bits + bank bits + column bits log2(local avalon data width/memory DQ width)
For multiple chip selects: width = chip bits* + row bits + bank bits +
column bits - log2(local avalon data width/memory data width)
If the local Avalon data width is 32, the memory DQ width is 8, the bank
address is 3 bits wide, the row is 12 bits wide and the column is 8 bits
wide, the local address is 21 bits wide. To map local_address to
bank, row and column address:
avl_addr is 21 bits wide
avl_addr[20:9] = row address[11:0]
avl_addr[8:6] = bank address [2:0]
avl_addr[5:0] = column address[7:2]
The IP core ignores the two least significant bits of the column address
on the memory side because the local data width is four times that of
the memory data bus width (Multi-Port Frontend).
Input
avl_be[]
(2)
Byte enable signal, which you use to mask off individual bytes during
writes. avl_be is active high; mem_dm is active low.
To map avl_wdata and avl_be to mem_dq and mem_dm, consider a
full-rate design with 32-bit avl_wdata and 16-bit mem_dq.
avl_wdata = < 22334455 >< 667788AA >< BBCCDDEE>
avl_be = < 1100 >< 0110 >< 1010 >
These values map to:
Mem_dq = <4455><2233><88AA><6677><DDEE><BBCC>
Mem_dm = <1 1 ><0 0 ><0 1 ><1 0 ><0 1 ><0 1 >
Input
avl_burstbegin
(3)
The Avalon burst begin strobe, which indicates the beginning of an
Avalon burst. Unlike all other Avalon-MM signals, the burst begin signal
is not dependant on avl_ready.
For write transactions, assert this signal at the beginning of each burst
transfer and keep this signal high for one cycle per burst transfer, even
if the slave deasserts avl_ready. The IP core samples this signal at
the rising edge of phy_clk when avl_write_req is asserted. After
continued...
External Memory Interface Handbook Volume 3: Reference Material
366
8 Functional Description—HPC II Controller
Signal Name
Direction
Description
the slave deasserts the avl_ready signal, the master keeps all the
write request signals asserted until avl_ready signal becomes high
again.
For read transactions, assert this signal for one clock cycle when read
request is asserted and avl_addr from which the data should be read
is given to the memory. After the slave deasserts avl_ready
(waitrequest_n in Avalon interface), the master keeps all the read
request signals asserted until avl_ready becomes high again.
Input
Read request signal. You cannot assert read request and write request
signals at the same time. The controller must deassert
reset_phy_clk_n before you can assert avl_read_req.
Input
User-controlled refresh request. If Enable User Auto-Refresh
Controls option is turned on, local_refresh_req becomes available
and you are responsible for issuing sufficient refresh requests to meet
the memory requirements. This option allows complete control over
when refreshes are issued to the memory including grouping together
multiple refresh commands. Refresh requests take priority over read
and write requests, unless the IP core is already processing the
requests.
Input
Controls which chip to issue the user refresh to. The IP core uses this
active high signal with local_refresh_req. This signal is as wide as
the memory chip select. This signal asserts a high value to each bit that
represents the refresh for the corresponding memory chip.
(4)
avl_read_req
local_refresh_req
local_refresh_chip
For example: If local_refresh_chip signal is assigned with a value
of 4’b0101, the controller refreshes the memory chips 0 and 2, and
memory chips 1 and 3 are not refreshed.
avl_size[]
avl_wdata[]
Input
Controls the number of beats in the requested read or write access to
memory, encoded as a binary number. In UniPHY, the IP core supports
Avalon burst lengths from 1 to 1024. The IP core derives the width of
this signal based on the burst count that you specify in the Maximum
Avalon-MM burst length option. With the derived width, you specify a
value ranging from 1 to the local maximum burst count specified.
This signal must remain stable only during the first transaction of a
burst. The constantBurstBehavior property is always false for
UniPHY controllers.
Input
Write data bus. The width of avl_wdata is twice that of the memory
data bus for a full-rate controller, four times the memory data bus for a
half-rate controller, and eight times the memory data bus for a quarterrate controller. If Generate power-of-2 data bus widths for Qsys
and SOPC Builder is enabled, the width is rounded down to the
nearest power of 2.
Input
Write request signal. You cannot assert read request and write request
signal at the same time. The controller must deassert
reset_phy_clk_n before you can assert avl_write_req.
Input
User control of autoprecharge. If you turn on Enable Auto- Precharge
Control, the local_autopch_req signal becomes available and you
can request the controller to issue an autoprecharge write or
autoprecharge read command.
These commands cause the memory to issue a precharge command to
the current bank at the appropriate time without an explicit precharge
command from the controller. This feature is particularly useful if you
know the current read or write is the last one you intend to issue to the
currently open row. The next time you need to use that bank, the
access could be quicker as the controller does not need to precharge the
bank before activating the row you wish to access.
(5)
(6)
avl_write_req
(7)
local_autopch_req
(8)
continued...
External Memory Interface Handbook Volume 3: Reference Material
367
8 Functional Description—HPC II Controller
Signal Name
Direction
Description
Upon receipt of the local_autopch_req signal, the controller
evaluates the pending commands in the command buffer and
determines the most efficient autoprecharge operation to perform,
reordering commands if necessary.
The controller must deassert reset_phy_clk_n before you can assert
local_autopch_req.
Input
local_self_rfsh_chip
Controls which chip to issue the user refresh to. The IP core uses this
active high signal with local_self_rfsh_req. This signal is as wide
as the memory chip select. This signal asserts a high value to each bit
that represents the refresh for the corresponding memory chip.
For example: If local_self_rfsh_chip signal is assigned with a
value of 4’b0101, the controller refreshes the memory chips 0 and 2,
and memory chips 1 and 3 are not refreshed.
Input
User control of the self-refresh feature. If you turn on Enable SelfRefresh Controls, you can request that the controller place the
memory devices into a self-refresh state by asserting this signal. The
controller places the memory in the self-refresh state as soon as it can
without violating the relevant timing parameters and responds by
asserting local_self_rfsh_ack. You can hold the memory in the
self-refresh state by keeping this signal asserted. You can release the
memory from the self-refresh state at any time by deasserting
local_self_rfsh_req and the controller responds by deasserting
local__self_rfsh_ack when it has successfully brought the memory
out of the self-refresh state.
Output
When the memory initialization, training, and calibration are complete,
the PHY sequencer asserts ctrl_usr_mode_rdy to the memory
controller, which then asserts this signal to indicate that the memory
interface is ready for use.
Output
When the memory initialization, training, and calibration completes
successfully, the controller asserts this signal coincident with
local_init_done to indicate the memory interface is ready for use.
Output
When the memory initialization, training, or calibration fails, the
controller asserts this signal to indicate that calibration failed. The
local_init_done signal will not assert when local_cal_fail
asserts.
Output
Read data bus. The width of avl_rdata is twice that of the memory
data bus for a full rate controller; four times the memory data bus for a
half rate controller. If Generate power-of-2 data bus widths for
Qsys and SOPC Builder is enabled, the width is rounded down to the
nearest power of 2.
Output
Asserted if the current read data has an error. This signal is only
available if you turn on Enable Error Detection and Correction
Logic. The controller asserts this signal with the avl_rdata_valid
signal.
If the controller encounters double-bit errors, no correction is made and
the controller asserts this signal.
Output
Read data valid signal. The avl_rdata_valid signal indicates that
valid data is present on the read data bus.
Output
The avl_ready signal indicates that the controller is ready to accept
request signals. If controller asserts the avl_ready signal in the clock
cycle that it asserts a read or write request, the controller accepts that
request. The controller deasserts the avl_ready signal to indicate that
it cannot accept any more requests. The controller can buffer eight read
or write requests, after which the avl_ready signal goes low.
local_self_rfsh_req
local_init_done
local_cal_success
local_cal_fail
avl_rdata[]
(9)
avl_rdata_error
(10)
avl_rdata_valid
(11)
avl_ready
(12)
continued...
External Memory Interface Handbook Volume 3: Reference Material
368
8 Functional Description—HPC II Controller
Signal Name
Direction
Description
The avl_ready signal is deasserted when any of the following are true:
•
•
•
•
write data FIFO register is full.
controller is waiting for write data when in ECC mode.
Refresh request acknowledge, which the controller asserts for one clock
cycle every time it issues a refresh. Even if you do not turn on Enable
User Auto-Refresh Controls, local_refresh_ack still indicates to
the local interface that the controller has just issued a refresh
command.
Output
Self refresh request acknowledge signal. The controller asserts and
deasserts this signal in response to the local_self_rfsh_req signal.
Output
Auto power-down acknowledge signal. The controller asserts this signal
for one clock cycle every time auto power-down is issued.
Output
Interrupt signal from the ECC logic. The controller asserts this signal
when the ECC feature is turned on, and the controller detects an error.
local_self_rfsh_ack
local_power_down_ack
(13)
Timing bank Pool is full.
FIFO register that stores read data from the memory device is
Output
local_refresh_ack
ecc_interrupt
The
The
full.
The
The
Notes to Table:
1. For the hard memory controller with multiport front end available in Arria V and Cyclone V devices, avl_addr becomes a
per port value, avl_addr_#, where # is a numeral from 0–5, based on the number of ports selected in the Controller
tab.
2. For the hard memory controller with multiport front end available in Arria V and Cyclone V devices, avl_be becomes a
per port value, avl_be_#, where # is a numeral from 0–5, based on the number of ports selected in the Controller tab.
3. For the hard memory controller with multiport front end available in Arria V and Cyclone V devices, avl_burstbegin
becomes a per port value, avl_burstbegin_#, where # is a numeral from 0–5, based on the number of ports selected in
the Controller tab.
4. For the hard memory controller with multiport front end available in Arria V and Cyclone V devices, avl_read_req
becomes a per port value, avl_read_req_#, where # is a numeral from 0–5, based on the number of ports selected in
the Controller tab.
5. For the hard memory controller with multiport front end available in Arria V and Cyclone V devices, avl_size becomes a
per port value, avl_size_#, where # is a numeral from 0–5, based on the number of ports selected in the Controller
tab.
6. For the hard memory controller with multiport front end available in Arria V and Cyclone V devices, avl_wdata becomes
a per port value, avl_wdata_#, where # is a numeral from 0–5, based on the number of ports selected in the Controller
tab.
7. For the hard memory controller with multiport front end available in Arria V and Cyclone V devices, avl_write_req
becomes a per port value, avl_write_req_#, where # is a numeral from 0–5, based on the number of ports selected in
the Controller tab.
8. This signal is not applicable to the hard memory controller.
9. For the hard memory controller with multiport front end available in Arria V and Cyclone V devices, avl_rdata becomes
a per port value, avl_rdata_#, where # is a numeral from 0–5, based on the number of ports selected in the Controller
tab.
10.This signal is not applicable to the hard memory controller.
11.For the hard memory controller with multiport front end available in Arria V and Cyclone V devices, avl_rdata_valid
becomes a per port value, avl_rdata_valid_#, where # is a numeral from 0–5, based on the number of ports selected in
the Controller tab.
12.For the hard memory controller with multiport front end available in Arria V and Cyclone V devices, avl_ready becomes
a per port value, avl_ready_#, where # is a numeral from 0–5, based on the number of ports selected in the Controller
tab.
13.This signal is not applicable to the hard memory controller.
8.5.3 Controller Interface Signals
The following table lists the controller interface signals.
External Memory Interface Handbook Volume 3: Reference Material
369
8 Functional Description—HPC II Controller
Table 97.
Interface Signals
Signal Name
Direction
Description
Bidirectional
Memory data bus. This bus is half the width of the local read
and write data busses.
Bidirectional
Memory data strobe signal, which writes data into the
memory device and captures read data into the Intel device.
Bidirectional
Inverted memory data strobe signal, which with the mem_dqs
signal improves signal integrity.
mem_dq[]
mem_dqs[]
mem_dqs_n[]
Output
Clock for the memory device.
Output
Inverted clock for the memory device.
Output
Memory address bus.
Output
Address or command parity signal generated by the PHY and
sent to the DIMM. DDR3 SDRAM only.
Output
Memory bank address bus.
Output
Memory column address strobe signal.
Output
Memory clock enable signals.
Output
Memory chip select signals.
Output
Memory data mask signal, which masks individual bytes
during writes.
Output
Memory on-die termination control signal.
Output
Memory row address strobe signal.
Output
Memory write enable signal.
Output
This signal is not used and should be ignored. The PHY does
not have capture circuity to trigger on the falling edge of
mem_error_out_n. See below for a complete description of
mem_error_out_n, and recommendations.
mem_ck
mem_ck_n
mem_addr[]
mem_ac_parity
(1)
mem_ba[]
mem_cas_n
mem_cke[]
mem_cs_n[]
mem_dm[]
mem_odt
mem_ras_n
mem_we_n
parity_error_n
mem_err_out_n
(1)
(1)
Input
This is an output of registered DIMMs. When an address-andcommand parity error is detected, DDR3 registered DIMMs
assert the mem_err_out_n signal in accordance with the
memory buffer configuration. Unlike ECC on the data bus, the
controller does not automatically correct errors on the
address-and-command bus. You should connect this pin to
continued...
External Memory Interface Handbook Volume 3: Reference Material
370
8 Functional Description—HPC II Controller
Signal Name
Direction
Description
your own falling edge detection circuitry in order to capture
when a parity error occurs. Upon error detection, action may
be taken such as causing a system interrupt, or an
appropriate event.
Note:
1. This signal is for registered DIMMs only.
8.5.4 CSR Interface Signals
The following table lists the CSR interface signals.
Table 98.
CSR Interface Signals
Signal Name
Direction
Description
Input
Register map address.The width of csr_addr is 16 bits.
Input
Byte-enable signal, which you use to mask off individual bytes during
writes. csr_be is active high.
csr_addr[]
csr_be[]
Output
csr_clk
(1)
Clock for the configuration and status register (CSR) interface, which is
the same as afi_clk and is always synchronous relative to the main
data slave interface.
Input
Write data bus. The width of csr_wdata is 32 bits.
Input
Write request signal. You cannot assert csr_write_req and
csr_read_req signals at the same time.
Input
Read request signal. You cannot assert csr_read_req and
csr_write_req signals at the same time.
csr_wdata[]
csr_write_req
csr_read_req
Output
Read data bus. The width of csr_rdata is 32 bits.
Output
Read data valid signal. The csr_rdata_valid signal indicates that
valid data is present on the read data bus.
Output
The csr_waitrequest signal indicates that the HPC II is busy and not
ready to accept request signals. If the csr_waitrequest signal goes
high in the clock cycle when a read or write request is asserted, that
request is not accepted. If the csr_waitrequest signal goes low, the
HPC II is then ready to accept more requests.
csr_rdata[]
csr_rdata_valid
csr_waitrequest
Note to Table:
1. Applies only to the hard memory controller with multiport front end available in Arria V and Cyclone V devices.
8.5.5 Soft Controller Register Map
The soft controller register map allows you to control the soft memory controller
settings.
Note:
Dynamic reconfiguration is not currently supported.
External Memory Interface Handbook Volume 3: Reference Material
371
8 Functional Description—HPC II Controller
The following table lists the register map for the controller.
Table 99.
Soft Controller Register Map
Address
0x100
0x110
0x120
Default
Access
0
Bit
Reserved.
Name
0
—
Reserved for future use.
1
Reserved.
0
—
Reserved for future use.
2
Reserved.
0
—
Reserved for future use.
7:3
Reserved.
0
—
Reserved for future use.
13:8
Reserved.
0
—
Reserved for future use.
30:14
Reserved.
0
—
Reserved for future use.
15:0
AUTO_PD_CYCLES
16
Description
The number of idle clock cycles
after which the controller should
place the memory into powerdown mode. The controller is
considered to be idle if there are
no commands in the command
queue. Setting this register to 0
disables the auto power-down
mode. The default value of this
register depends on the values
set during the generation of the
design.
0x0
Read write
Reserved.
0
—
Reserved for future use.
17
Reserved.
0
—
Reserved for future use.
18
Reserved.
0
—
Reserved for future use.
19
Reserved.
0
—
Reserved for future use.
21:20
ADDR_ORDER
00
Read write
22
Reserved.
0
—
Reserved for future use.
24:23
Reserved.
0
—
Reserved for future use.
30:24
Reserved
0
—
Reserved for future use.
7:0
Column address width
—
Read write
The number of column address
bits for the memory devices in
your memory interface. The
range of legal values is 7-12.
15:8
Row address width
—
Read write
The number of row address bits
for the memory devices in your
memory interface. The range of
legal values is 12-16.
19:16
Bank address width
—
Read write
The number of bank address bits
for the memory devices in your
memory interface. The range of
legal values is 2-3.
23:20
Chip select address width
—
Read write
The number of chip select
address bits for the memory
devices in your memory
interface. The range of legal
00 - Chip, row, bank, column.01
- Chip, bank, row, column.10 reserved for future use.11 Reserved for future use.
continued...
External Memory Interface Handbook Volume 3: Reference Material
372
8 Functional Description—HPC II Controller
Address
Bit
Name
Default
Access
Description
values is 0-2. If there is only one
single chip select in the memory
interface, set this bit to 0.
31:24
Reserved.
0
—
0x121
31:0
Data width representation
(word)
—
Read only
The number of DQS bits in the
memory interface. This bit can be
used to derive the width of the
memory interface by multiplying
this value by the number of DQ
pins per DQS pin (typically 8).
0x122
7:0
Chip select representation
—
Read only
The number of chip select in
binary representation. For
example, a design with 2 chip
selects has the value of
00000011.
31:8
Reserved.
0
—
3:0
tRCD
—
Read write
The activate to read or write a
timing parameter. The range of
legal values is 2-11 cycles.
7:4
tRRD
—
Read write
The activate to activate a timing
parameter. The range of legal
values is 2-8 cycles.
11:8
tRP
—
Read write
The precharge to activate a
timing parameter. The range of
legal values is 2-11 cycles.
15:12
tMRD
—
Read write
The mode register load time
parameter. This value is not used
by the controller, as the controller
derives the correct value from
the memory type setting.
23:16
tRAS
—
Read write
The activate to precharge a
timing parameter. The range of
legal values is 4-29 cycles.
31:24
tRC
—
Read write
The activate to activate a timing
parameter. The range of legal
values is 8-40 cycles.
3:0
tWTR
—
Read write
The write to read a timing
parameter. The range of legal
values is 1-10 cycles.
7:4
tRTP
—
Read write
The read to precharge a timing
parameter. The range of legal
values is 2-8 cycles.
15:8
tFAW
—
Read write
The four-activate window timing
parameter. The range of legal
values is 6-32 cycles.
31:16
Reserved.
0
—
15:0
tREFI
—
Read write
The refresh interval timing
parameter. The range of legal
values is 780-6240 cycles.
23:16
tRFC
—
Read write
The refresh cycle timing
parameter. The range of legal
values is 12-255 cycles.
0x123
0x124
0x125
Reserved for future use.
Reserved for future use.
Reserved for future use.
continued...
External Memory Interface Handbook Volume 3: Reference Material
373
8 Functional Description—HPC II Controller
Address
0x126
0x130
0x131
Bit
Name
Default
Access
Description
31:24
Reserved.
0
—
Reserved for future use.
3:0
Reserved.
0
—
Reserved for future use.
7:4
Reserved.
0
—
Reserved for future use.
11:8
Reserved.
0
—
Reserved for future use.
15:12
Reserved.
0
—
Reserved for future use.
23:16
Burst Length
—
Read write
31:24
Reserved.
0
—
0
ENABLE_ECC
1
Read write
When this bit equals 1, it enables
the generation and checking of
ECC. This bit is only active if ECC
was enabled during IP
parameterization.
1
ENABLE_AUTO_CORR
—
Read write
When this bit equals 1, it enables
auto-correction when a single-bit
error is detected.
2
GEN_SBE
0
Read write
When this bit equals 1, it enables
the deliberate insertion of singlebit errors, bit 0, in the data
written to memory. This bit is
used only for testing purposes.
3
GEN_DBE
0
Read write
When this bit equals 1, it enables
the deliberate insertion of
double-bit errors, bits 0 and 1, in
the data written to memory. This
bit is used only for testing
purposes.
4
ENABLE_INTR
1
Read write
When this bit equals 1, it enables
the interrupt output.
5
MASK_SBE_INTR
0
Read write
When this bit equals 1, it masks
the single-bit error interrupt.
6
MASK_DBE_INTR
0
Read write
When this bit equals 1, it masks
the double-bit error interrupt
7
CLEAR
0
Read write
When this bit equals 1, writing to
this self-clearing bit clears the
interrupt signal, and the error
status and error address
registers.
8
MASK_CORDROP_INTR
0
Read write
When this bit equals 1, the
dropped autocorrection error
interrupt is dropped.
9
Reserved.
0
—
0
SBE_ERROR
0
Read only
Set to 1 when any single-bit
errors occur.
1
DBE_ERROR
0
Read only
Set to 1 when any double-bit
errors occur.
2
CORDROP_ERROR
0
Read only
Value is set to 1 when any
controller-scheduled
autocorrections are dropped.
7:3
Reserved.
0
—
Value must match memory burst
length.
Reserved for future use.
Reserved for future use.
Reserved for future use.
continued...
External Memory Interface Handbook Volume 3: Reference Material
374
8 Functional Description—HPC II Controller
Address
Bit
Name
Default
Access
Description
15:8
SBE_COUNT
0
Read only
Reports the number of single-bit
errors that have occurred since
the status register counters were
last cleared.
23:16
DBE_COUNT
0
Read only
Reports the number of double-bit
errors that have occurred since
the status register counters were
last cleared.
31:24
CORDROP_COUNT
0
Read only
Reports the number of controllerscheduled autocorrections
dropped since the status register
counters were last cleared.
0x132
31:0
ERR_ADDR
0
Read only
The address of the most recent
ECC error. This address is a
memory burst-aligned local
address.
0x133
31:0
CORDROP_ADDR
0
Read only
The address of the most recent
autocorrection that was dropped.
This is a memory burst-aligned
local address.
0x134
0
REORDER_DATA
—
Read write
15:1
Reserved.
0
—
23:16
STARVE_LIMIT
0
Read write
31:24
Reserved.
0
—
Reserved for future use.
Number of commands that can
be served before a starved
command.
Reserved for future use.
8.5.6 Hard Controller Register Map
The hard controller register map allows you to control the hard memory controller
settings.
Note:
Dynamic reconfiguration is not currently supported.
The following table lists the register map for the hard controller.
Table 100.
Hard Controller Register Map
Address
Bit
Name
Default
Access
0x000
3:0
0x001
Description
CFG_CAS_WR_LAT
0
Read/Write
Memory write latency.
4:0
CFG_ADD_LAT
0
Read/Write
Memory additive latency.
0x002
4:0
CFG_TCL
0
Read/Write
Memory read latency.
0x003
3:0
CFG_TRRD
0
Read/Write
The activate to activate
different banks timing
parameter.
0x004
5:0
CFG_TFAW
0
Read/Write
The four-activate window
timing parameter.
0x005
7:0
CFG_TRFC
0
Read/Write
The refresh cycle timing
parameter.
continued...
External Memory Interface Handbook Volume 3: Reference Material
375
8 Functional Description—HPC II Controller
Address
Bit
Name
Default
Access
Description
0x006
12:0
CFG_TREFI
0
Read/Write
The refresh interval timing
parameter.
0x008
3:0
CFG_TREFI
0
Read/Write
The activate to read/write
timing parameter.
0x009
3:0
CFG_TRP
0
Read/Write
The precharge to activate
timing parameter.
0x00A
3:0
CFG_TWR
0
Read/Write
The write recovery timing.
0x00B
3:0
CFG_TWTR
0
Read/Write
The write to read timing
parameter.
0x00C
3:0
CFG_TRTP
0
Read/Write
The read to precharge timing
parameter.
0x00D
4:0
CFG_TRAS
0
Read/Write
The activate to precharge
timing parameter.
0x00E
5:0
CFG_TRC
0
Read/Write
The activate to activate timing
parameter.
0x00F
15:0
CFG_AUTO_PD_CYCLES
0
Read/Write
The number of idle clock cycles
after which the controller
should place the memory into
power-down mode.
0x011
9:0
CFG_SELF_RFSH_EXIT_CYCLES
0
Read/Write
The self-refresh exit cycles.
0x013
9:0
CFG_PDN_EXIT_CYCLES
0
Read/Write
The power down exit cycles.
0x015
3:0
CFG_TMRD
0
Read/Write
Mode register timing
parameter.
0x016
4:0
CFG_COL_ADDR_WIDTH
0
Read/Write
The number of column address
bits for the memory devices in
your memory interface.
0x017
4:0
CFG_ROW_ADDR_WIDTH
0
Read/Write
The number of row address bits
for the memory devices in your
memory interface.
0x018
2:0
CFG_BANK_ADDR_WIDTH
0
Read/Write
The number of bank address
bits for the memory devices in
your memory interface.
0x019
2:0
CFG_CS_ADDR_WIDTH
0
Read/Write
The number of chip select
address bits for the memory
devices in your memory
interface.
0x035
3:0
CFG_TCCD
0
Read/Write
CAS#-to-CAS# command
delay.
0x035
3:0
CFG_WRITE_ODT_CHIP
0
Read/Write
CAS#-to-CAS# command
delay.
0x037
3:0
CFG_READ_ODT_CHIP
0
Read/Write
Read ODT Control.
0x040
2:0
CFG_TYPE
0
Read/Write
Selects memory type.
7:3
CFG_BURST_LENGTH
0
Read/Write
Configures burst length as a
static decimal value.
9:8
CFG_ADDR_ORDER
0
Read/Write
Address order selection.
10
CFG_ENABLE_ECC
0
Read/Write
Enable the generation and
checking of ECC.
continued...
External Memory Interface Handbook Volume 3: Reference Material
376
8 Functional Description—HPC II Controller
Address
Bit
Name
Default
Access
Description
11
CFG_ENABLE_AUTO_CORR
0
Read/Write
Enable auto correction when
single bit error is detected.
12
CFG_GEN_SBE
0
Read/Write
When this bit equals 1, it
enables the deliberate insertion
of single-bit errors, bit 0, in the
data written to memory. This
bit is used only for testing
purposes.
13
CFG_GEN_DBE
0
Read/Write
When this bit equals 1, it
enables the deliberate insertion
of double-bit errors, bits 0 and
1, in the data written to
memory. This bit is used only
for testing purposes.
14
CFG_REORDER_DATA
0
Read/Write
Enable Data Reordering.
15
CFG_USER_RFSH
0
Read/Write
Enable User Refresh.
16
CFG_REGDIMM_ENABLE
0
Read/Write
REG DIMM Configuration.
17
CFG_ENABLE_DQS_TRACKING
0
Read/Write
Enable DQS Tracking.
18
CFG_OUTPUT_REGD
0
Read/Write
Enable Registered Output.
19
CFG_ENABLE_NO_DM
0
Read/Write
No Data Mask Configuration.
20
CFG_ENABLE_ECC_CODE_OVE
RWRITES
0
Read/Write
Enable ECC Code Overwrite in
Double Error Bit Detection.
0x043
7:0
CFG_INTERFACE_WIDTH
0
Read/Write
Memory Interface Width.
0x044
3:0
CFG_DEVICE_WIDTH
0
Read/Write
Memory Device Width.
0x045
0
CFG_CAL_REQ
0
Read/Write
Request re-calibration.
CFG_CLOCK_OFF
0
Read/Write
Disable memory clock.
0
STS_CAL_SUCCESS
0
Read Only
Calibration Success.
1
STS_CAL_FAIL
0
Read Only
Calibration Fail.
2
STS_SBE_ERROR
0
Read Only
Single Bit Error Detected.
3
STS_DBE_ERROR
0
Read Only
Double Bit Error Detected.
4
STS_CORR_DROPPED
0
Read Only
Auto Correction Dropped.
0
CFG_ENABLE_INTR
0
Read/Write
Enable Interrupt
1
CFG_MASK_SBE_INTR
0
Read/Write
Mask Single Bit Error Interrupt.
2
CFG_MASK_DBE_INTR
0
Read/Write
Mask Double Bit Error
Interrupt.
3
CFG_MASK_DBE_INTR
0
Write Clear
Clear Interrupt.
6:1
0x047
0x048
0x049
7:0
STS_SBD_COUNT
0
Read Only
Reports the number of SBE
errors that have occurred since
the status register counters
were last cleared.
0x04A
7:0
STS_DBE_COUNT
0
Read Only
Reports the number of SBE
errors that have occurred since
the status register counters
were last cleared.
continued...
External Memory Interface Handbook Volume 3: Reference Material
377
8 Functional Description—HPC II Controller
Address
Bit
Name
Default
Access
Description
STS_ERR_ADDR
0
Read Only
The address of the most recent
ECC error.
0x04B
31:0
0x04F
0
CFG_MASK_CORR_DROPPED_I
NTR
0
Read/Write
Auto Correction Dropped Count.
0x050
7:0
CFG_MASK_CORR_DROPPED_I
NTR
0
Read Only
Auto Correction Dropped Count.
0x051
31:0
STS_CORR_DROPPED_ADDR
0
Read Only
Auto Correction Dropped
Address.
0x055
5:0
CFG_STARVE_LIMIT
0
Read/Write
Starvation Limit.
0x056
1:0
CFG_MEM_BL
0
Read/Write
Burst Length.
2
CFG_MEM_BL
0
Read/Write
ECC Enable.
0x057
1:0
CFG_MEM_BL
0
Read/Write
Specifies controller interface
width.
0x058
11:0
CMD_PORT_WIDTH
0
Read/Write
Specifies per command port
data width.
0x05A
11:0
CMD_FIFO_MAP
0
Read/Write
Specifies command port to
Write FIFO association.
0x05C
11:0
CFG_CPORT_RFIFO_MAP
0
Read/Write
Specifies command port to
Read FIFO association.
23:12
CFG_RFIFO_CPORT_MAP
0
Read/Write
Port assignment (0 - 5)
associated with each of the N
FIFO.
31:24
CFG_WFIFO_CPORT_MAP
0
Read/Write
Port assignment (0 - 5)
associated with each of the N
FIFO. (con't)
3:0
CFG_WFIFO_CPORT_MAP
0
Read/Write
Port assignment (0 - 5)
associated with each of the N
FIFO.
CFG_CPORT_TYPE
0
Read/Write
Command port type.
2:0
CFG_CLOSE_TO_FULL
0
Read/Write
Indicates when the FIFO has
this many empty entries left.
5:3
CFG_CLOSE_TO_EMPTY
0
Read/Write
Indicates when the FIFO has
this many valid entries left.
11:6
CFG_CLOSE_TO_EMPTY
0
Read/Write
Port works in synchronous
mode.
12
CFG_INC_SYNC
0
Read/Write
Set the number of FF as clock
synchronizer.
5:0
CFG_ENABLE_BONDING
0
Read/Write
Enables bonding for each of the
control ports.
7:6
CFG_DELAY_BONDING
0
Read/Write
Set to the value used for the
bonding input to bonding
output delay.
0x069
5:0
CFG_AUTO_PCH_ENABLE
0
Read/Write
Control auto-precharage
options.
0x06A
17:0
MP_SCHEDULER_PRIORITY
0
Read/Write
Set absolute user priority of the
port
0x06D
29:0
RCFG_ST_WT
0
Read/Write
Set static weight of the port.
0x05D
14:4
0x062
0x067
continued...
External Memory Interface Handbook Volume 3: Reference Material
378
8 Functional Description—HPC II Controller
Address
Bit
Name
Default
Access
Description
31:30
RCFG_SUM_PRI_WT
0
Read/Write
Set the sum of static weights
for particular user priority.
0x06E
31:0
RCFG_SUM_PRI_WT
0
Read/Write
Set the sum of static weights
for particular user priority.
0x06F
29:0
RCFG_SUM_PRI_WT
0
Read/Write
Set the sum of static weights
for particular user priority.
0x0B9
0
CFG_DISABLE_MERGING
0
Read/Write
Set to a one to disable
command merging.
8.6 Sequence of Operations
Various blocks pass information in specific ways in response to write, read, and readmodify-write commands.
Write Command
When a requesting master issues a write command together with write data, the
following events occur:
•
The input interface accepts the write command and the write data.
•
The input interface passes the write command to the command generator and the
write data to the write data buffer.
•
The command generator processes the command and sends it to the timing bank
pool.
•
Once all timing requirements are met and a write-data-ready notification has been
received from the write data buffer, the timing bank pool sends the command to
the arbiter.
•
When rank timing requirements are met, the arbiter grants the command request
from the timing bank pool and passes the write command to the AFI interface.
•
The AFI interface receives the write command from the arbiter and requests the
corresponding write data from the write data buffer.
•
The PHY receives the write command and the write data, through the AFI
interface.
Read Command
When a requesting master issues a read command, the following events occur:
•
The input interface accepts the read command.
•
The input interface passes the read command to the command generator.
•
The command generator processes the command and sends it to the timing bank
pool.
•
Once all timing requirements are met, the timing bank pool sends the command to
the arbiter.
•
When rank timing requirements are met, the arbiter grants the command request
from the timing bank pool and passes the read command to the AFI interface.
•
The AFI interface receives the read command from the arbiter and passes the
command to the PHY.
External Memory Interface Handbook Volume 3: Reference Material
379
8 Functional Description—HPC II Controller
•
The PHY receives the read command through the AFI interface, and returns read
data through the AFI interface.
•
The AFI interface passes the read data from the PHY to the read data buffer.
•
The read data buffer sends the read data to the master through the input
interface.
Read-Modify-Write Command
A read-modify-write command can occur when enabling ECC for partial write, and for
ECC correction commands. When a read-modify-write command is issued, the
following events occur:
•
The command generator issues a read command to the timing bank pool.
•
The timing bank pool and arbiter passes the read command to the PHY through
the AFI interface.
•
The PHY receives the read command, reads data from the memory device, and
returns the read data through the AFI interface.
•
The read data received from the PHY passes to the ECC block.
•
The read data is processed by the write data buffer.
•
When the write data buffer issues a read-modify-write data ready notification to
the command generator, the command generator issues a write command to the
timing bank pool. The arbiter can then issue the write request to the PHY through
the AFI interface.
•
When the PHY receives the write request, it passes the data to the memory
device.
8.7 Document Revision History
Date
Version
Changes
May 2017
2017.05.08
Rebranded as Intel.
October 2016
2016.10.31
Maintenance release.
May 2016
2016.05.02
Maintenance release.
November 2015
2015.11.02
•
•
Changed instances of Quartus II to Quartus Prime.
Added CFG_GEN_SBE and CFG_GEN_DBE to Hard Controller Register Map
table.
May 2015
2015.05.04
Maintenance release.
December 2014
2014.12.15
•
•
Renamed Controller Register Map to Soft Controller Register Map.
Added Hard Controller Register Map.
August 2014
2014.08.15
•
Added "asynchronous" to descriptions of mp_cmd_reset_n_#_reset_n,
mp_rfifo_reset_n_#_reset_n, and mp_wfifo_reset_n_#_reset_n
signals in the MPFE Signals table.
Added Reset description to Hard Memory Controller section.
Added full-rate hard memory controller information for Arria V and
Cyclone V to description of avl_addr[] in the Local Interface Signals
table.
•
•
•
Reworded avl_burstbegin description in the Local Interface Signals
table.
continued...
External Memory Interface Handbook Volume 3: Reference Material
380
8 Functional Description—HPC II Controller
Date
December 2013
Version
2013.12.16
Changes
•
•
•
•
•
•
•
•
Removed references to ALTMEMPHY.
Removed references to SOPC Builder.
Removed Half-Rate Bridge information.
Modified Burst Merging description.
Expanded description of avl_ready in Local Interface Signals table.
Added descriptions of local_cal_success and local_cal_fail to Local
Interface Signals table.
Modified description of avl_size in Local Interface Signals table.
Added guidance to initialize memory before use.
November 2012
2.1
•
•
•
•
Added Controller Register Map information.
Added Burst Merging information.
Updated User-Controlled Refresh Interface information.
Changed chapter number from 4 to 5.
June 2012
2.0
•
•
Added LPDDR2 support.
Added Feedback icon.
November 2011
1.1
•
•
•
•
Revised Figure 5–1.
Added AXI to Avalon-ST Converter information.
Added AXI Data Slave Interface information.
Added Half-Rate Bridge information.
External Memory Interface Handbook Volume 3: Reference Material
381
9 Functional Description—QDR II Controller
9 Functional Description—QDR II Controller
The QDR II, QDR II+, and QDR II+ Xtreme controller translates memory requests
from the Avalon Memory-Mapped (Avalon-MM) interface to AFI, while satisfying timing
requirements imposed by the memory configuration. QDR II, QDR II+, and QDR II+
Xtreme SRAM has unidirectional data buses, therefore read and write operations are
highly independent of each other and each has its own interface and state machine.
9.1 Block Description
The following figure shows a block diagram of the QDR II, QDR II+, and QDR II+
Xtreme SRAM controller architecture.
Figure 162. QDR II, QDR II+, and QDR II+ Xtreme SRAM Controller Architecture Block
Diagram
Controller
Write
Data
FIFO
AFI
Command
Issuing
FSM
Avalon-MM Slave
Write Interface
Avalon-MM Slave
Write Interface
Avalon-MM Slave
Read Interface
9.1.1 Avalon-MM Slave Read and Write Interfaces
The read and write blocks accept read and write requests, respectively, from the
Avalon-MM interface. Each block has a simple state machine that represents the state
of the command and address registers, which stores the command and address when
a request arrives.
The read data passes through without the controller registering it, as the PHY takes
care of read latency. The write data goes through a pipeline stage to delay for a fixed
number of cycles as specified by the write latency. In the full-rate burst length of four
controller, the write data is also multiplexed into a burst of 2, which is then
multiplexed again in the PHY to become a burst of 4 in DDR.
Intel Corporation. All rights reserved. Intel, the Intel logo, Altera, Arria, Cyclone, Enpirion, MAX, Nios, Quartus
and Stratix words and logos are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other
countries. Intel warrants performance of its FPGA and semiconductor products to current specifications in
accordance with Intel's standard warranty, but reserves the right to make changes to any products and services
at any time without notice. Intel assumes no responsibility or liability arising out of the application or use of any
information, product, or service described herein except as expressly agreed to in writing by Intel. Intel
customers are advised to obtain the latest version of device specifications before relying on any published
information and before placing orders for products or services.
*Other names and brands may be claimed as the property of others.
ISO
9001:2008
Registered
9 Functional Description—QDR II Controller
The user interface to the controller has separate read and write Avalon-MM interfaces
because reads and writes are independent of each other in the memory device. The
separate channels give efficient use of available bandwidth.
9.1.2 Command Issuing FSM
The command-issuing finite-state machine (FSM) has two states: INIT and
INIT_COMPLETE. In the INIT_COMPLE TE state, commands are issued immediately
as requests arrive using combinational logic and do not require state transitions.
9.1.3 AFI
The QDR II, QDR II+, and QDR II+ Xtreme controller communicates with the PHY
using the AFI interface.
In the full-rate burst-length-of-two configuration, the controller can issue both read
and write commands in the same clock cycle. In the memory device, both commands
are clocked on the positive edge, but the read address is clocked on the positive edge,
while the write address is clocked on the negative edge. Care must be taken on how
these signals are ordered in the AFI.
For the half-rate burst-length-of-four configuration, the controller also issues both
read and write commands, but the AFI width is doubled to fill two memory clocks per
controller clock. Because the controller issues only one write command and one read
command per controller clock, the AFI read and write signals corresponding to the
other memory cycle are tied to no operation (NOP).
For the full-rate burst-length-of-four configuration, the controller alternates between
issuing read and write commands every clock cycle. The memory device requires two
clock cycles to complete the burst-length-of-four operation and requires an
interleaving of read and write commands.
For information on the AFI, refer to AFI 4.0 Specification in chapter 1, Functional
Description - UniPHY.
Related Links
•
AFI 4.0 Specification on page 120
The Altera PHY interface (AFI) 4.0 defines communication between the
controller and physical layer (PHY) in the external memory interface IP.
•
Functional Description—UniPHY on page 13
9.2 Avalon-MM and Memory Data Width
The following table lists the data width ratio between the memory interface and the
Avalon-MM interface.
The half-rate controller does not support burst-of-2 devices because it under-uses the
available memory bandwidth. Regardless of full or half-rate decision and the device
burst length, the Avalon-MM interface must supply all the data for the entire memory
burst in a single clock cycle. Therefore the Avalon-MM data width of the full-rate
controller with burst-of-4 devices is four times the memory data width. For widthexpanded configurations, the data width is further multiplied by the expansion factor
(not shown in table 5-1 and 5-2).
External Memory Interface Handbook Volume 3: Reference Material
383
9 Functional Description—QDR II Controller
Table 101.
Data Width Ratio
Memory Burst Length
QDR II 2-word burst
Half-Rate Designs
Full-Rate Designs
No Support
2:1
QDR II, QDR II+, and QDR II+
Extreme 4-word burst
4:1
9.3 Signal Descriptions
The following tables lists the signals of the controller’s Avalon-MM slave interface.
Table 102.
Avalon-MM Slave Read Signals
UniPHY Signal Name
Arria 10 Signal Name
Width (UniPHY)
1
avl_r_ready
amm_ready_0
avl_r_read_req
amm_read_0
avl_r_addr
amm_address_0
avl_r_rdata_valid
amm_readdatavalid_0
avl_r_rdata
amm_readdata_0
avl_r_size
amm_burstcount_0
Width (Arria
10)
1
Direction
Out
waitrequest_n
1
1
In
read
15–25
17–23
In
address
1
1
Out
readdatavalid
9, 18, 36 (or 8,
16, 32 if powerof-2-bus is
enabled in the
controller)
9, 18, 36 (or 8,
16, 32 if powerof-2-bus is
enabled in the
controller)
log_2(MAX_BURS
T_SIZE) + 1
ceil(log2(CTRL_Q
DR2_AVL_MAX_B
URST_COUNT+1))
Out
readdata
In
burstcount
Note:
To obtain the actual signal width, you must multiply the widths in the above table by
the data width ratio and the width expansion ratio.
Table 103.
Avalon-MM Slave Write Signals
UniPHY Signal Name
Arria 10 Signal Name
Width (UniPHY)
1
avl_w_ready
amm_ready_1
avl_w_write_req
amm_write_1
avl_w_addr
amm_address_1
Avalon-MM Signal
Type
Width (Arria
10)
1
Direction
Avalon-MM Signal
Type
Out
waitrequest_n
1
1
In
write
15–25
17–23
In
address
continued...
External Memory Interface Handbook Volume 3: Reference Material
384
9 Functional Description—QDR II Controller
UniPHY Signal Name
Arria 10 Signal Name
avl_w_wdata
amm_writedata_1
avl_w_be
amm_byteenable_1
avl_w_size
amm_burstcount_1
Note:
Width (UniPHY)
Width (Arria
10)
Direction
9, 18, 36 (or 8,
16, 32 if powerof-2-bus is
enabled in the
controller)
9, 18, 36 (or 8,
16, 32 if powerof-2-bus is
enabled in the
controller)
In
1,2,4
1,2,4
In
Avalon-MM Signal
Type
writedata
byteenable
log_2(MAX_BURS
T_SIZE) + 1
ceil(log2(CTRL_Q
DR2_AVL_MAX_B
URST_COUNT+1))
In
burstcount
To obtain the actual signal width, you must multiply the widths in the above table by
the data width ratio and the width expansion ratio.
Related Links
Avalon Interface Specifications
9.4 Document Revision History
Date
Version
Changes
May 2017
2017.05.08
Rebranded as Intel.
October 2016
2016.10.31
Maintenance release.
May 2016
2016.05.02
Maintenance release.
November 2015
2015.11.02
Maintenance release.
May 2015
2015.05.04
Maintenance release.
December 2014
2014.12.15
Maintenance release.
August 2014
2014.08.15
•
•
•
•
Added QDR II+ Xtreme throughout the chapter.
Added full-rate, burst-length-of-four information to AFI section.
Revised Avalon-MM Slave Read Signals table to include Arria 10
information.
Revised Avalon-MM Slave Write Signals table to include Arria 10
information.
December 2013
2013.12.16
Removed references to SOPC Builder.
November 2012
3.3
Changed chapter number from 5 to 6.
June 2012
3.2
Added Feedback icon.
November 2011
3.1
Harvested Controller chapter from 11.0 QDR II and QDR II+ SRAM Controller
with UniPHY User Guide.
External Memory Interface Handbook Volume 3: Reference Material
385
10 Functional Description—QDR-IV Controller
10 Functional Description—QDR-IV Controller
The QDR-IV controller translates memory requests from the Avalon Memory-Mapped
(Avalon-MM) interface to AFI, while satisfying timing requirements imposed by the
memory configuration. QDR-IV has two independent bidirectional data ports, each of
which can perform read and write operations on each memory clock cycle, subject to
bank address and bus turnaround restrictions. To maximize data bus utilization, the
QDR-IV controller provides eight separate Avalon interfaces, one for each QDR-IV data
port and time slot (at quarter rate).
10.1 Block Description
The following figure shows a block diagram of the QDR-IV controller architecture.
Figure 163. QDR-IV Controller Architecture Block Diagram
QDR-IV Controller
Avalon-MM Slave 0
.
..
Avalon-MM Slave 7
Avalon
FSM
Queue
Scheduler
Address
Parity
Address/Write
data bus
inversion
AFI
AFI to
AVL
Note:
Read data bus
inversion
To achieve maximum performance for a QDR-IV interface on an Arria 10 device, read
data bus inversion must be enabled.
10.1.1 Avalon-MM Slave Read and Write Interfaces
To maximize available bandwidth utilization, the QDR-IV external memory controller
provides eight separate bidirectional Avalon interfaces—one channel for each of the
two QDR-IV data ports in each of the four memory time slots. At the memory device,
the PHY ensures that port A commands are issued on the rising edge of the clock, and
port B commands are issued on the falling edge of the clock.
The Avalon finite state machine (FSM) implements the standard Avalon-MM interface
for each channel. New commands arrive in a queue, before being sent to the PHY by
the scheduler. The scheduler ensures that bank policy and bus turnaround restrictions
are met. The controller handles address/data bus inversion, if you have enabled those
features.
Intel Corporation. All rights reserved. Intel, the Intel logo, Altera, Arria, Cyclone, Enpirion, MAX, Nios, Quartus
and Stratix words and logos are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other
countries. Intel warrants performance of its FPGA and semiconductor products to current specifications in
accordance with Intel's standard warranty, but reserves the right to make changes to any products and services
at any time without notice. Intel assumes no responsibility or liability arising out of the application or use of any
information, product, or service described herein except as expressly agreed to in writing by Intel. Intel
customers are advised to obtain the latest version of device specifications before relying on any published
information and before placing orders for products or services.
*Other names and brands may be claimed as the property of others.
ISO
9001:2008
Registered
10 Functional Description—QDR-IV Controller
The controller schedules commands in a simple round-robin fashion, always
maintaining the time slot relationship shown in the following table. At each AFI cycle,
the scheduler attempts to issue up to eight commands—one from each channel. If a
particular command cannot be issued due to bank policy or bus turnaround violations,
only commands from the preceding channels are issued. The remaining commands are
issued in the following one or more cycles, depending on whether further violations
occur. Read data passes through without the controller registering it, after AFI to
Avalon bus conversion. The PHY implements read latency requirements.
Table 104.
Avalon-MM to Memory Time Slot and Port Mapping
Avalon-MM Slave Interface
Memory Time Slot
Memory Port
0
0
A
1
0
B
2
1
A
3
1
B
4
2
A
5
2
B
6
3
A
7
3
B
10.1.2 AFI
The QDR-IV controller communicates with the PHY using the AFI interface.
The controller supports a quarter-rate burst-length-of-two configuration, and can issue
up to eight read and/or write commands per controller clock cycle. The controller may
have to reduce the actual number of commands issued to avoid banking and bus
turnaround violations. If you use QDR-IV devices with banked operation, you cannot
access the same bank in the same memory clock cycle. Write latency is much shorter
than read latency, therefore you must be careful not to place write data at the same
time that read data is driven on the bus, on the same port. In addition, when
switching between read and write operations, a delay may be required to avoid signal
reflection and allow enough time for dynamic on-chip termination (OCT) to work. If
necessary, the scheduler in the controller can delay the issuing of commands to avoid
these issues.
For information on the AFI, refer to AFI 4.0 Specification in Functional Description Arria 10 EMIF IP.
Related Links
•
AFI 4.0 Specification on page 120
The Altera PHY interface (AFI) 4.0 defines communication between the
controller and physical layer (PHY) in the external memory interface IP.
•
Functional Description—Intel Arria 10 EMIF IP on page 144
Intel Arria 10 devices can interface with external memory devices clocking at
frequencies of up to 1.3 GHz. The external memory interface IP component for
Arria 10 devices provides a single parameter editor for creating external
memory interfaces, regardless of memory protocol.
External Memory Interface Handbook Volume 3: Reference Material
387
10 Functional Description—QDR-IV Controller
10.2 Avalon-MM and Memory Data Width
The following table lists the data width ratio between the memory interface and the
Avalon-MM interface.
QDR-IV memory devices use a burst length of 2. Because the Avalon-MM interface
must supply all of the data for the entire memory burst in a single controller clock
cycle, the Avalon-MM data width of the controller is double the memory data width.
For width-expanded configurations, the data width is further multiplied by the
expansion factor.
Table 105.
Data Width Ratio
Memory Burst Length
Controller
2
2:1
10.3 Signal Descriptions
The following tables lists the signals of the controller’s Avalon-MM slave interface.
Table 106.
Avalon-MM Slave Interface Signals
Signal Name
Width
1
Direction
Out
amm_ready_i
waitrequest_n
1
In
amm_read_i
read
1
In
amm_write_i
write
21-25
In
amm_address_i
address
1
Out
amm_readdatavalid_i
readdatavalid
18, 36, 72
Out
amm_readdata_i
readdata
18, 36, 72
In
amm_writedata_i
amm_burstcount_i
Avalon-MM Signal Type
writedata
ceil(log2(CTRL_QDR4_AVL_MAX_BUR
ST_COUNT+1))
1
In
burstcount
Out
emif_usr_clk
clk
1
Out
emif_usr_reset_n
reset_n
1
global_reset_n
External Memory Interface Handbook Volume 3: Reference Material
388
In
reset_n
10 Functional Description—QDR-IV Controller
Note:
1.
In the above table, the _i suffix represents the Avalon-MM interface index, which
has the range of 0-7.
2. To obtain the actual signal width for the data ports (readdata, writedata), you
must multiply the widths in the above table by the data width ratio and the width
expansion ratio.
3. •
emif_usr_clk: User clock domain.
•
emif_usr_reset_n: Reset for the clock domain. Asynchronous assertion and
synchronous deassertion (to the emif_usr_clk clock signal).
•
global_reset_n: Asynchronous reset causes the memory interface to be
reset and recalibrated.
Related Links
Avalon Interface Specifications
10.4 Document Revision History
Date
Version
Changes
May 2017
2017.05.08
Rebranded as Intel.
October 2016
2016.10.31
Maintenance release.
May 2016
2016.05.02
Maintenance release.
November 2015
2015.11.02
Maintenance release.
May 2015
2015.05.04
Maintenance release.
December 2014
2014.12.15
Initial Release.
External Memory Interface Handbook Volume 3: Reference Material
389
11 Functional Description—RLDRAM II Controller
11 Functional Description—RLDRAM II Controller
The RLDRAM II controller translates memory requests from the Avalon MemoryMapped (Avalon-MM) interface to AFI, while satisfying timing requirements imposed
by the memory configurations.
11.1 Block Description
The following figure shows a block diagram of the RLDRAM II controller architecture.
Figure 164. RLDRAM II Controller Architecture Block Diagram
Controller
Write
Data
FIFO
AFI
Command
Issuing
FSM
Avalon-MM Slave
Write Interface
Avalon-MM Slave
Write Interface
Avalon-MM Slave
Read Interface
11.1.1 Avalon-MM Slave Interface
The Avalon-MM slave interface accepts read and write requests. A simple state
machine represents the state of the command and address registers, which stores the
command and address when a request arrives.
Intel Corporation. All rights reserved. Intel, the Intel logo, Altera, Arria, Cyclone, Enpirion, MAX, Nios, Quartus
and Stratix words and logos are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other
countries. Intel warrants performance of its FPGA and semiconductor products to current specifications in
accordance with Intel's standard warranty, but reserves the right to make changes to any products and services
at any time without notice. Intel assumes no responsibility or liability arising out of the application or use of any
information, product, or service described herein except as expressly agreed to in writing by Intel. Intel
customers are advised to obtain the latest version of device specifications before relying on any published
information and before placing orders for products or services.
*Other names and brands may be claimed as the property of others.
ISO
9001:2008
Registered
11 Functional Description—RLDRAM II Controller
The Avalon-MM slave interface decomposes the Avalon-MM address to the memory
bank, column, and row addresses. The IP automatically maps the bank address to the
LSB of the Avalon address vector.
The Avalon-MM slave interface includes a burst adaptor, which has two parts:
•
The first part is a read and write request combiner that groups requests to
sequential addresses into the native memory burst. Given that the second request
arrives within the read and write latency window of the first request, the controller
can combine and satisfy both requests with a single memory transaction.
•
The second part is the burst divider in the front end of the Avalon-MM interface,
which breaks long Avalon bursts into individual requests of sequential addresses,
which then pass to the controller state machine.
11.1.2 Write Data FIFO Buffer
The write data FIFO buffer accepts write data from the Avalon-MM interface. The AFI
controls the subsequent consumption of the FIFO buffer write data.
11.1.3 Command Issuing FSM
The command issuing finite-state machine (FSM) has three states.
The controller is in the INIT state when the PHY initializes the memory. Upon
receiving the afi_cal_success signal, the state transitions to INIT_COMPLETE. If
the calibration fails, afi_cal_fail is asserted and the state transitions to
INIT_FAIL. The PHY receives commands only in the INIT_COMPLETE state.
When a refresh request arrives at the state machine at the same time as a read or
write request, the refresh request takes precedence. The read or write request waits
until there are no more refresh requests, and is issued immediately if timing
requirements are met.
11.1.4 Refresh Timer
With automatic refresh, the refresh timer periodically issues refresh requests to the
command issuing FSM. The refresh interval can be set at generation.
11.1.5 Timer Module
The timer module contains one DQ timer and eight bank timers (one per bank). The
DQ timer tracks how often read and write requests can be issued, to avoid bus
contention. The bank timers track the cycle time (tRC).
The 8-bit wide output bus of the bank timer indicates to the command issuing FSM
whether each bank can be issued a read, write, or refresh command.
11.1.6 AFI
The RLDRAM II controller communicates with the PHY using the AFI interface. For
information on the AFI, refer to AFI 4.0 Specification in Functional Description UniPHY.
External Memory Interface Handbook Volume 3: Reference Material
391
11 Functional Description—RLDRAM II Controller
Related Links
•
AFI 4.0 Specification on page 120
The Altera PHY interface (AFI) 4.0 defines communication between the
controller and physical layer (PHY) in the external memory interface IP.
•
Functional Description—UniPHY on page 13
11.2 User-Controlled Features
The General Settings tab of the parameter editor contains several features which are
disabled by default. You can enable these features as required for your external
memory interface.
11.2.1 Error Detection Parity
The error detection feature asserts an error signal if it detects any corrupted data
during the read process.
The error detection parity protection feature creates a simple parity encoder block
which processes all read and write data. For every 8 bits of write data, a parity bit is
generated and concatenated to the data before it is written to the memory. During the
subsequent read operation, the parity bit is checked against the data bits to ensure
data integrity.
When you enable the error detection parity protection feature, the local data width is
reduced by one. For example, a nine-bit memory interface will present eight bits of
data to the controller interface.
You can enable error detection parity protection in the Controller Settings section of
the General Settings tab of the parameter editor.
11.2.2 User-Controlled Refresh
The user-controlled refresh feature allows you to take control of the refresh process
that the controller normally performs automatically. You can control when refresh
requests occur, and, if there are multiple memory devices, you control which bank
receives the refresh signal.
When you enable this feature, you disable auto-refresh, and assume responsibility for
maintaining the necessary average periodic refresh rate. You can enable usercontrolled refresh in the Controller Settings section of the General Settings tab of
the parameter editor.
11.3 Avalon-MM and Memory Data Width
The following table lists the data width ratio between the memory interface and the
Avalon-MM interface. The half-rate controller does not support burst-of-2 devices
because it under-uses the available memory bandwidth.
External Memory Interface Handbook Volume 3: Reference Material
392
11 Functional Description—RLDRAM II Controller
Table 107.
Data Width Ratio
Memory Burst Length
2-word
Half-Rate Designs
Full-Rate Designs
No Support
4-word
2:1
4:1
8-word
11.4 Signal Descriptions
The following table lists the signals of the controller’s Avalon-MM slave interface.
For information on the AFI signals, refer to AFI 4.0 Specification in Functional
Description - UniPHY.
Table 108.
Avalon-MM Slave Signals
Signal
avl_size
Width
1 to 11
1
Direction
In
Avalon-MM Signal Type
burstcount
Out
avl_ready
Description
—
—
waitrequest_n
1
In
avl_read_req
—
read
1
In
avl_write_req
—
write
 25
In
avl_addr
—
address
1
Out
avl_rdata_valid
—
readdatavalid
18, 36, 72, 144
Out
avl_rdata
—
readdata
18, 36, 72, 144
avl_wdata
In
—
writedata
Note:
If you are using Qsys, the data width of the Avalon-MM interface is restricted to
powers of two. Non-power-of-two data widths are available with the IP Catalog.
Note:
The RLDRAM II controller does not support the byteenable signal. If the RLDRAM II
controller is used with the Avalon-MM Efficiency Monitor and Protocol Checker, data
corruption can occur if the byteenable signal on the efficiency monitor is used. For
example, this can occur if using the JTAG Avalon-MM Master component to drive the
efficiency monitor.
Related Links
•
AFI 4.0 Specification on page 120
The Altera PHY interface (AFI) 4.0 defines communication between the
controller and physical layer (PHY) in the external memory interface IP.
External Memory Interface Handbook Volume 3: Reference Material
393
11 Functional Description—RLDRAM II Controller
•
Functional Description—UniPHY on page 13
11.5 Document Revision History
Date
Version
Changes
May 2017
2017.05.08
Rebranded as Intel.
October 2016
2016.10.31
Maintenance release.
May 2016
2016.05.02
Maintenance release.
November 2015
2015.11.02
Maintenance release.
May 2015
2015.05.04
Maintenance release.
December 2014
2014.12.15
Maintenance release.
August 2014
2014.08.15
Removed occurrence of MegaWizard Plug-In Manager.
December 2013
2013.12.16
Removed references to SOPC Builder.
November 2012
3.3
Changed chapter number from 6 to 7.
June 2012
3.2
Added Feedback icon.
November 2011
3.1
Harvested Controller chapter from 11.0 RLDRAM II Controller with UniPHY IP
User Guide.
External Memory Interface Handbook Volume 3: Reference Material
394
12 Functional Description—RLDRAM 3 PHY-Only IP
12 Functional Description—RLDRAM 3 PHY-Only IP
The RLDRAM 3 PHY-only IP works with a customer-supplied memory controller to
translate memory requests from user logic to RLDRAM 3 memory devices, while
satisfying timing requirements imposed by the memory configurations.
12.1 Block Description
The RLDRAM 3 UniPHY-based IP is a PHY-only offering which you can use with a thirdparty controller or a controller that you develop yourself. The following figure shows a
block diagram of the RLDRAM 3 system architecture.
Figure 165. RLDRAM 3 System Architecture
FPGA
User
Logic
Avalon-MM
Controller
AFI
(Custom or
third-party)
PHY
(RLDRAM III
UniPHY)
RLDRAM III
Memory
Device
12.2 Features
The RLDRAM 3 UniPHY-based IP supports features available from major RLDRAM 3
device vendors at speeds of up to 800 MHz.
The following list summarizes key features of the RLDRAM 3 UniPHY-based IP:
•
support for Arria V GZ and Stratix V devices
•
standard AFI interface between the PHY and the memory controller
•
quarter-rate and half-rate AFI interface
•
maximum frequency of 533 MHz for half-rate operation and 800 MHz for quarterrate operation
•
burst length of 2, 4, or 8
•
x18 and x36 memory organization
•
common I/O device support
•
nonmultiplexed addressing
•
multibank write and refresh protocol (programmable through mode register)
•
optional use of data mask pins
Intel Corporation. All rights reserved. Intel, the Intel logo, Altera, Arria, Cyclone, Enpirion, MAX, Nios, Quartus
and Stratix words and logos are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other
countries. Intel warrants performance of its FPGA and semiconductor products to current specifications in
accordance with Intel's standard warranty, but reserves the right to make changes to any products and services
at any time without notice. Intel assumes no responsibility or liability arising out of the application or use of any
information, product, or service described herein except as expressly agreed to in writing by Intel. Intel
customers are advised to obtain the latest version of device specifications before relying on any published
information and before placing orders for products or services.
*Other names and brands may be claimed as the property of others.
ISO
9001:2008
Registered
12 Functional Description—RLDRAM 3 PHY-Only IP
12.3 RLDRAM 3 AFI Protocol
The RLDRAM 3 UniPHY-based IP communicates with the memory controller using an
AFI interface that follows the AFI 4.0 specification. To maximize bus utilization
efficiency, the RLDRAM 3 UniPHY-based IP can issue multiple memory read/write
operations within a single AFI cycle.
The following figure illustrates AFI bus activity when a quarter-rate controller issues
four consecutive burst-length 2 read requests.
Figure 166. AFI Bus Activity for Quarter-Rate Controller Issuing Four Burst-Length 2
Read Requests
Read Latency
afi_clk
afi_cs_n[3:0]
4’b0000
4’b1111
afi_ref_n[3:0]
4’b1111
afi_we_n[3:0]
4’b1111
afi_rdata_en_full[3:0]
4’b1111
4’b0000
afi_rdata_valid[3:0]
4’b1111
afi_rdata[...]
Data
The controller does not have to begin a read or write command using channel 0 of the
AFI bus. The flexibility afforded by being able to begin a command on any bus channel
can facilitate command scheduling in the memory controller.
The following figure illustrates the AFI bus activity when a quarter-rate controller
issues a single burst-length 4 read command to the memory on channel 1 of the AFI
bus.
Figure 167. AFI Bus Activity for Quarter-Rate Controller Issuing One Burst-Length 4 Read
Request
Read Latency
afi_clk
afi_cs_n[3:0]
4’b1101
4’b1111
afi_ref_n[3:0]
4’b1111
afi_we_n[3:0]
4’b1111
afi_rdata_en_full[3:0]
4’b0110
afi_rdata_valid[3:0]
afi_rdata[...]
External Memory Interface Handbook Volume 3: Reference Material
396
4’b0000
4’b0110
Data
12 Functional Description—RLDRAM 3 PHY-Only IP
Note:
For information on the AFI, refer to AFI 4.0 Specification in Functional Description UniPHY.
Related Links
•
AFI 4.0 Specification on page 120
The Altera PHY interface (AFI) 4.0 defines communication between the
controller and physical layer (PHY) in the external memory interface IP.
•
Functional Description—UniPHY on page 13
12.4 RLDRAM 3 Controller with Arria 10 EMIF Interfaces
The following table lists the RLDRAM 3 signals available for each interface when using
Arria 10 EMIF IP.
Signal
Interface Type
Description
pll_ref_clk interface
pll_ref_clock
Clock input
Clock input to the PLL inside EMIF.
Clock output
AFI clock output from the PLL inside
EMIF.
Clock output
AFI clock output from the PLL inside
EMIF running at half speed.
Conduit
Interface signal between the PHY and
the memory device.
Conduit
Memory interface status signal.
afi_clk_interface
afi_clock
afi_half_clk_interface
afi_half_clock
Memory interface
mem_a
mem_ba
mem_ck
mem_ck_n
mem_cs_n
mem_dk
mem_dk_n
mem_dm
mem_dq
mem_qk
mem_qk_n
mem_ref_n
mem_we_n
mem_reset_n
Status interface
local_init_done
local_cal_success
continued...
External Memory Interface Handbook Volume 3: Reference Material
397
12 Functional Description—RLDRAM 3 PHY-Only IP
Signal
Interface Type
Description
local_cal_fail
local_cal_request
oct interface
oct_rzqin
Conduit
OCT reference resistor pins for RZQ.
Avalon-MM slave
Altera PHY interface (AFI) signal
between the PHY and the memory
controller.
afi_interface
afi_addr
afi_ba
afi_cs_n
afi_we_n
afi_ref_n
afi_wdata_valid
afi_wdata
afi_dm
afi_rdata
afi_rdata_en_full
afi_rdata_valid
afi_rst_n
afi_cal_success
afi_cal_fail
afi_wlat
afi_rlat
12.5 Document Revision History
Date
Version
Changes
May 2017
2017.05.08
Rebranded as Intel.
October 2016
2016.10.31
Maintenance release.
May 2016
2016.05.02
Maintenance release.
November 2015
2015.11.02
Maintenance release.
May 2015
2015.05.04
Maintenance release.
December 2014
2014.12.15
Maintenance release.
August 2014
2014.08.15
Added RLDRAM 3 Controller with Arria 10 EMIF Interfaces.
December 2013
2013.12.16
Maintenance release.
November 2012
1.0
Initial release.
External Memory Interface Handbook Volume 3: Reference Material
398
13 Functional Description—Example Designs
13 Functional Description—Example Designs
Generation of your external memory interface IP creates two independent example
designs. These example designs illustrate how to instantiate and connect the memory
interface for both synthesis and simulation flows.
13.1 Arria 10 EMIF IP Example Designs Quick Start Guide
A new interface and more automated design example flow is available for Arria 10
external memory interfaces.
The Example Designs tab is available in the parameter editor when you specify an
Arria 10 target device. This tab allows you to select from a list of presets for Intel
FPGA development kits and target interface protocols. All tabs are automatically
parameterized with appropriate values, based on the preset that you select. You can
specify that the system create directories for simulation and synthesis file sets, and
generate the file sets automatically.
You can generate an example design specifically for an Intel FPGA development kit, or
for any EMIF IP that you generate.
Figure 168. Using the Example Design
Design Example
Design
Example
Generation
Compilation
(Simulator)
Functional
Simulation
Compilation
(Quartus Prime)
Timing Analysis
(Quartus Prime)
Hardware
Testing
13.1.1 Typical Example Design Workflow
When you generate your EMIF IP, the system creates an example design that includes
a traffic generator and external memory interface controller.
Intel Corporation. All rights reserved. Intel, the Intel logo, Altera, Arria, Cyclone, Enpirion, MAX, Nios, Quartus
and Stratix words and logos are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other
countries. Intel warrants performance of its FPGA and semiconductor products to current specifications in
accordance with Intel's standard warranty, but reserves the right to make changes to any products and services
at any time without notice. Intel assumes no responsibility or liability arising out of the application or use of any
information, product, or service described herein except as expressly agreed to in writing by Intel. Intel
customers are advised to obtain the latest version of device specifications before relying on any published
information and before placing orders for products or services.
*Other names and brands may be claimed as the property of others.
ISO
9001:2008
Registered
13 Functional Description—Example Designs
Figure 169. Example Design for Synthesis
Synthesis Example Design
Avalon-MM
Traffic Generator
AFI
Controller
PHY
Memory
IP
Figure 170. Example Design for Simulation
Simulation Example Design
Abstract instance of the synthesis example design
Traffic Generator
Avalon-MM
Controller
AFI
PHY
Memory
Memory
Model
IP
Status
Checker
Workflow Overview
Figure 171. Generating a Simulation or Synthesis File set
Start the
Parameter Editor
Select Family
and Device
Specify IP
Parameters
Check Simulation or Synthesis
under Example Design Files
Generate
Example Design
If you select Simulation or Synthesis under Example Design Files on the
Example Designs tab, the system creates a complete simulation file set or a
complete synthesis file set, in accordance with your selection.
13.1.2 Example Designs Interface Tab
If you have selected an Arria 10 device, the parameter editor includes an Example
Designs tab which allows you to parameterize and generate your example designs.
External Memory Interface Handbook Volume 3: Reference Material
400
13 Functional Description—Example Designs
Figure 172. Example Designs Tab in the External Memory Interfaces Parameter Editor
Available Example Designs Section
The Select design pulldown allows you to select the desired example design. At
present, EMIF Example Design is the only available choice, and is selected by
default.
Example Design Files Section
The example design supports generation, simulation, and Quartus Prime compilation
flows for any selected device. The Simulation and Synthesis checkboxes in this
section allow you to specify whether to generate a simulation file set, a synthesis file
set, or both. Check Simulation to use the example design for simulation, or check
Synthesis to use the example design for compilation and hardware. The system
creates the specified file sets when you click the Generate Example Design button.
Note:
If you don't select the Simulation or Synthesis checkbox, the destination directory
will contain Qsys design files, which are not compilable by Quartus Prime directly, but
can be viewed or edited under Qsys.
•
To create a compilable project, you must run the quartus_sh -t
make_qii_design.tcl script in the destination directory.
•
To create a simulation project, you must run the quartus_sh -t
make_sim_design.tcl script in the destination directory.
Generated HDL Format Section
The Simulation HDL format pulldown allows you to specify the target format for the
generated design. At present, only Verilog filesets are supported.
External Memory Interface Handbook Volume 3: Reference Material
401
13 Functional Description—Example Designs
Target Development Kit Section
The Select board pulldown in this section applies the appropriate development kit pin
assignments to the example design.
•
This setting is available only when you turn on the Synthesis checkbox in the
Example Design Files section.
•
This setting must match the applied development kit present, or else an error
message appears.
If the value None appears in the Select board pulldown, it indicates that the current
parameter selections do not match any development kit configurations. You may apply
a development kit-specific IP and related parameter settings by selecting one of the
presets from the preset library. When you apply a preset, the current IP and other
parameter settings are set to match the selected preset. If you want to save your
current settings, you should do so before you select a preset. If you do select a preset
without saving your prior settings, you can always save the new preset settings under
a different name
If you want to generate the example design for use on your own board, set Select
board to None, generate the example design, and then add pin location constraints.
Note:
In the 15.1 release, programming file (.sof/.pof) generation is supported for Arria
10 engineering samples only. If you would like to learn more about support for Arria
10 production devices in 15.1, contact your Intel FPGA representative or use the
support link on www.altera.com.
13.1.3 Development Kit Preset Workflow
Arria 10 users can take advantage of development kit presets to create an example
design that is automatically configured and ready for hardware testing or simulation.
The use of presets greatly simplifies the workflow and reduces the steps necessary to
have an EMIF example design working on an Arria 10 development kit.
Figure 173. Generating an Example Design for Use With an Arria 10 Development Kit
Start the
Parameter Editor
Select Arria 10
Device
Select Traffic Generator 2.0
on Diagnostics tab
(if desired)
Apply a
Development Kit Preset
Set Parameters on
Example Designs tab
External Memory Interface Handbook Volume 3: Reference Material
402
Generate
Example Design
Compile
the Design
13 Functional Description—Example Designs
Follow these steps to generate the example design:
1.
Under Memory Interfaces and Controllers (Tools > Qsys>IP
Catalog>Library>Memory Interfaces and Controllers), select Arria
10 External Memory Interfaces. The parameter editor appears.
2.
Specify a top-level name and the folder for your custom IP variation, and specify
an Arria 10 device. Click OK.
3.
Select the appropriate preset for your development kit from the Presets Library.
Click Apply.
4. By default, the example design will be generated with the traditional Traffic
Generator. If you would rather use the new EMIF Configurable Traffic Generator
2.0, select Use configurable Avalon traffic generator 2.0 on the Diagnostics
tab of the parameter editor.
5.
On the Example Designs tab in the parameter editor, make the following
settings:
a.
Under Available Example Designs, only EMIF Example Design is
available, and is selected by default.
b.
Under Example Design Files, select Simulation or Synthesis to have the
system create a simulation file set or synthesis file set.
c.
Under Generated HDL Format, only Verilog is available, and is selected by
default.
d.
Under Target Development Kit, select the desired Arria 10 development kit.
The development kit that you select here must match the development kit
preset that you selected in step 1, or else an error message appears.
Note: The Target Development Kit section is available only if you have
specified Synthesis under Example Design Files.
6. Click the Generate Example Design button at the top of the parameter editor.
The software generates all files necessary to run simulations and hardware tests
on the selected development kit.
Related Links
Support Center Web Page on www.altera.com
13.1.4 Compiling and Simulating the Design
This section explains how to simulate your EMIF example design.
Figure 174. Procedure for Simulating the Design
Run
Simulation Script
Change to
Simulation Directory
Analyze
Results
1.
Change to the simulation directory.
2.
Run the simulation script for the simulator of your choice. Refer to the table below.
3.
Analyze the results.
External Memory Interface Handbook Volume 3: Reference Material
403
13 Functional Description—Example Designs
Table 109.
Steps to Run Simulation
Simulator
Riviera-PRO
Working Directory
Instructions
<variation_name>_example_de
sign\sim\ aldec
To simulate the example design using the Riviera-PRO simulator,
follow these steps:
a. At the Linux* or Windows* shell command prompt, change
directory to: <variation_name>_example_design\sim\
aldec.
b. Execute the rivierapro_setup.tcl script by typing the
following command at the Linux or Windows command
prompt: vsim -do rivierapro_setup.tcl .
c. To compile and elaborate the design after the script loads,
type: ld_debug.
d. Type run -all to run the simulation.
e. A successful simulation ends with the following message, “--SIMULATION PASSED ---“
NCSim
<variation_name>_example_de
sign\sim\cadence\
To simulate the example design using the NCSim simulator,
follow these steps:
a. At the Linux shell command prompt, change directory to
<variation_name>_example_design\sim\cadence\.
b. Run the simulation by typing the following at the command
prompt: sh ncsim_setup.sh.
c. A successful simulation ends with the following message, “--SIMULATION PASSED ---“
ModelSim*
<variation_name>_example_de
sign\sim\mentor\
To simulate the example design using the Modelsim simulator,
follow these steps:
a. At the Linux or Windows shell command prompt, change
directory to <variation_name>_example_design\sim
\mentor\.
b. Perform one of the following:
• Execute the msim_setup.tcl script by typing the
following command at the Linux or Windows command
prompt: vsim -do msim_setup.tcl.
•
Type the following command at the ModelSim command
prompt: do msim_setup.tcl.
c. A successful simulation ends with the following message, “--SIMULATION PASSED ---“
VCS
<variation_name>_example_de
sign\sim\synopsys\vcsmx\
To simulate the example design using the VCS simulator, follow
these steps:
a. At the Linux shell command prompt, change directory to
<variation_name>_example_design\sim\synopsys
\vcsmx\.
b. Run the simulation by typing the following at the command
prompt: sh vcsmx_setup.sh.
c. A successful simulation ends with the following message, “--SIMULATION PASSED ---“
13.1.5 Compiling and Testing the Design in Hardware
This section explains how to compile and test your EMIF example design in hardware.
Figure 175. Procedure for Testing in Hardware
Arrange to Control
and Monitor Pins
Compile
the Design
External Memory Interface Handbook Volume 3: Reference Material
404
Configure Board
with the SOF file
Issue Reset and
Monitor for Success
13 Functional Description—Example Designs
Follow these steps to compile and test the design in hardware:
1.
2.
Before compiling a design, select a method for controlling and monitoring the
following pins:
•
global_reset_reset_n
•
<variation_name>_status_local_cal_fail
•
<variation_name>_status_local_cal_success
•
<variation_name>_tg_0_traffic_gen_fail
•
<variation_name>_tg_0_traffic_gen_pass
•
<variation_name>_tg_0_traffic_gen_timeout
•
Additional pins that may exist if the example design contains more than one
traffic generator.
Assign pins by one of the following means, as appropriate for your situation:
•
If you have selected a development kit preset, the above pins are
automatically assigned to switches and LEDs on the board. See the
documentation accompanying your specific development kit for details.
•
If you are using a non-development kit board, assign the above pins to
appropriate locations on your board.
•
Alternatively, the above pins can be controlled and monitored using Intel insystem sources and probes. To instantiate these in your design, add the
following line to your (.qsf) file: set_global_assignment -name
VERILOG_MACRO "ALTERA_EMIF_ENABLE_ISSP=1"
3. Compile the design by clicking Processing ➤ Start Compilation in the Quartus
Prime software.
4. Configure the FPGA on the board using the generated (.sof) file.
5.
Issue a reset on the global_reset_reset_n port. If the traffic generator test
completes successfully, the <variation_name>_status_local_cal_success
and <variation_name>_tg_0_traffic_gen_pass signals will go high,
indicating success.
6.
If you are using the Traffic Generator 2.0, you can configure and monitor the
traffic generator using the EMIF Debug Toolkit. Refer to Configuring the Traffic
Generator 2.0.
13.2 Testing the EMIF Interface Using the Traffic Generator 2.0
The EMIF Configurable Traffic Generator 2.0 can assist in debugging and stress-testing
your external memory interface. The traffic generator 2.0 supports Arria 10 and later
device families.
Key Features
The traffic generator 2.0 has the following key features:
•
Is a standalone, soft-logic device that resides in the FPGA core.
•
Is independent of the FPGA architecture and the external memory protocol in use.
•
Offers configuration options for the generation of reads and writes, addressing,
data, and data mask.
External Memory Interface Handbook Volume 3: Reference Material
405
13 Functional Description—Example Designs
For information on using the Traffic Generator 2.0 from within the EMIF Debug Toolkit,
refer to Traffic Generator 2.0 in External Memory Interface Debug Toolkit.
13.2.1 Configurable Traffic Generator 2.0 Configuration Options
The EMIF Configurable Traffic Generator 2.0 offers a range of configuration options for
fast debugging of your external memory interface. You configure the traffic generator
by modifying address-mapped registers through the simulation test bench file or by
creating a configuration test stage.
Configuration Syntax
The test bench includes an example test procedure.
The syntax for writing to a configuraton register is as follows:
tg_send_cfg_write_<index>(<Register Name>, <Value to be
written>);
The index value represents the index of the memory interface (which is usually 0, but
can be 0/1 in ping-pong PHY mode).
The register name values are listed in the tables of configuration options, and can also
be found in the altera_emif_avl_tg_defs.sv file, in the same directory as the
test bench.
The final configuration command must be a write of any value to TG_START, which
starts the traffic generator for the specified interface.
Configuration Options
Configuration options are divided into read/write, address, data, and data mask
generation categories.
Table 110.
Configuration Options for Read/Write Generation
Toolkit GUI Tab
Toolkit GUI Parameter
--
--
Loops
Register Name
Description
TG_START
Tells the system to perform a write
to this register to initiate traffic
generation test.
Loops
TG_LOOP_COUNT
Specifies the number of read/write
loops to perform before completion
of the test stage. A loop is a single
iteration of writes and reads. Upon
completion, the base address is
either incremented (SEQ or
RAND_SEQ) or replaced by a
newly generated random address
(RAND or RAND_SEQ).
Writes per block
TG_WRITE_COUNT
Specifies the number of writes to
be performed in each loop of the
test.
Reads per block
TG_READ_COUNT
Specifies the number of reads to
be performed in each loop of the
test. This register must have the
same value as TG_WRITE_COUNT.
--
TG_WRITE_REPEAT_COUNT
Specifies the number of times each
write transaction is repeated.
continued...
External Memory Interface Handbook Volume 3: Reference Material
406
13 Functional Description—Example Designs
Toolkit GUI Tab
Register Name
Description
--
TG_READ_REPEAT_COUNT
Specifies the number of times each
read transaction is repeated.
Reads per block
TG_BURST_LENGTH
Configures the length of each
write/read burst to memory.
--
--
TG_CLEAR_FIRST_FAIL
Clears the record of first failure
occurrence. New failure
information such as expected data,
read data, and fail address, is
written to the corresponding
registers following the next failure.
--
--
TG_TEST_BYTEEN
Toggles the byte-enable (data
mask enable) register within the
traffic generator which allows the
current test to use byte-enable
signals.
--
--
TG_DATA_MODE
Specifies the source of data used
for data signal generation during
the test. Set to 0 to use pseudorandom data. Set to 1 to use userspecified values stored in the static
data generators.
--
--
TG_BYTEEN_MODE
Specifies the source of data used
for byte-enable signal generation
during the test. Set to 0 to use
pseudo-random data. Set to 1 to
use user-specified values stored in
the static data generators.
Table 111.
Configuration Options for Address Generation
Toolkit GUI Tab
Address
Toolkit GUI Parameter
Toolkit GUI Parameter
Register Name
Description
Start Address
TG_SEQ_START_ADDR_WR_L
Specifies the sequential start
address (lower 32 bits).
Start address
TG_SEQ_START_ADDR_WR_H
Specifies the sequential start
address (upper 32 bits)
Address mode
TG_ADDR_MODE_WR
Specifies how write addresses are
generated by writing the value in
parentheses to the register
address. Values are: randomize
the address for every write (0),
increment sequentially from a
specified address (1), increment
sequentially from a random
address (2), or perform one hot
addressing (3).
Number of sequential
address
TG_RAND_SEQ_ADDRS_WR
Specifies the number of times to
increment sequentially on the
random base address before
generating a new random write
address.
Return to start address
TG_RETURN_TO_START_ADDR
Return to start address in
deterministic sequential address
mode (if 1 is written to
TG_ADDR_MODE_WR) after every
loop of transactions.
continued...
External Memory Interface Handbook Volume 3: Reference Material
407
13 Functional Description—Example Designs
Toolkit GUI Tab
Toolkit GUI Parameter
Register Name
Description
Rank | Mask Mode
TG_RANK_MASK_EN
Specifies the rank masking mode
by writing the value in parentheses
to the register address. Values
are: disable rank masking (0),
maintain a static rank mask (1),
cycle through rank masks
incrementally (2).
Bank Address | Mask Mode
TG_BANK_MASK_EN
Specifies the bank masking mode
by writing the value in parentheses
to the register address. Values
are: disable bank masking (0),
maintain a static bank mask (1),
cycle through bank masks
incrementally (2), cycle through
only three consecutive bank masks
(3).
Row | Mask Mode
TG_ROW_MASK_EN
Specifies the mode for row
masking by writing the value in
parentheses to the register
address. Values are: disable row
masking (0), maintain a static row
mask (1), cycle through row
masks incrementally (2).
Bank Group | Mask Mode
TG_BG_MASK_EN
Specifies the mode for bank group
masking by writing the value in
parentheses to the register
address. Values are: disable bank
group masking (0), maintain a
static bank group mask (1), cycle
through bank group masks
incrementally (2).
Rank | Mask Value
TG_RANK_MASK
Specifies the initial rank to be
masked into the generated traffic
generator address.
Bank Address | Mask Value
TG_BANK_MASK
Specifies the initial bank to be
masked into the generated traffic
generator address.
Row | Mask Value
TG_ROW_MASK
Specifies the initial row to be
masked into the generated traffic
generator address.
Bank Group | Mask Value
TG_BG_MASK
Specifies the initial bank group to
be masked into the generated
traffic generator address.
Sequential Address
Increment
TG_SEQ_ADDR_INCR
Specifies the increment to use
when sequentially incrementing
the address. This value is used by
both deterministic sequential
addressing and random sequential
addressing. (Refer to
TG_ADDR_MODE_WR)
Start Address
TG_SEQ_START_ADDR_RD_L
Specifies the sequential start read
address (lower 32 bits).
continued...
External Memory Interface Handbook Volume 3: Reference Material
408
13 Functional Description—Example Designs
Toolkit GUI Tab
Table 112.
Data
Description
TG_SEQ_START_ADDR_RD_H
Specifies the sequential start read
address (upper 32 bits).
Address Mode
TG_ADDR_MODE_RD
Similar to TG_ADDR_MODE_WR
but for reads.
Number of sequential
address
TG_RAND_SEQ_ADDRS_RD
Specifies the number of times to
increment the random sequential
read address.
Toolkit GUI Parameter
Register Name
Description
Seed/Fixed Pattern
TG_DATA_SEED
Specifies an initial value to the data
generator corresponding to the
index value.
PRBS and Fixed Pattern
radio buttons
TG_DATA_MODE
Specifies whether to treat the initial
value of the data generator of
corresponding index as a seed for
generating pseudo-random data
(value of 0) or to keep the initial
value static (value of 1).
Configuration Options for Data Mask Generation
Toolkit GUI Tab
Data
Register Name
Configuration Options for Data Generation
Toolkit GUI Tab
Table 113.
Toolkit GUI Parameter
Toolkit GUI Parameter
Register Name
Description
Seed/Fixed Pattern
TG_BYTEEN_SEED
Specifies an initial value to the
byte-enable generator
corresponding to the index value.
PRBS and Fixed Pattern
radio buttons
TG_BYTEEN_MODE
Specifies whether to treat the
initial value of the byte-enable
generator of corresponding index
as a seed and generate pseudorandom data (value of 0) or to
keep the initial value static (value
of 1).
13.2.1.1 Test Information
In the test bench file, register reads are encoded in a similar syntax to register writes.
The following example illustrates the syntax of a register read:
integer <Variable Name>;
tg_send_cfg_read_<index>(<Register Name>, <Variable Name>);
In hardware, you can probe the registers storing the test information (such as pnf
per bit persist, first fail read address, first fail read data, and
first fail expected data).
External Memory Interface Handbook Volume 3: Reference Material
409
13 Functional Description—Example Designs
Table 114.
Test Information Read-Accessible Through Register Addresses
Register Name
Description
TG_PASS
Returns a high value if the traffic generator passes at the
end of all the test stages.
TG_FAIL
Returns a high value if the traffic generator fails at the end
of all the test stages.
TG_FAIL_COUNT_L
Reports the failure count (lower 32 bits).
TG_FAIL_COUNT_H
Reports the failure count(upper 32 bits).
TG_FIRST_FAIL_ADDR_L
Reports the address of the first failure (lower 32 bits).
TG_FIRST_FAIL_ADDR_H
Reports the address of the first failure (upper 32 bits).
TG_FIRST_FAIL_IS_READ
First failure is Read Failure.
TG_FIRST_FAIL_IS_WRITE
First failure is Write Failure.
TG_VERSION
Reports the traffic generator version number.
TG_NUM_DATA_GEN
Reports the number of data generators.
TG_NUM_BYTEEN_GEN
Reports the number of byte-enable generators.
TG_RANK_ADDR_WIDTH
Reports the rank address width.
TG_BANK_ADDR_WIDTH
Reports the bank address width.
TG_ROW_ADDR_WIDTH
Reports the row address width.
TG_BANK_GROUP_WIDTH
Reports the bank group width.
TG_RDATA_WIDTH
Reports the width of all data and PNF signals within the
traffic generator.
TG_DATA_PATTERN_LENGTH
Reports the length of the static pattern to be loaded into
static per-pin data generators.
TG_BYTEEN_PATTERN_LENGTH
Reports the length of the static pattern to be loaded into
static per-pin byte-enable generators.
TG_MIN_ADDR_INCR
Reports the minimum address increment permitted for
sequential and random-sequential address generation.
TG_ERROR_REPORT
Reports error bits. Refer to Error Report Register Bits for
details.
TG_PNF
Read the persistent PNF per bit as an array of 32-bit
entries.
TG_FAIL_EXPECTED_DATA
Reports the first failure expected data. This is read as an
array of 32-bit entries.
TG_FAIL_READ_DATA
Reports the first failure read data. This is read as an array
of 32-bit entries.
The configuration error report register contains information on common
misconfigurations of the traffic generator. The bit corresponding to a given
configuration error is set high when that configuration error is detected. Ensure that
all bits in this register are low, to avoid test failures or unexpected behavior due to
improper configuration.
External Memory Interface Handbook Volume 3: Reference Material
410
13 Functional Description—Example Designs
Table 115.
Error Report Register Bits
Bit Index (LSB = 0x0)
Bit Name
Description of Error
0
ERR_MORE_READS_THAN_WRITES
You have requested more read
transactions per loop than write
transactions per loop.
1
ERR_BURSTLENGTH_GT_SEQ_ADDR_I
NCR
The configured burst length is greater
than the configured sequential address
increment when address generation is
in sequential or random sequential
mode.
2
ERR_ADDR_DIVISIBLE_BY_GT_SEQ_A
DDR_INCR
The configured sequential address
increment is less than the minimum
required address increment when
address generation is in sequential or
random sequential mode.
3
ERR_SEQ_ADDR_INCR_NOT_DIVISIBL
E
The configured sequential address
increment is not a multiple of the
minimum required address increment
when address generation is in
sequential or random sequential mode.
4
ERR_READ_AND_WRITE_START_ADDR
S_DIFFER
The configured start addresses for
reads and writes are different when
address generation is in sequential
mode.
5
ERR_ADDR_MODES_DIFFERENT
The configured address modes for
reads and writes are different.
6
ERR_NUMBER_OF_RAND_SEQ_ADDRS
_DIFFERENT
The configured number of random
sequential addresses for reads and
writes are different when address
generation is in random sequential
mode.
7
ERR_REPEATS_SET_TO_ZERO
The number of read and/or write
repeats is set to 0.
8
ERR_BOTH_BURST_AND_REPEAT_MOD
E_ACTIVE
The burst length is set greater than 1
and read/write requests are set greater
than 1.
9-31
Reserved
--
13.2.2 Performing Your Own Tests Using Traffic Generator 2.0
If you want, you can create your own configuration test stage for the EMIF
Configurable Traffic Generator.
The general flow of a configuration test stage, including the default test stages, is as
follows:
•
Configure the number of loops to be completed by the traffic generator
(TG_LOOP_COUNT).
•
Configure the number of writes and reads to be complete per loop
(TG_WRITE_COUNT and TG_READ_COUNT respectively).
•
Choose the burst length of each write and read (TG_BURST_LENGTH).
•
Select starting write address by writing to the lower and upper bits of the address
register (TG_SEQ_START_ADDR_WR_L and TG_SEQ_START_ADDR_WR_H,
respectively).
External Memory Interface Handbook Volume 3: Reference Material
411
13 Functional Description—Example Designs
•
Select write address generation mode (TG_ADDR_MODE_WR).
•
Select starting read address by writing to the lower and upper bits of the address
register (TG_SEQ_START_ADDR_RD_L and TG_SEQ_START_ADDR_RD_H,
respectively).
•
Select read address generation mode (TG_ADDR_MODE_RD).
•
If applicable, select sequential address increment (TG_SEQ_ADDR_INCR).
•
Write initial values/seeds to the data and byte-enable generators (TG_DATA_SEED
and TG_BYTEEN_SEED).
•
Select generation mode of the data and byte-enable generators (TG_DATA_MODE
and TG_BYTEEN_MODE).
•
Initiate test (TG_START).
Simulation
For a comprehensive example of how to write your own configuration test for
simulation, refer to the test bench file, located at
<example_design_directory>/sim/ed_sim/
altera_emif_tg_avl_2_<>/sim/altera_emif_avl_tg_2_tb.sv
To iterate over the data generators or byte-enable generators, you must read the
number of data generators and number of byte-enable generators. These values are
mapped to read-accessible registers TG_NUM_DATA_GEN and TG_NUM_BYTEEN_GEN,
respectively. The following example illustrates how one would configure the data
generators to continuously output the pattern 0x5a, using the simulation test bench:
integer num_data_generators;
…
tg_send_cfg_read_0(TG_NUM_DATA_GEN, num_data_generators);
tg_send_cfg_write_0(TG_DATA_MODE, 32'h1);
for (i = 0; i < num_data_generators; i = i + 1) begin
tg_send_cfg_write_0(TG_DATA_SEED + i, 32'h5A);
end
Hardware
Configuration test stages in hardware must be inserted into the RTL, and will resemble
the single read/write, byte-enable, and block read/write stages in the default test
pattern. In most cases, you can modify one of the existing stages to create the
desired custom test stage. The stages are linear, finite, state machines that write
predetermined values to the configuration address-mapped registers. As always, the
last state in configuration is a write to address 0x0 or TG_START. The state machine
then waits for the traffic generator to return a signal signifying its completion of the
test stage.
Refer to the aforementioned default test stages as examples of hardware test stages.
The default test stages are contained within the following files:
•
<example_design_directory>/qii/ed_synth/
altera_emif_tg_avl_2_<>/synth/altera_emif_avl_tg_2_rw_stage.sv
•
<example_design_directory>/qii/ed_synth/
altera_emif_tg_avl_2_<>/synth/
altera_emif_avl_tg_2_byteenable_test_stage.sv
External Memory Interface Handbook Volume 3: Reference Material
412
13 Functional Description—Example Designs
You can also configure the configurable traffic generator in real time using the EMIF
Debug Toolkit. The configuration settings available in the Toolkit interface are detailed
in the Configurable Traffic Generator 2.0 Configuration Options topic.
13.2.3 Signal Splitter Component
The signal splitter (altera_emif_sig_splitter) is an internal IP component which
receives a single signal as its input and passes that signal directly to n outputs, where
n is an interger value equal to or greater than 1.The signal splitter is useful because
Qsys does not directly allow one-to-many connections for conduit interfaces.
The signal splitter contains no logic or memory elements. When you configure the
signal splitter to have exactly one output port, it is functionally identical to a single
wire, and can be replaced by one with no loss of performance.
The rzq_splitter is an instantiation of the signal splitter component specifically for
the RZQ signal. The signal splitter facilitates the sharing of one RZQ signal among
multiple memory interfaces in an EMIF example design.
13.3 UniPHY-Based Example Designs
For UniPHY-based interfaces, two independent example designs are created, each
containing independent RTL files and other project files. You should compile or
simulate these designs separately, and the files should not be mixed. Nonetheless, the
designs are related, as the simulation example design builds upon the design of the
synthesis example design.
13.3.1 Synthesis Example Design
The synthesis example design contains the major blocks shown in the figure below.
•
A traffic generator, which is a synthesizable Avalon-MM example driver that
implements a pseudo-random pattern of reads and writes to a parameterized
number of addresses. The traffic generator also monitors the data read from the
memory to ensure it matches the written data and asserts a failure otherwise.
•
An instance of the UniPHY memory interface, which includes a memory controller
that moderates between the Avalon-MM interface and the AFI interface, and the
UniPHY, which serves as an interface between the memory controller and external
memory devices to perform read and write operations.
Figure 176. Synthesis Example Design
Synthesis Example Design
Traffic Generator
Avalon-MM
Controller
AFI
PHY
Memory
IP
External Memory Interface Handbook Volume 3: Reference Material
413
13 Functional Description—Example Designs
If you are using the Ping Pong PHY feature, the synthesis example design includes two
traffic generators issuing commands to two independent memory devices through two
independent controllers and a common PHY, as shown in the following figure.
Figure 177. Synthesis Example Design for Ping Pong PHY
Synthesis Example Design
Traffic
Generator 0
Traffic
Generator 1
Avalon-MM
Avalon-MM
Memory
AFI
Controller 0
AFI
Controller 1
PHY
Memory
IP
If you are using RLDRAM 3, the traffic generator in the synthesis example design
communicates directly with the PHY using AFI, as shown in the following figure.
Figure 178. Synthesis Example Design for RLDRAM 3 Interfaces
Synthesis Example Design
Traffic Generator
AFI
PHY
Memory
IP
You can obtain the synthesis example design by generating your IP core. The files
related to the synthesis example design reside at < variation_name >_
example_design / example_project. The synthesis example design includes a
Quartus Prime project file (< variation_name >_ example_design /
example_project /< variation_name >_example.qpf). The Quartus Prime project
file can be compiled in the Quartus Prime software, and can be run on hardware.
Note:
If one or more of the PLL Sharing Mode, DLL Sharing Mode, or OCT Sharing
Mode parameters are set to any value other than No Sharing, the synthesis example
design will contain two traffic generator/memory interface instances. The two traffic
generator/memory interface instances are related only by shared PLL/DLL/OCT
connections as defined by the parameter settings. The traffic generator/memory
interface instances demonstrate how you can make such connections in your own
designs.
External Memory Interface Handbook Volume 3: Reference Material
414
13 Functional Description—Example Designs
13.3.2 Simulation Example Design
The simulation example design contains the major blocks shown in the following
figure.
•
An instance of the synthesis example design. As described in the previous section,
the synthesis example design contains a traffic generator and an instance of the
UniPHY memory interface. These blocks default to abstract simulation models
where appropriate for rapid simulation.
•
A memory model, which acts as a generic model that adheres to the memory
protocol specifications. Frequently, memory vendors provide simulation models for
specific memory components that you can download from their websites.
•
A status checker, which monitors the status signals from the UniPHY IP and the
traffic generator, to signal an overall pass or fail condition.
Figure 179. Simulation Example Design
Simulation Example Design
Abstract instance of the synthesis example design
Traffic Generator
Avalon-MM
Controller
AFI
PHY
Memory
Memory
Model
IP
Status
Checker
If you are using the Ping Pong PHY feature, the simulation example design includes
two traffic generators issuing commands to two independent memory devices through
two independent controllers and a common PHY, as shown in the following figure.
External Memory Interface Handbook Volume 3: Reference Material
415
13 Functional Description—Example Designs
Figure 180. Simulation Example Design for Ping Pong PHY
Simulation Example Design
Abstract Instance of the Synthesis Example Design
Traffic
Generator 0
Avalon-MM
Controller 0
AFI
Memory
Memory
Model 0
Memory
Memory
Model 1
PHY
Traffic
Generator 1
Avalon-MM
Controller 1
AFI
IP
Status Checker
If you are using RLDRAM 3, the traffic generator in the simulation example design
communicates directly with the PHY using AFI, as shown in the following figure.
Figure 181. Simulation Example Design for RLDRAM 3 Interfaces
Synthesis Example Design
Abstract instance of the synthesis example design
Traffic Generator
AFI
PHY
Memory
Memory
Model
IP
You can obtain the simulation example design by generating your IP core. The files
related to the simulation example design reside at < variation_name >_
example_design /simulation. After obtaining the generated files, you must still
generate the simulation example design RTL for your desired HDL language. The file <
variation_name >_ example_design /simulation/README.txt contains details
about how to generate the IP and to run the simulation in ModelSim or ModelSim Intel FPGA Edition.
13.3.3 Traffic Generator and BIST Engine
The traffic generator and built-in self test (BIST) engine for Avalon-MM memory
interfaces generates Avalon-MM traffic on an Avalon-MM master interface. The traffic
generator creates read and write traffic, stores the expected read responses internally,
and compares the expected responses to the read responses as they arrive. If all
reads report their expected response, the pass signal is asserted; however, if any read
responds with unexpected data a fail signal occurs.
External Memory Interface Handbook Volume 3: Reference Material
416
13 Functional Description—Example Designs
Each operation generated by the traffic generator is a single write or block of writes
followed by a single read or block of reads to the same addresses, which allows the
driver to precisely determine the data that should be expected when the read data is
returned by the memory interface. The traffic generator comprises a traffic generation
block, the Avalon-MM interface and a read comparison block. The traffic generation
block generates addresses and write data, which are then sent out over the AvalonMM interface. The read comparison block compares the read data received from the
Avalon-MM interface to the write data from the traffic generator. If at any time the
data received is not the expected data, the read comparison block records the failure,
finishes reading all the data, and then signals that there is a failure and the traffic
generator enters a fail state. If all patterns have been generated and compared
successfully, the traffic generator enters a pass state.
Figure 182. Example Driver Operations
Initialize
Sequential
Addresses
Individual
Reads/Writes
Random
Addresses
Sequential/Random
Addresses
Block
Reads/Writes
User-Defined
Stages (Optional)
Pass
User-defined
Addresses
(Optional)
Fail
Within the traffic generator, there are the following main states:
•
Generation of individual read and writes
•
Generation of block read and writes
•
The pass state
•
The fail state
Within each of the generation states there are the following substates:
•
Sequential address generation
•
Random address generation
•
Mixed sequential and random address generation
For each of the states and substates, the order and number of operations generated
for each substate is parameterizable—you can decide how many of each address
pattern to generate, or can disable certain patterns entirely if you want. The
sequential and random interleave substate takes in additions to the number of
operations to generate. An additional parameter specifies the ratio of sequential to
random addresses to generate randomly.
External Memory Interface Handbook Volume 3: Reference Material
417
13 Functional Description—Example Designs
13.3.3.1 Read and Write Generation
The traffic generator block can generate individual or block reads and writes.
Individual Read and Write Generation
During the traffic generator’s individual read and write generation state, the traffic
generation block generates individual write followed by individual read Avalon-MM
transactions, where the address for the transactions is chosen according to the specific
substate. The width of the Avalon-MM interface is a global parameter for the driver,
but each substate can have a parameterizable range of burst lengths for each
operation.
Block Read and Write Generation
During the traffic generator’s block read and write generation state, the traffic
generator block generates a parameterizable number of write operations followed by
the same number of read operations. The specific addresses generated for the blocks
are chosen by the specific substates. The burst length of each block operation can be
parameterized by a range of acceptable burst lengths.
13.3.3.2 Address and Burst Length Generation
The traffic generator block can perform sequential or random addressing.
Sequential Addressing
The sequential addressing substate defines a traffic pattern where addresses are
chosen in sequential order starting from a user definable address. The number of
operations in this substate is parameterizable.
Random Addressing
The random addressing substate defines a traffic pattern where addresses are chosen
randomly over a parameterizable range. The number of operations in this substate is
parameterizable.
Sequential and Random Interleaved Addressing
The sequential and random interleaved addressing substate defines a traffic pattern
where addresses are chosen to be either sequential or random based on a
parameterizable ratio. The acceptable address range is parameterizable as is the
number of operations to perform in this substate.
13.3.3.3 Traffic Generator Signals
The following table lists the signals used by the traffic generator.
Table 116.
Traffic Generator Signals
Signal
Signal Type
clk
Input clock to traffic generator. For 28nm devices, use the AFI clock. For 20nm devices, use the
emif_usr_clk.
reset_n
Active low reset input. For 28nm devices, typically connected to afi_reset_n. For 20nm devices,
typically connectd to emif_usr_reset_n.
continued...
External Memory Interface Handbook Volume 3: Reference Material
418
13 Functional Description—Example Designs
Signal
Signal Type
avl_ready
Refer to Avalon Interface Specification.
avl_write_req
Refer to Avalon Interface Specification.
avl_read_req
Refer to Avalon Interface Specification.
avl_addr
Refer to Avalon Interface Specification.
avl_size
Refer to Avalon Interface Specification.
avl_wdata
Refer to Avalon Interface Specification.
avl_rdata
Refer to Avalon Interface Specification.
avl_rdata_valid
Refer to Avalon Interface Specification.
pnf_per_bit
Output. Bitwise pass/fail for last traffic generator read compare of AFI interface. No errors produces
value of all 1s.
pnf_per_bit_persist
Output. Cumulative bitwise pass/fail for all previous traffic generator read compares of AFI interface.
No errors produces value of all 1s.
pass
Active high output when traffic generator tests complete successfully.
fail
Active high output when traffic generator test does not complete successfully.
test_complete
Active high output when traffic generator test completes.
For information about the Avalon signals and the Avalon interface, refer to Avalon
Interface Specifications.
Related Links
Avalon Interface Specifications
13.3.3.4 Traffic Generator Add-Ons
Some optional components that can be useful for verifying aspects of the controller
and PHY operation are generated in conjunction with certain user-specified options.
These add-on components are self-contained, and are not part of the controller or
PHY, nor the traffic generator.
External Memory Interface Handbook Volume 3: Reference Material
419
13 Functional Description—Example Designs
User Refresh Generator
The user refresh generator sends refresh requests to the memory controller when user
refresh is enabled. The memory controller returns an acknowledgement signal and
then issues the refresh command to the memory device.
The user refresh generator is created when you turn on Enable User Refresh on the
Controller Settings tab of the parameter editor.
13.3.3.5 Traffic Generator Timeout Counter
The traffic generator timeout counter uses the Avalon interface clock.
When a test fails due to driver failure or timeout, the fail signal is asserted. When a
test has failed, the traffic generator must be reset with the reset_n signal.
13.3.4 Creating and Connecting the UniPHY Memory Interface and the
Traffic Generator in Qsys
The traffic generator can be used in Qsys as a stand-alone component for use within a
larger system.
To create the system in Qsys, perform the following steps:
1. Start Qsys.
2. On the Project Settings tab, select the required device from the Device Family
list.
3.
In the Component Library, choose a UniPHY memory interface to instantiate. For
example, under Library > Memories and Memory Controllers > External
Memory Interfaces, select DDR3 SDRAM Controller with UniPHY.
4.
Configure the parameters for your instantiation of the memory interface.
5.
In the Component Library, find the example driver and instantiate it in the system.
For example, under Library > Memories and Memory Controllers > Pattern
Generators, select Avalon-MM Traffic Generator and BIST Engine.
6.
Configure the parameters for your instantiation of the example driver.
Note: The Avalon specification stipulates that Avalon-MM master interfaces issue
byte addresses, while Avalon-MM slave interfaces accept word addresses.
The default for the Avalon-MM Traffic Generator and BIST Engine is to issue
word addresses. When using Qsys, you must enable the Generate per
byte address setting in the traffic generator.
7. Connect the interfaces as illustrated in the following figure. At this point, you can
generate synthesis RTL, Verilog or VHDL simulation RTL, or a simulation testbench
system.
External Memory Interface Handbook Volume 3: Reference Material
420
13 Functional Description—Example Designs
13.3.4.1 Notes on Configuring UniPHY IP in Qsys
This section includes notes and tips on configuring the UniPHY IP in Qsys.
•
The address ranges shown for the Avalon-MM slave interface on the UniPHY
component should be interpreted as byte addresses that an Avalon-MM master
would address, despite the fact that this range is modified by configuring the word
addresses width of the Avalon-MM slave interface on the UniPHY controller.
•
The afi_clk clock source is the associated clock to the Avalon-MM slave interface
on the memory controller. This is the ideal clock source to use for all IP
components connected on the same Avalon network. Using another clock would
cause Qsys to automatically instantiate clock-crossing logic, potentially degrading
performance.
•
The afi_clk clock rate is determined by the Rate on Avalon-MM interface
setting on the UniPHY PHY Settings tab. The afi_half_clk clock interface has
a rate which is further halved. For example, if Rate on Avalon-MM interface is
set to Half, the afi_clk rate is half of the memory clock frequency, and the
afi_half_clk is one quarter of the memory clock frequency.
•
The global_reset input interface can be used to reset the UniPHY memory
interface and the PLL contained therein. The soft_reset input interface can be
used to reset the UniPHY memory interface but allow the PLL to remain locked.
You can use the soft_reset input to reset the memory but to maintain the AFI
clock output to other components in the system.
External Memory Interface Handbook Volume 3: Reference Material
421
13 Functional Description—Example Designs
•
Do not connect a reset request from a system component (such as a Nios II
processor) to the UniPHY global_reset_n port. Doing so would reset the
UniPHY PLL, which would propagate as a reset condition on afi_reset back to
the requester; the resulting reset loop could freeze the system.
•
Qsys generates an interconnect fabric for each Avalon network. The interconnect
fabric is capable of burst and width adaptation. If your UniPHY memory controller
is configured with an Avalon interface data width which is wider than an AvalonMM master interface connected to it, you must enable the byte enable signal on
the Avalon-MM slave interface, by checking the Enable Avalon-MM byte-enable
signal checkbox on the Controller Settings tab in the parameter editor.
•
If you have a point-to-point connection from an Avalon-MM master to the AvalonMM slave interface on the memory controller, and if the Avalon data width and
burst length settings match, then the Avalon interface data widths may be
multiples of either a power of two or nine. Otherwise, you must enable Generate
power-of-2 data bus widths for Qsys or SOPC Builder on the Controller
Settings tab of the parameter editor.
13.4 Document Revision History
Date
May 2017
Version
2017.05.08
Changes
•
•
•
•
Retitled EMIF Configurable Traffic Generator 2.0 Reference to Testing the
EMIF Interface Using the Traffic Generator 2.0.
Removed Configurable Traffic Generator Parameters section.
Consolidated Traffic Generator 2.0 usage information in External Memory
Interface Debug Toolkit chapter.
Rebranded as Intel.
October 2016
2016.10.31
Maintenance release.
May 2016
2016.05.02
•
•
•
•
•
•
•
•
Modified step 6 of Compiling and Testing the Design in Hardware.
Added Configuring the Traffic Generator 2.0.
Added The Traffic Generator 2.0 Report.
Changed heading from EMIF Configurable Traffic Generator 2.0 for Arria
10 EMIF IP to EMIF Configurable Traffic Generator 2.0 Reference.
Added Bypass the traffic generator repeated-writes/ repeated-reads test
pattern, Bypass the traffic generator stress pattern, and Export Traffic
Generator 2.0 configuration interface to Configurable Traffic Generator
Parameters table. Removed Number of Traffic Generator 2.0 configuration
interfaces.
Added TG_WRITE_REPEAT_ COUNT, TG_READ_REPEAT_ COUNT,
TG_DATA_MODE, and TG_BYTEEN_MODE to Configuration Options for
Read/Write Generation table.
Added Error Report Register Bits table to Test Information section.
Corrected paths in Simulation and Hardware sections of Performing Your
Own Tests Using Traffic Generator 2.0 topic.
November 2015
2015.11.02
•
•
•
•
Added EMIF Configurable Traffic Generator 2.0 for Arria 10 EMIF IP.
Added Arria 10 EMIF IP Example Design Quick Start Guide.
Changed order of sections in chapter.
Changed instances of Quartus II to Quartus Prime.
May 2015
2015.05.04
Maintenance release.
December 2014
2014.12.15
Maintenance release.
August 2014
2014.08.15
Removed occurrence of MegaWizard Plug-In Manager.
December 2013
2013.12.16
Removed references to SOPC Builder.
continued...
External Memory Interface Handbook Volume 3: Reference Material
422
13 Functional Description—Example Designs
Date
November 2012
Version
1.3
Changes
•
•
Added block diagrams of simulation and synthesis example designs for
RLDRAM 3 and Ping Pong PHY.
Changed chapter number from 7 to 9.
June 2012
1.2
Added Feedback icon.
November 2011
1.1
•
•
•
Added Synthesis Example Design and Simulation Example Design
sections.
Added Creating and Connecting the UniPHY Memory Interface and the
Traffic Generator in Qsys.
Revised Example Driver section as Traffic Generator and BIST Engine.
External Memory Interface Handbook Volume 3: Reference Material
423
14 Introduction to UniPHY IP
14 Introduction to UniPHY IP
The UniPHY IP is an interface between a memory controller and memory devices and
performs read and write operations to the memory. The UniPHY IP creates the
datapath between the memory device and the memory controller and user logic in
various Intel devices.
The Intel FPGA DDR2, DDR3, and LPDDR2 SDRAM controllers with UniPHY, QDR II and
QDR II+ SRAM controllers with UniPHY, RLDRAM II controller with UniPHY, and
RLDRAM 3 PHY-only IP provide low latency, high-performance, feature-rich interfaces
to industry-standard memory devices. The DDR2, QDR II and QDR II+, and RLDRAM
II controllers with UniPHY offer full-rate and half-rate interfaces, while the DDR3
controller with UniPHY and the RLDRAM 3 PHY-only IP offer half-rate and quarter-rate
interfaces, and the LPDDR2 controller with UniPHY offers a half-rate interface.
When you generate your external memory interface IP core, the system creates an
example top-level project, consisting of an example driver, and your controller custom
variation. The controller instantiates an instance of the UniPHY datapath.
The example top-level project is a fully-functional design that you can simulate,
synthesize, and use in hardware. The example driver is a self-test module that issues
read and write commands to the controller and checks the read data to produce the
pass, fail, and test-complete signals.
If the UniPHY datapath does not match your requirements, you can create your own
memory interface datapath using the ALTDLL, ALTDQ_DQS, ALTDQ_DQS2, ALTDQ, or
ALTDQS IP cores, available in the Quartus Prime software, but you are then
responsible for all aspects of the design including timing analysis and design
constraints.
14.1 Release Information
The following table provides information about this release of the DDR2 and DDR3
SDRAM, QDR II and QDR II+ SRAM, and RLDRAM II controllers with UniPHY, and the
RLDRAM 3 PHY-only IP.
Table 117.
Release Information
Item
Protocol
DDR2, DDR3,
LPDDR2
QDR II
RLDRAM II
RLDRAM 3
Version
13.1
13.1
13.1
13.1
Release Date
November 2013
November 2013
November 2013
November 2013
Ordering Code
IP-DDR2/UNI
IP-DDR3/UNI
IP-SDRAM/LPDDR2
IP-QDRII/UNI
IP-RLDII/UNI
Intel Corporation. All rights reserved. Intel, the Intel logo, Altera, Arria, Cyclone, Enpirion, MAX, Nios, Quartus
and Stratix words and logos are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other
countries. Intel warrants performance of its FPGA and semiconductor products to current specifications in
accordance with Intel's standard warranty, but reserves the right to make changes to any products and services
at any time without notice. Intel assumes no responsibility or liability arising out of the application or use of any
information, product, or service described herein except as expressly agreed to in writing by Intel. Intel
customers are advised to obtain the latest version of device specifications before relying on any published
information and before placing orders for products or services.
*Other names and brands may be claimed as the property of others.
—
ISO
9001:2008
Registered
14 Introduction to UniPHY IP
Intel verifies that the current version of the Quartus Prime software compiles the
previous version of each IP core. The Intel FPGA IP Library Release Notes and Errata
report any exceptions to this verification. Intel does not verify compilation with IP core
versions older than one release.
Related Links
Intel FPGA IP Library Release Notes
14.2 Device Support Levels
The following terms define the device support levels for Intel FPGA IP cores.
Intel FPGA IP Core Device Support Levels
•
Preliminary support—Intel verifies the IP core with preliminary timing models
for this device family. The IP core meets all functional requirements, but might still
be undergoing timing analysis for the device family. You can use it in production
designs with caution.
•
Final support—Intel verifies the IP core with final timing models for this device
family. The IP core meets all functional and timing requirements for the device
family and can be used in production designs.
14.3 Device Family and Protocol Support
The following table shows the level of support offered by each of the UniPHY-based
external memory interface protocols for Intel device families.
Table 118.
Device Family Support
Device Family
Support Level
DDR2
DDR3
No support
No support
No support
Final
No support
No support
Arria II GZ
Final
Final
No support
Final
Final
No support
Arria V
Refer to the What’s New in Intel FPGA IP on www.altera.com.
Arria V GZ
Refer to the What’s New in
Intel FPGA IP on
www.altera.com.
Cyclone V
Refer to the What’s New in Intel FPGA IP on www.altera.com.
Stratix®
Final
Final (Only
Vcc = 1.1V
supported)
No support
Final
Final (Only
Vcc = 1.1V
supported)
No support
Stratix IV
Final
Final
No support
Final
Final
No support
Stratix V
Refer to the What’s New in
Intel FPGA IP on
www.altera.com.
No support
Refer to the What’s New in Intel FPGA IP on
www.altera.com.
Other device families
No support
No support
No support
Arria®
II GX
III
No support
LPDDR2
No support
QDR II
RLDRAM II
RLDRAM 3
No support
Refer to the What’s New in Intel FPGA IP on
www.altera.com.
No support
No support
No support
For information about features and supported clock rates for external memory
interfaces, refer to the External Memory Specification Estimator.
External Memory Interface Handbook Volume 3: Reference Material
425
14 Introduction to UniPHY IP
Related Links
•
What's New in Intel FPGA IP
•
External Memory Interface Spec Estimator
14.4 UniPHY-Based External Memory Interface Features
The following table summarizes key feature support for Intel’s UniPHY-based external
memory interfaces.
Table 119.
Feature Support
Key Feature
Protocol
DDR2
DDR3
LPDDR2
QDR II
RLDRAM
II
RLDRAM 3
High-performance controller II (HPC
II)
Yes
Yes
Yes
—
—
—
Half-rate core logic and user interface
Yes
Yes
Yes
Yes
Yes
Yes
Full-rate core logic and user interface
Yes
—
—
Yes
Yes
—
—
—
—
Yes
Yes
Yes
Yes
Quarter-rate core logic and user
interface
—
(1)
Yes
Dynamically generated Nios II-based
sequencer
Yes
Yes
Yes
Choice of RTL-based or dynamically
generated Nios® II-based sequencer
—
—
—
Yes
Yes
Available Efficiency Monitor and
Protocol Checker (14)
DDR3L support
—
(2) (3)
Yes
(12)
(12)
—
Yes
—
Yes
Yes
(13)
—
—
—
—
(4) (5)
—
—
—
—
Yes
UDIMM and RDIMM in any form factor
Yes
Multiple components in a single-rank
UDIMM or RDIMM layout
Yes
Yes
—
—
—
—
LRDIMM
—
Yes
—
—
—
—
Burst length (half-rate)
8
—
8 or 16
4
4 or 8
2, 4, or 8
Burst length (full-rate)
4
—
—
2 or 4
2, 4, or 8
—
Burst length (quarter-rate)
—
8
—
—
—
2, 4, or 8
Burst length of 8 and burst chop of 4
(on the fly)
—
Yes
—
—
—
—
(9) (10)
—
—
—
Yes
Yes
—
—
—
32 bits
72 bits
72 bits
72 bits
With leveling
Yes
Yes
240 MHz
and above
(10)
Without leveling
Below 240
MHz
Maximum data width
144 bits
(6)
Yes
—
144 bits
(6)
Reduced controller latency
—
—
—
Read latency
—
—
—
Yes
(2) (7)
1.5 (QDR
II) 2 or 2.5
(QDR II+)
Yes
(2) (7)
—
—
—
continued...
External Memory Interface Handbook Volume 3: Reference Material
426
14 Introduction to UniPHY IP
Key Feature
Protocol
DDR2
DDR3
LPDDR2
QDR II
RLDRAM
II
RLDRAM 3
ODT (in memory device)
—
Yes
—
QDR II+
only
Yes
Yes
x36 emulation mode
—
—
—
Yes (8) (10)
—
—
Notes:
1. For Arria V, Arria V GZ, and Stratix V devices only.
2. Not available in Arria II GX devices.
3. Nios II-based sequencer not available for full-rate interfaces.
4. For DDR3, the DIMM form is not supported in Arria II GX, Arria II GZ, Arria V, or Cyclone V devices.
5. Arria II GZ uses leveling logic for discrete devices in DDR3 interfaces to achieve high speeds, but that leveling cannot
be used to implement the DIMM form in DDR3 interfaces.
6. For any interface with data width above 72 bits, you must use software timing analysis of your complete design to
determine the maximum clock rate.
7. The maximum achievable clock rate when reduced controller latency is selected must be attained through
Quatrus Prime software timing analysis of your complete design.
8. Emulation mode allows emulation of a larger memory-width interface using multiple smaller memory-width interfaces.
For example, an x36 QDR II or QDR II+ interface can be emulated using two x18 interfaces.
9. The leveling delay on the board between first and last DDR3 SDRAM component laid out as a DIMM must be less than
0.69 tCK.
10.Leveling is not available for Arria V or Cyclone V devices.
11.x36 emulation mode is not supported in Arria V, Arria V GZ, Cyclone V, or Stratix V devices.
12.The RTL-based sequencer is not available for QDR II or RLDRAM II interfaces on Arria V devices.
13.For Arria V, Arria V GZ, Cyclone V, and Stratix V.
14.The Efficiency Monitor and Protocol Checker is is not available for QDR II and QDR II+ SRAM, or for the MAX 10 device
family, or for Arria V or Cyclone V designs using the Hard Memory Controller.
14.5 System Requirements
For system requirements and installation instructions, refer to Intel FPGA Software
Installation and Licensing manual.
The DDR2, DDR3, and LPDDR2 SDRAM controllers with UniPHY, QDR II and QDR II+
SRAM controllers with UniPHY, RLDRAM II controller with UniPHY, and RLDRAM 3 PHYonly IP are part of the Intel FPGA IP Library, which Intel distributes with the Quartus
Prime software.
Related Links
Intel FPGA Software Installation and Licensing Manual
14.6 Intel FPGA IP Core Verification
Intel has carried out extensive random, directed tests with functional test coverage
using industry-standard models to ensure the functionality of the external memory
controllers with UniPHY. Intel’s functional verification of the external memory
controllers with UniPHY use modified Denali models, with certain assertions disabled.
14.7 Resource Utilization
The following topics provide resource utilization data for the external memory
controllers with UniPHY for supported device families.
External Memory Interface Handbook Volume 3: Reference Material
427
14 Introduction to UniPHY IP
14.7.1 DDR2, DDR3, and LPDDR2 Resource Utilization in Arria V Devices
The following table shows typical resource usage of the DDR2, DDR3, and LPDDR2
SDRAM controllers with UniPHY in the current version of Quartus Prime software for
Arria V devices.
Table 120.
Resource Utilization in Arria V Devices
Protocol
Memory
Width (Bits)
Combinationa
l ALUTS
Logic
Registers
M10K Blocks
Memory
(Bits)
Hard Memory
Controiler
Controller
DDR2 (Half
rate)
8
2286
1404
4
6560
0
64
2304
1379
17
51360
0
DDR2
(Fullrate)
32
0
0
0
0
1
DDR3 (Half
rate)
8
2355
1412
4
6560
0
64
2372
1440
17
51360
0
DDR3 (Full
rate)
32
0
0
0
0
1
LPDDR2 (Half
rate)
8
2230
1617
4
6560
0
32
2239
1600
10
25760
0
DDR2 (Half
rate)
8
1652
2015
34
141312
0
64
1819
2089
34
174080
0
DDR2
(Fullrate)
32
1222
1415
34
157696
1
DDR3 (Half
rate)
8
1653
1977
34
141312
0
64
1822
2090
34
174080
0
DDR3 (Full
rate)
32
1220
1428
34
157696
0
LPDDR2 (Half
rate)
8
2998
3187
35
150016
0
32
3289
3306
35
174592
0
DDR2 (Half
rate)
8
4555
3959
39
148384
0
64
4991
4002
52
225952
0
DDR2
(Fullrate)
32
1776
1890
35
158208
1
DDR3 (Half
rate)
8
4640
3934
39
148384
0
64
5078
4072
52
225952
0
DDR3 (Full
rate)
32
1774
1917
35
158208
1
LPDDR2 (Half
rate)
8
5228
4804
39
156576
0
32
5528
4906
45
200352
0
PHY
Total
External Memory Interface Handbook Volume 3: Reference Material
428
14 Introduction to UniPHY IP
14.7.2 DDR2 and DDR3 Resource Utilization in Arria II GZ Devices
The following table shows typical resource usage of the DDR2 and DDR3 SDRAM
controllers with UniPHY in the current version of Quartus Prime software for
Arria II GZ devices.
Table 121.
Resource Utilization in Arria II GZ Devices
Protocol
Memory
Width
(Bits)
Combinatio
nal ALUTS
Logic
Registers
Mem ALUTs
M9K Blocks
M144K
Blocks
Memory
(Bits)
Controller
DDR2 (Half
rate)
DDR2 (Full
rate)
DDR3 (Half
rate)
8
1,781
1,092
10
2
0
4,352
16
1,784
1,092
10
4
0
8,704
64
1,818
1,108
10
15
0
34,560
72
1,872
1,092
10
17
0
39,168
8
1,851
1,124
10
2
0
2,176
16
1,847
1,124
10
2
0
4,352
64
1,848
1,124
10
8
0
17,408
72
1,852
1,124
10
9
0
19,574
8
1,869
1,115
10
2
0
4,352
16
1,868
1,115
10
4
0
8,704
64
1,882
1,131
10
15
0
34,560
72
1,888
1,115
10
17
0
39,168
8
2,560
2,042
183
22
0
157,696
16
2,730
2,262
183
22
0
157,696
64
3,606
3,581
183
22
0
157,696
72
3,743
3,796
183
22
0
157,696
8
2,494
1,934
169
22
0
157,696
16
2,652
2,149
169
22
0
157,696
64
3,519
3,428
169
22
0
157,696
72
3,646
3,642
169
22
0
157,696
8
2,555
2,032
187
22
0
157,696
16
3,731
2,251
187
22
0
157,696
64
3,607
3,572
187
22
0
157,696
72
3,749
3,788
187
22
0
157,696
8
4,341
3,134
193
24
0
4,374
16
4,514
3,354
193
26
0
166,400
64
5,424
4,689
193
37
0
PHY
DDR2 (Half
rate)
DDR2 (Full
rate)
DDR3 (Half
rate)
Total
DDR2 (Half
rate)
192,256
continued...
External Memory Interface Handbook Volume 3: Reference Material
429
14 Introduction to UniPHY IP
Protocol
DDR2 (Full
rate)
DDR3 (Half
rate)
Memory
Width
(Bits)
Combinatio
nal ALUTS
Logic
Registers
Mem ALUTs
M9K Blocks
M144K
Blocks
Memory
(Bits)
72
5,615
4,888
193
39
0
196,864
8
4,345
3,058
179
24
0
159,872
16
4,499
3,273
179
24
0
162,048
64
5,367
4,552
179
30
0
175,104
72
5,498
4,766
179
31
0
177,280
8
4,424
3,147
197
24
0
162,048
16
5,599
3,366
197
26
0
166,400
64
5,489
4,703
197
37
0
192,256
72
5,637
4,903
197
39
0
196,864
14.7.3 DDR2 and DDR3 Resource Utilization in Stratix III Devices
The following table shows typical resource usage of the DDR2 and DDR3 SDRAM
controllers with UniPHY in the current version of Quartus Prime software for Stratix III
devices.
Table 122.
Resource Utilization in Stratix III Devices
Protocol
Memory
Width
(Bits)
Combinatio
nal ALUTS
Logic
Registers
Mem ALUTs
M9K Blocks
M144K
Blocks
Memory
(Bits)
Controller
DDR2 (Half
rate)
DDR2 (Full
rate)
DDR3 (Half
rate)
8
1,807
1,058
0
4
0
4,464
16
1,809
1,058
0
6
0
8,816
64
1,810
1,272
10
14
0
32,256
72
1,842
1,090
10
17
0
39,168
8
1,856
1,093
0
4
0
2,288
16
1,855
1,092
0
4
0
4,464
64
1,841
1,092
0
10
0
17,520
72
1,834
1,092
0
11
0
19,696
8
1,861
1,083
0
4
0
4,464
16
1,863
1,083
0
6
0
8,816
64
1,878
1,295
10
14
0
32,256
72
1,895
1,115
10
17
0
39,168
8
2,591
2,100
218
6
1
157,696
16
2,762
2,320
218
6
1
157,696
64
3,672
3,658
242
6
1
157,696
72
3,814
3,877
242
6
1
PHY
DDR2 (Half
rate)
157,696
continued...
External Memory Interface Handbook Volume 3: Reference Material
430
14 Introduction to UniPHY IP
Protocol
DDR2 (Full
rate)
DDR3 (Half
rate)
Memory
Width
(Bits)
Combinatio
nal ALUTS
Logic
Registers
Mem ALUTs
M9K Blocks
M144K
Blocks
Memory
(Bits)
8
2,510
1,986
200
6
1
157,696
16
2,666
2,200
200
6
1
157,696
64
3,571
3,504
224
6
1
157,696
72
3,731
3,715
224
6
1
157,696
8
2,591
2,094
224
6
1
157,696
16
2,765
2,314
224
6
1
157,696
64
3,680
3,653
248
6
1
157,696
72
3,819
3,871
248
6
1
157,696
8
4,398
3,158
218
10
1
162,160
16
4,571
3,378
218
12
1
166,512
64
5,482
4,930
252
20
1
189,952
72
5,656
4,967
252
23
1
196,864
8
4,366
3,079
200
10
1
159,984
16
4,521
3,292
200
10
1
162,160
64
5,412
4,596
224
16
1
175,216
72
5,565
4,807
224
17
1
177,392
8
4,452
3,177
224
10
1
162,160
16
4,628
3,397
224
12
1
166,512
64
5,558
4,948
258
20
1
189,952
72
5,714
4,986
258
23
1
196,864
Total
DDR2 (Half
rate)
DDR2 (Full
rate)
DDR3 (Half
rate)
14.7.4 DDR2 and DDR3 Resource Utilization in Stratix IV Devices
The following table shows typical resource usage of the DDR2 and DDR3 SDRAM
controllers with UniPHY in the current version of Quartus Prime software for Stratix IV
devices.
Table 123.
Resource Utilization in Stratix IV Devices
Protocol
Memory
Width
(Bits)
Combinatio
nal ALUTS
Logic
Registers
Mem ALUTs
M9K Blocks
M144K
Blocks
Memory
(Bits)
Controller
DDR2 (Half
rate)
8
1,785
1,090
10
2
0
4,352
16
1,785
1,090
10
4
0
8,704
64
1,796
1,106
10
15
0
34,560
72
1,798
1,090
10
17
0
39,168
continued...
External Memory Interface Handbook Volume 3: Reference Material
431
14 Introduction to UniPHY IP
Protocol
DDR2 (Full
rate)
DDR3 (Half
rate)
Memory
Width
(Bits)
Combinatio
nal ALUTS
Logic
Registers
Mem ALUTs
M9K Blocks
M144K
Blocks
Memory
(Bits)
8
1,843
1,124
10
2
0
2,176
16
1,845
1,124
10
2
0
4,352
64
1,832
1,124
10
8
0
17,408
72
1,834
1,124
10
9
0
19,584
8
1,862
1,115
10
2
0
4,352
16
1,874
1,115
10
4
0
8,704
64
1,880
1,131
10
15
0
34,560
72
1,886
1,115
10
17
0
39,168
8
2,558
2,041
183
6
1
157,696
16
2,728
2,262
183
6
1
157,696
64
3,606
3,581
183
6
1
157,696
72
3,748
3,800
183
6
1
157,696
8
2,492
1,934
169
6
1
157,696
16
2,652
2,148
169
6
1
157,696
64
3,522
3,428
169
6
1
157,696
72
3,646
3,641
169
6
1
157,696
8
2,575
2,031
187
6
1
157,696
16
2,732
2,251
187
6
1
157,696
64
3,602
3,568
187
6
1
157,696
72
3,750
3,791
187
6
1
157,696
8
4,343
3,131
193
8
1
162,048
16
4,513
3,352
193
10
1
166,400
64
5,402
4,687
193
21
1
192,256
72
5,546
4,890
193
23
1
196,864
8
4,335
3,058
179
8
1
159,872
16
4,497
3,272
179
8
1
162,048
64
5,354
4,552
179
14
1
175,104
72
5,480
4,765
179
15
1
177,280
8
4,437
3,146
197
8
1
162,048
16
4,606
3,366
197
10
1
166,400
64
5,482
4,699
197
21
1
192,256
72
5,636
4,906
197
23
1
196,864
PHY
DDR2 (Half
rate)
DDR2 (Full
rate)
DDR3 (Half
rate)
Total
DDR2 (Half
rate)
DDR2 (Full
rate)
DDR3 (Half
rate)
External Memory Interface Handbook Volume 3: Reference Material
432
14 Introduction to UniPHY IP
14.7.5 DDR2 and DDR3 Resource Utilization in Arria V GZ and Stratix V
Devices
The following table shows typical resource usage of the DDR2 and DDR3 SDRAM
controllers with UniPHY in the current version of Quartus Prime software for Arria V GZ
and Stratix V devices.
Table 124.
Resource Utilization in Arria V GZ and Stratix V Devices
Protocol
Memory Width
(Bits)
Combinational
LCs
Logic Registers
M20K Blocks
Memory (Bits)
Controller
DDR2 (Half rate)
DDR2 (Full rate)
DDR3 (Quarter
rate)
DDR3 (Half rate)
8
1,787
1,064
2
4,352
16
1,794
1,064
4
8,704
64
1,830
1,070
14
34,304
72
1,828
1,076
15
38,400
8
2,099
1,290
2
2,176
16
2,099
1,290
2
4,352
64
2,126
1,296
7
16,896
72
2,117
1,296
8
19,456
8
2,101
1,370
4
8,704
16
2,123
1,440
7
16,896
64
2,236
1,885
28
69,632
72
2,102
1,870
30
74,880
8
1,849
1,104
2
4,352
16
1,851
1,104
4
8,704
64
1,853
1,112
14
34,304
72
1,889
1,116
15
38,400
8
2,567
1,757
13
157,696
16
2,688
1,809
13
157,696
64
3,273
2,115
13
157,696
72
3,377
2,166
13
157,696
8
2,491
1,695
13
157,696
16
2,578
1,759
13
157,696
64
3,062
2,137
13
157,696
72
3,114
2,200
13
157,696
8
2,209
2,918
18
149,504
16
2,355
3,327
18
157,696
64
3,358
5,228
18
182,272
72
4,016
6,318
18
198,656
PHY
DDR2 (Half rate)
DDR2 (Full rate)
DDR3 (Quarter
rate)
continued...
External Memory Interface Handbook Volume 3: Reference Material
433
14 Introduction to UniPHY IP
Protocol
Memory Width
(Bits)
DDR3 (Half rate)
Combinational
LCs
Logic Registers
M20K Blocks
Memory (Bits)
8
2,573
1,791
13
157,696
16
2,691
1,843
13
157,696
64
3,284
2,149
13
157,696
72
3,378
2,200
13
157,696
8
4,354
2,821
15
162,048
16
4,482
2,873
17
166,400
64
5,103
3,185
27
192,000
72
5,205
3,242
28
196,096
8
4,590
2,985
15
159,872
16
4,677
3,049
15
162,048
64
5,188
3,433
20
174,592
72
5,231
3,496
21
177,152
8
4,897
4,844
23
158,720
16
5,065
5,318
26
175,104
64
6,183
7,669
47
252,416
72
6,705
8,744
49
274,048
8
4,422
2,895
15
162,048
16
4,542
2,947
17
166,400
64
5,137
3,261
27
192,000
72
5,267
3,316
28
196,096
Total
DDR2 (Half rate)
DDR2 (Full rate)
DDR3 (Quarter
rate)
DDR3 (Half rate)
14.7.6 QDR II and QDR II+ Resource Utilization in Arria V Devices
The following table shows typical resource usage of the QDR II and QDR II+ SRAM
controllers with UniPHY in the current version of Quartus Prime software for Arria V
devices.
Table 125.
Resource Utilization in Arria V Devices
PHY Rate
Memory
Width (Bits)
Combinationa
l ALUTs
Logic
Registers
M10K Blocks
Memory
(Bits)
Hard Memory
Controiler
Controller
Half
9
98
120
0
0
0
18
96
156
0
0
0
36
94
224
0
0
0
9
234
257
0
0
0
18
328
370
0
0
0
PHY
Half
continued...
External Memory Interface Handbook Volume 3: Reference Material
434
14 Introduction to UniPHY IP
PHY Rate
Memory
Width (Bits)
Combinationa
l ALUTs
Logic
Registers
M10K Blocks
Memory
(Bits)
Hard Memory
Controiler
36
522
579
0
0
0
9
416
377
0
0
0
18
542
526
0
0
0
36
804
803
0
0
0
Total
Half
14.7.7 QDR II and QDR II+ Resource Utilization in Arria II GX Devices
The following table shows typical resource usage of the QDR II and QDR II+ SRAM
controllers with UniPHY in the current version of Quartus Prime software for
Arria II GX devices.
Table 126.
PHY Rate
Half
Full
Resource Utilization in Arria II GX Devices
Memory Width (Bits)
Combinational ALUTS
Logic Registers
Memory (Bits)
M9K Blocks
9
620
701
0
0
18
921
1122
0
0
36
1534
1964
0
0
9
584
708
0
0
18
850
1126
0
0
36
1387
1962
0
0
14.7.8 QDR II and QDR II+ Resource Utilization in Arria II GZ, Arria V GZ,
Stratix III, Stratix IV, and Stratix V Devices
The following table shows typical resource usage of the QDR II and QDR II+ SRAM
controllers with UniPHY in the current version of Quartus Prime software for
Arria II GZ, Arria V GZ, Stratix III, Stratix IV, and Stratix V devices.
Table 127.
PHY Rate
Half
Full
Resource Utilization in Arria II GZ, Arria V GZ, Stratix III, Stratix IV, and
Stratix V Devices
Memory Width (Bits)
Combinational ALUTS
Logic Registers
Memory (Bits)
M9K Blocks
9
602
641
0
0
18
883
1002
0
0
36
1457
1724
0
0
9
586
708
0
0
18
851
1126
0
0
36
1392
1962
0
0
14.7.9 RLDRAM II Resource Utilization in Arria V Devices
The following table shows typical resource usage of the RLDRAM II controller with
UniPHY in the current version of Quartus Prime software for Arria V devices.
External Memory Interface Handbook Volume 3: Reference Material
435
14 Introduction to UniPHY IP
Table 128.
Resource Utilization in Arria V Devices (Part 1 of 2)
PHY Rate
Memory
Width (Bits)
Combinationa
l ALUTs
Logic
Registers
M10K Blocks
Memory
(Bits)
Hard Memory
Controller
Controller
Half
9
353
303
1
288
0
18
350
324
2
576
0
36
350
402
4
1152
0
9
295
474
0
0
0
18
428
719
0
0
0
36
681
1229
0
0
0
9
705
777
1
288
0
18
871
1043
2
576
0
36
1198
1631
4
1152
0
PHY
Half
Total
Half
14.7.10 RLDRAM II Resource Utilization in Arria II GZ, Arria V GZ, Stratix
III, Stratix IV, and Stratix V Devices
The following table shows typical resource usage of the RLDRAM II controller with
UniPHY in the current version of Quartus Prime software for Arria II GZ, Arria V GZ,
Stratix III, Stratix IV, and Stratix V devices.
Table 129.
Resource Utilization in Arria II GZ, Arria V GZ, Stratix III, Stratix IV, and
Stratix V Devices (1)
PHY Rate
Half
Full
Memory Width
(Bits)
Combinational
ALUTS
Logic Registers
Memory (Bits)
M9K Blocks
9
829
763
288
1
18
1145
1147
576
2
36
1713
1861
1152
4
9
892
839
288
1
18
1182
1197
576
1
36
1678
1874
1152
2
Note to Table:
1. Half-rate designs use the same amount of memory as full-rate designs, but the data is organized in a different way
(half the width, double the depth) and the design may need more M9K resources.
14.8 Document Revision History
Date
Version
Changes
May 2017
2017.05.08
Rebranded as Intel.
October 2016
2016.10.31
Maintenance release.
continued...
External Memory Interface Handbook Volume 3: Reference Material
436
14 Introduction to UniPHY IP
Date
Version
Changes
May 2016
2016.05.02
Maintenance release.
November 2015
2015.11.02
Changed instances of Quartus II to Quartus Prime.
May 2015
2015.05.04
Maintenance release.
December 2014
2014.12.15
Maintenance release.
August 2014
2014.08.15
Removed occurrences of MegaWizard Plug-In Manager.
December 2013
2013.12.16
•
•
Removed references to ALTMEMPHY.
Removed references to HardCopy.
November 2012
2.1
•
•
•
•
Added RLDRAM 3 support.
Added LRDIMM support.
Added Arria V GZ support.
Changed chapter number from 8 to 10.
June 2012
2.0
•
•
•
Added LPDDR2 support.
Moved Protocol Support Matrix to Volume 1.
Added Feedback icon.
November 2011
1.1
•
Combined Release Information, Device Family Support, Features list, and
Unsupported
Features list for DDR2, DDR3, QDR II, and RLDRAM II.
Added Protocol Support Matrix.
Combined Resource Utilization information for DDR2, DDR3, QDR II, and
RLDRAM II.
Updated data for 11.1.
•
•
•
External Memory Interface Handbook Volume 3: Reference Material
437
15 Latency for UniPHY IP
15 Latency for UniPHY IP
Intel defines read and write latencies in terms of memory device clock cycles, which
are always full-rate. There are two types of latencies that exists while designing with
memory controllers—read and write latencies, which have the following definitions:
•
Read latency—the amount of time it takes for the read data to appear at the local
interface after initiating the read request.
•
Write latency—the amount of time it takes for the write data to appear at the
memory interface after initiating the write request.
Latency of the memory interface depends on its configuration and traffic patterns,
therefore you should simulate your system to determine precise latency values. The
numbers presented in this chapter are typical values meant only as guidelines.
Latency found in simulation may differ from latency found on the board, because
functional simulation does not consider board trace delays and differences in process,
voltage, and temperature. For a given design on a given board, the latency found may
differ by one clock cycle (for full-rate designs), or two clock cycles (for quarter-rate or
half-rate designs) upon resetting the board. The same design can yield different
latencies on different boards.
Note:
For a half-rate controller, the local side frequency is half of the memory interface
frequency. For a full-rate controller, the local side frequency is equal to the memory
interface frequency.
15.1 DDR2 SDRAM LATENCY
The following table shows the DDR2 SDRAM latency in full-rate memory clock cycles.
Table 130.
DDR2 SDRAM Controller Latency (In Full-Rate Memory Clock Cycles)
(1) (2)
Latency in Full-Rate Memory Clock Cycles
Rate
Half
Controller
Address &
Command
10
PHY
Address &
Command
EWL: 3
Memory
Maximum
Read
3–7
PHY Read
Return
6
Controller
Read
Return
4
OWL: 4
Full
5
0
3–7
4
10
Round Trip
Round Trip
Without
Memory
EWL: 26–30
EWL: 23
OWL: 27–31
OWL: 24
22–26
19
Notes to Table:
1. EWL = Even write latency
2. OWL = Odd write latency
Intel Corporation. All rights reserved. Intel, the Intel logo, Altera, Arria, Cyclone, Enpirion, MAX, Nios, Quartus
and Stratix words and logos are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other
countries. Intel warrants performance of its FPGA and semiconductor products to current specifications in
accordance with Intel's standard warranty, but reserves the right to make changes to any products and services
at any time without notice. Intel assumes no responsibility or liability arising out of the application or use of any
information, product, or service described herein except as expressly agreed to in writing by Intel. Intel
customers are advised to obtain the latest version of device specifications before relying on any published
information and before placing orders for products or services.
*Other names and brands may be claimed as the property of others.
ISO
9001:2008
Registered
15 Latency for UniPHY IP
15.2 DDR3 SDRAM LATENCY
The following table shows the DDR3 SDRAM latency in full-rate memory clock cycles.
Table 131.
DDR3 SDRAM Controller Latency (In Full-Rate Memory Clock Cycles)
(1) (2)
(3) (4)
Latency in Full-Rate Memory Clock Cycles
Rate
Quarter
Half
Controller
Address &
Command
20
10
Full
5
PHY
Address &
Command
EWER : 8
Memory
Maximum
Read
5–11
PHY Read
Return
EWER: 16
Controller
Read
Return
8
Round Trip
Round Trip
Without
Memory
EWER: 57–
63
EWER: 52
EWOR: 8
EWOR: 17
EWOR:58–
64
EWOR: 53
OWER: 11
OWER: 17
OWER:61–
67
OWER: 56
OWOR: 11
OWOR: 14
OWOR: 58–
64
OWOR: 53
EWER: 29–
35
EWER: 24
EWER: 3
5–11
EWER: 7
4
EWOR: 3
EWOR: 6
EWOR: 28–
34
EWOR: 23
OWER: 4
OWER: 6
OWER: 29–
35
OWER: 24
OWOR: 4
OWOR: 7
OWOR: 30–
36
OWOR: 25
24–30
19
0
5–11
4
10
Notes to Table:
1. EWER = Even write latency and even read latency
2. EWOR = Even write latency and odd read latency
3. OWER = Odd write latency and even read latency
4. OWOR = Odd write latency and odd read latency
15.3 LPDDR2 SDRAM LATENCY
The following table shows the LPDDR2 SDRAM latency in full-rate memory clock
cycles.
Table 132.
LPDDR2 SDRAM Controller Latency (In Full-Rate Memory Clock Cycles)
(1) (2)
(3) (4)
Latency in Full-Rate Memory Clock Cycles
Rate
Half
Controller
Address &
Command
10
PHY
Address &
Command
EWER: 3
EWOR: 3
Memory
Maximum
Read
5–11
PHY Read
Return
EWER: 7
EWOR: 6
Controller
Read
Return
4
Round Trip
Round Trip
Without
Memory
EWER: 29–
35
EWER: 24
EWOR: 28–
34
EWOR: 23
continued...
External Memory Interface Handbook Volume 3: Reference Material
439
15 Latency for UniPHY IP
Latency in Full-Rate Memory Clock Cycles
Rate
Full
Controller
Address &
Command
5
PHY
Address &
Command
Memory
Maximum
Read
PHY Read
Return
Controller
Read
Return
Round Trip
Round Trip
Without
Memory
OWER: 4
OWER: 6
OWER: 29–
35
OWER: 24
OWOR: 4
OWOR: 7
OWOR: 30–
36
OWOR: 25
24–30
19
0
5–11
4
10
Notes to Table:
1. EWER = Even write latency and even read latency
2. EWOR = Even write latency and odd read latency
3. OWER = Odd write latency and even read latency
4. OWOR = Odd write latency and odd read latency
15.4 QDR II and QDR II+ SRAM Latency
The following table shows the latency in full-rate memory clock cycles.
Table 133.
QDR II Latency (In Full-Rate Memory Clock Cycles)
(1)
Latency in Full-Rate Memory Clock Cycles
Rate
Controller
Address &
Command
PHY
Address &
Command
Memory
Maximum
Read
PHY Read
Return
Controller
Read
Return
Round Trip
Round Trip
Without
Memory
Half 1.5 RL
2
5.5
1.5
7.0
0
16
14.5
Half 2.0 RL
2
5.5
2.0
6.5
0
16
14.0
Half 2.5 RL
2
5.5
2.5
6.0
0
16
13.5
Full 1.5 RL
2
1.5
1.5
4.0
1
10
8.5
Full 2.0 RL
2
1.5
2.0
4.5
1
11
9.0
Full 2.5 RL
2
1.5
2.5
4.0
1
11
8.5
Note to Table:
1. RL = Read latency
15.5 RLDRAM II Latency
The following table shows the latency in full-rate memory clock cycles.
External Memory Interface Handbook Volume 3: Reference Material
440
15 Latency for UniPHY IP
Table 134.
RLDRAM II Latency (In Full-Rate Memory Clock Cycles)
(1) (2)
Latency in Full-Rate Memory Clock Cycles
Rate
Controller
Address &
Command
Half
4
PHY
Address &
Command
EWL: 1
Memory
Maximum
Read
3–8
EWL: 4
OWL: 2
Full
2
1
PHY Read
Return
Controller
Read
Return
0
OWL: 4
3–8
4
0
Round Trip
Round Trip
Without
Memory
EWL: 12–17
EWL: 9
OWL: 13–18
OWL: 10
10–15
7
Notes to Table:
1. EWL = Even write latency
2. OWL = Odd write latency
15.6 RLDRAM 3 Latency
The following table shows the latency in full-rate memory clock cycles.
Table 135.
RLDRAM 3 Latency (In Full-Rate Memory Clock Cycles)
Latency in Full-Rate Memory Clock Cycles
Rate
PHY Address
& Command
Memory
Maximum
Read
PHY Read
Return
Controller
Read Return
Round Trip
Round Trip
Without
Memory
Quarter
7
3–16
18
0
28–41
25
Half
4
3–16
6
0
13–26
10
15.7 Variable Controller Latency
The variable controller latency feature allows you to take advantage of lower latency
for variations designed to run at lower frequency. When deciding whether to vary the
controller latency from the default value of 1, be aware of the following
considerations:
•
Reduced latency can help achieve a reduction in resource usage and clock cycles
in the controller, but might result in lower fMAX.
•
Increased latency can help acheive greater fMAX, but might consume more clock
cycles in the controller and result in increased resource usage.
If you select a latency value that is inappropriate for the target frequency, the system
displays a warning message in the text area at the bottom of the parameter editor.
You can change the controller latency by altering the value of the Controller Latency
setting in the Controller Settings section of the General Settings tab of the QDR II
and QDR II+ SRAM controller with UniPHY parameter editor.
15.8 Document Revision History
External Memory Interface Handbook Volume 3: Reference Material
441
15 Latency for UniPHY IP
Date
Version
Changes
May 2017
2017.05.08
Rebranded as Intel.
October 2016
2016.10.31
Maintenance release.
May 2016
2016.05.02
Maintenance release.
November 2015
2015.11.02
Maintenance release.
May 2015
2015.05.04
Maintenance release.
December 2014
2014.12.15
Maintenance release.
August 2014
2014.08.15
Maintenance release.
December 2013
2013.12.16
Updated latency data for QDR II/II+.
November 2012
2.1
•
•
Added latency information for RLDRAM 3.
Changed chapter number from 9 to 11.
June 2012
2.0
•
•
Added latency information for LPDDR2.
Added Feedback icon.
November 2011
1.0
•
Consolidated latency information from 11.0 DDR2 and DDR3 SDRAM
Controller with UniPHY User Guide, QDR II and QDR II+ SRAM Controller
with UniPHY User Guide, and RLDRAM II Controller with UniPHY IP User
Guide.
Updated data for 11.1.
•
External Memory Interface Handbook Volume 3: Reference Material
442
16 Timing Diagrams for UniPHY IP
16 Timing Diagrams for UniPHY IP
The following topics contain timing diagrams for UniPHY-based external memory
interface IP for supported protocols.
16.1 DDR2 Timing Diagrams
This topic contains timing diagrams for UniPHY-based external memory interface IP for
DDR2 protocols.
Intel Corporation. All rights reserved. Intel, the Intel logo, Altera, Arria, Cyclone, Enpirion, MAX, Nios, Quartus
and Stratix words and logos are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other
countries. Intel warrants performance of its FPGA and semiconductor products to current specifications in
accordance with Intel's standard warranty, but reserves the right to make changes to any products and services
at any time without notice. Intel assumes no responsibility or liability arising out of the application or use of any
information, product, or service described herein except as expressly agreed to in writing by Intel. Intel
customers are advised to obtain the latest version of device specifications before relying on any published
information and before placing orders for products or services.
*Other names and brands may be claimed as the property of others.
ISO
9001:2008
Registered
16 Timing Diagrams for UniPHY IP
The following figures present timing diagrams based on a Stratix III device:
Figure 183. Full-Rate DDR2 SDRAM Read
afi_clk
avl_ready
[1]
avl_read_req
avl_size[1:0]
X
2
X
avl_addr[24:0]
X
0
X
avl_burstbegin
avl_rdata_valid
X
avl_rdata[15:0]
X
afi_cs_n
[8]
afi_ras_n
afi_cas_n
[2]
[4]
afi_we_n
afi_ba[1:0]
afi_addr[13:0]
X
0
X
0
X
X
0
X
0
X
afi_rdata_en_full
afi_rdata_en
afi_rdata_valid
X
afi_rdata[15:0]
X
[7]
mem_ck
mem_cs_n
mem_ras_n
[3]
mem_cas_n
[5]
mem_we_n
mem_ba[1:0]
X
0
X
0
X
mem_a[13:0]
X
0
X
0
X
mem_dqs
mem_dq[7:0]
Notes for the above Figure:
1.
Controller receives read command.
2.
Controller issues activate command to PHY.
3.
PHY issues activate command to memory.
4.
Controller issues read command to PHY.
5.
PHY issues read command to memory.
6.
PHY receives read data from memory.
7.
Controller receives read data from PHY.
8.
User logic receives read data from controller.
External Memory Interface Handbook Volume 3: Reference Material
444
[6]
16 Timing Diagrams for UniPHY IP
Figure 184. Full-Rate DDR2 SDRAM Write
afi_clk
avl_ready
avl_write_req
avl_size[1:0]
avl_addr[24:0]
avl_burstbegin
avl_wdata[15:0]
avl_be[1:0]
[1]
X
X
X
2
0
X
X
X
X
3
1
X
[2]
afi_cs_n
afi_ras_n
[3]
afi_cas_n
afi_we_n
afi_ba[1:0]
X
0
X
0
X
afi_addr[13:0]
X
0
X
0
X
[5]
afi_dqs_burst
[7]
afi_wdata_valid
afi_wdata[15:0]
X
afi_dm_[1:0]
X
X
0
2
X
05
afi_wlat[5:0]
mem_ck
mem_cs_n
mem_ras_n
[4]
mem_cas_n
[6]
mem_we_n
mem_ba[1:0]
X
0
X
0
X
mem_a[13:0]
X
0
X
0
X
mem_dqs
[8]
mem_dq[7:0]
mem_dm
Notes for the above Figure:
1.
Controller receives write command.
2.
Controller receives write data.
3.
Controller issues activate command to PHY.
4.
PHY issues activate command to memory.
5.
Controller issues write command to PHY.
6.
PHY issues write command to memory.
7.
Controller sends write data to PHY.
8.
PHY sends write data to memory.
External Memory Interface Handbook Volume 3: Reference Material
445
16 Timing Diagrams for UniPHY IP
Figure 185. Half-Rate DDR2 SDRAM Read
afi_clk
avl_ready
[1]
avl_read_req
avl_size[1:0]
X
2
X
avl_addr[23:0]
X
0
X
avl_burstbegin
avl_rdata_valid
X
avl_rdata[31:0]
afi_cs_n[1:0]
3
1
afi_ras_n[1:0]
3
0
3
1
[8]
3
3
0
3
afi_cas_n[1:0]
X
3
[2]
3
afi_we_n[1:0]
[4]
afi_ba[3:0]
X
0
X
0
X
afi_addr[27:0]
X
0
X
0
X
afi_rdata_en_full[1:0]
0
3
0
afi_rdata_en[1:0]
0
3
0
afi_rdata_valid
afi_rdata[31:0]
X
X
mem_ck
[7]
mem_cs_n
mem_ras_n
[3]
mem_cas_n
[5]
mem_we_n
mem_ba[1:0]
X
0
X
0
X
mem_a[13:0]
X
0
X
0
X
mem_dqs
mem_dq[7:0]
Notes for the above Figure:
1.
Controller receives read command.
2.
Controller issues activate command to PHY.
3.
PHY issues activate command to memory.
4.
Controller issues read command to PHY.
5.
PHY issues read command to memory.
6.
PHY receives read data from memory.
7.
Controller receives read data from PHY.
8.
User logic receives read data from controller.
External Memory Interface Handbook Volume 3: Reference Material
446
[6]
16 Timing Diagrams for UniPHY IP
Figure 186. Half-Rate DDR2 SDRAM Write
afi_clk
avl_ready
[1]
avl_write_req
avl_size[1:0]
X
2
X
avl_addr[23:0]
X
0
X
avl_burstbegin
avl_wdata[15:0]
X
avl_be[3:0]
X
X
B D
X
[2]
afi_cs_n[1:0]
3
1
afi_ras_n[1:0]
3
0
afi_cas_n[1:0]
3
0
3
afi_we_n[1:0]
3
0
3
3
3
1
3
afi_ba[3:0]
X
0
X
0
X
afi_addr[27:0]
X
0
X
0
X
afi_dqs_burst[1:0]
0
afi_wdata_valid[1:0]
0
afi_wdata[31:0]
X
afi_dm_[3:0]
X
2
[3]
[5]
3
0
3
0
[7]
X
X
4 2
02
afi_wlat[5:0]
mem_ck
mem_cs_n
mem_ras_n
[4]
mem_cas_n
[6]
mem_we_n
mem_ba[1:0]
X
0
X
0
X
mem_a[13:0]
X
0
X
0
X
[8]
mem_dqs
mem_dq[7:0]
mem_dm
Notes for the above Figure:
1.
Controller receives write command.
2. Controller receives write data.
3. Controller issues activate command to PHY.
4.
PHY issues activate command to memory.
5.
Controller issues write command to PHY.
External Memory Interface Handbook Volume 3: Reference Material
447
16 Timing Diagrams for UniPHY IP
6.
PHY issues write command to memory.
7.
Controller sends write data to PHY.
8.
PHY sends write data to memory.
16.2 DDR3 Timing Diagrams
This topic contains timing diagrams for UniPHY-based external memory interface IP for
DDR3 protocols.
External Memory Interface Handbook Volume 3: Reference Material
448
16 Timing Diagrams for UniPHY IP
The following figures present timing diagrams based on a Stratix III device
Figure 187. Half-Rate DDR3 SDRAM Read
afi_clk
avl_ready
[1]
avl_read_req
avl_size[1:0]
2
X
avl_addr[25:0]
0
X
avl_burstbegin
avl_rdata_valid
X
avl_rdata[31:0]
afi_cs_n[1:0]
3
1
afi_ras_n[1:0]
3
0
3
[8]
3
1
3
3
afi_cas_n[1:0]
X
3
0
[2]
[4]
3
afi_we_n[1:0]
afi_ba[5:0]
X
0
X
0
X
afi_addr[29:0]
X
0
X
0
X
afi_rdata_en_full[1:0]
0
3
0
afi_rdata_en[1:0]
0
3
0
[7]
afi_rdata_valid[1:0]
X
afi_rdata[31:0]
X
mem_ck
mem_cs_n
mem_ras_n
[3]
mem_cas_n
[5]
mem_we_n
mem_ba[2:0]
X
0
X
0
X
mem_a[14:0]
X
0
X
0
X
mem_dqs
mem_dq[7:0]
External Memory Interface Handbook Volume 3: Reference Material
449
[6]
16 Timing Diagrams for UniPHY IP
Notes for the above Figure:
1.
Controller receives read command.
2.
Controller issues activate command to PHY.
3.
PHY issues activate command to memory.
4.
Controller issues read command to PHY.
5.
PHY issues read command to memory.
6.
PHY receives read data from memory.
7.
Controller receives read data from PHY.
8.
User logic receives read data from controller.
External Memory Interface Handbook Volume 3: Reference Material
450
16 Timing Diagrams for UniPHY IP
Figure 188. Half-Rate DDR3 SDRAM Writes
afi_clk
avl_ready
[1]
avl_write_req
avl_size[1:0]
X
2
X
avl_addr[25:0]
X
0
1X
avl_burstbegin
avl_wdata[31:0]
avl_be[3:0]
X
X
X
1 8
X
3
afi_cs_n[1:0]
3
1
afi_ras_n[1:0]
3
0
afi_cas_n[1:0]
3
0
3
[3]
afi_we_n[1:0]
3
0
3
[5]
afi_ba[5:0]
X
0
X
0
X
afi_addr[29:0]
X
0
X
0
X
afi_dqs_burst[1:0]
0
afi_wdata_valid[1:0]
0
afi_wdata[31:0]
X
afi_dm[3:0]
X
3
[2]
1
3
2
3
0
3
0
X
E
7
[7]
X
03
afi_wlat[5:0]
mem_ck
mem_cs_n
mem_ras_n
[4]
[6]
mem_cas_n
mem_we_n
mem_ba[2:0]
X
0
X
0
X
mem_a[14:0]
X
0
X
0
X
[8]
mem_dqs
mem_dq[7:0]
mem_dm
External Memory Interface Handbook Volume 3: Reference Material
451
16 Timing Diagrams for UniPHY IP
Notes for the above Figure:
1.
Controller receives write command.
2.
Controller receives write data.
3.
Controller issues activate command to PHY.
4.
PHY issues activate command to memory.
5.
Controller issues write command to PHY.
6.
PHY issues write command to memory.
7.
Controller sends write data to PHY.
8.
PHY sends write data to memory.
External Memory Interface Handbook Volume 3: Reference Material
452
16 Timing Diagrams for UniPHY IP
Figure 189. Quarter-Rate DDR3 SDRAM Reads
afi_clk
avl_ready
[1]
avl_read_req
avl_size[2:0]
X
1
X
avl_addr[24:0]
X
0
X
avl_burstbegin
avl_rdata_valid
X
X
avl_rdata[63:0]
afi_cs_n[3:0]
F
7
afi_ras_n[3:0]
F
0
F
[8]
F
0
F
afi_cas_n[3:0]
F
7
F
[2]
F
afi_we_n[3:0]
[4]
afi_ba[11:0]
X
0
X
0
X
afi_addr[59:0]
X
0
X
0
X
afi_rdata_en_full[3:0]
0
F
0
afi_rdata_en[3:0]
0
F
0
[7]
afi_rdata_valid[3:0]
X
afi_rdata[63:0]
X
X
mem_ck
mem_cs_n
mem_ras_n
[3]
mem_cas_n
[5]
mem_we_n
mem_ba[2:0]
X
0
X
0
X
mem_a[14:0]
X
0
X
0
X
[6]
mem_dqs
mem_dq[7:0]
External Memory Interface Handbook Volume 3: Reference Material
453
16 Timing Diagrams for UniPHY IP
Notes for the above Figure:
1.
Controller receives read command.
2.
Controller issues activate command to PHY.
3.
PHY issues activate command to memory.
4.
Controller issues read command to PHY.
5.
PHY issues read command to memory.
6.
PHY receives read data from memory
7.
Controller receives read data from PHY
8.
User logic receives read data from controller.
External Memory Interface Handbook Volume 3: Reference Material
454
16 Timing Diagrams for UniPHY IP
Figure 190. Quarter-Rate DDR3 SDRAM Writes
afi_clk
avl_ready
[1]
avl_write_req
avl_size[2:0]
0
1
0
avl_addr[24:0]
X
0
X
avl_burstbegin
avl_wdata[63:0]
X
X
avl_be[7:0]
X
X
afi_cs_n[3:0]
F
7
afi_ras_n[3:0]
F
0
F
[2]
7
F
afi_cas_n[3:0]
F
0
F
[3]
afi_we_n[3:0]
F
0
F
[5]
afi_ba[11:0]
X
0
X
0
X
afi_addr[59:0]
X
0
X
0
X
afi_dqs_burst[3:0]
0
afi_wdata_valid[3:0]
8
F
0
0
F
0
X
EC
X
[7]
afi_wdata[63:0]
afi_dm[7:0]
02
afi_wlat[5:0]
mem_ck
mem_cs_n
mem_ras_n
[4]
[6]
mem_cas_n
mem_we_n
mem_ba[2:0]
X
0
X
0
X
mem_a[14:0]
X
0
X
0
X
mem_dqs
[8]
mem_dq[7:0]
mem_dm
External Memory Interface Handbook Volume 3: Reference Material
455
16 Timing Diagrams for UniPHY IP
Notes for the above Figure:
1.
Controller receives write command.
2.
Controller receives write data.
3.
Controller issues activate command to PHY
4.
PHY issues activate command to memory.
5.
Controller issues write command to PHY
6.
PHY issues write command to memory
7.
Controller sends write data to PHY
8.
PHY sends write data to memory.
16.3 QDR II and QDR II+ Timing Diagrams
This topic contains timing diagrams for UniPHY-based external memory interface IP for
QDR II and QDR II+ protocols.
External Memory Interface Handbook Volume 3: Reference Material
456
16 Timing Diagrams for UniPHY IP
The following figures present timing diagrams, based on a Stratix III device:
Figure 191. Half-Rate QDR II and QDR II+ SRAM Read
afi_clk
avl_r_ready
[1]
avl_r_read_req
avl_r_size[2:0]
X
2
X
avl_r_addr[19:0]
X
0
X
avl_r_rdata_valid
[6]
X
avl_r_rdata[71:0]
afi_rps_n[1:0]
X
3
1
3
3
afi_wps_n[1:0]
afi_addr[39:0]
[2]
X
0 1
X
afi_rdata_en_full
afi_rdata_en
afi_rdata_valid
[5]
X
afi_rdata[71:0]
X
mem_k
mem_rps_n
[3]
mem_wps_n
mem_a[19:0]
X
0X1
X
mem_cq
[4]
mem_q[17:0]
Notes for the above Figure:
1.
Controller receives read command.
2.
Controller issues two read commands to PHY.
3.
PHY issues two read commands to memory.
4.
PHY receives read data from memory.
5.
Controller receives read data from PHY.
6.
User logic receives read data from controller.
External Memory Interface Handbook Volume 3: Reference Material
457
16 Timing Diagrams for UniPHY IP
Figure 192. Half-Rate QDR II and QDR II+ SRAM Write
afi_clk
avl_w_ready
[1]
avl_w_write_req
avl_w_size[2:0] X
2
X
avl_w_addr[19:0] X
0
X
3
afi_rps_n[1:0]
afi_wps_n[1:0]
afi_addr[39:0]
[2]
X
avl_w_wdata[71:0] X
3
X
afi_wdata_valid[1:0]
0
X
afi_bws_n[7:0]
X
3
X
1
0
3
0
afi_wdata[71:0]
[3]
2
X
[4]
X
0
mem_k
mem_rps_n
mem_wps_n
mem_a[19:0]
[5]
X
0 X 1
X
mem_d[17:0]
mem_bws_n[1:0]
[6]
X
0
Notes for the above Figure:
1. Controller receives write command.
2.
Controller receives write data.
3.
Controller issues two write commands to PHY.
4. Controller sends write data to PHY.
5.
PHY issues two write commands to memory.
6.
PHY sends write data to memory.
External Memory Interface Handbook Volume 3: Reference Material
458
X
16 Timing Diagrams for UniPHY IP
Figure 193. Full-Rate QDR II and QDR II+ SRAM Read
afi_clk
avl_r_ready
[1]
avl_r_read_req
avl_r_size[2:0]
X
2
X
avl_r_addr[20:0]
X
0
X
avl_r_rdata_valid
X
avl_r_rdata[35:0]
X
[6]
afi_rdata_valid
afi_rdata_en_full
afi_rdata_en
3
afi_wps_n[1:0]
afi_addr[41:0]
afi_rps_n[1:0]
1
0
X
3
X
3
2
X
afi_rdata[35:0]
X
[2]
[5]
mem_k
mem_rps_n
[3]
mem_wps_n
mem_a[20:0]
X
0 X 1
X
mem_cq
[4]
mem_q[17:0]
Notes for the above Figure:
1.
Controller receives read command.
2.
Controller issues two read commands to PHY.
3.
PHY issues two read commands to memory.
4.
PHY receives read data from memory.
5.
Controller receives read data from PHY.
6.
User logic receives read data from controller.
External Memory Interface Handbook Volume 3: Reference Material
459
16 Timing Diagrams for UniPHY IP
Figure 194. Full-Rate QDR II and QDR II+ SRAM Write
afi_clk
avl_w_ready
[1]
avl_w_write_req
avl_w_size[2:0]
X
2
X
avl_w_addr[20:0]
X
0
X
avl_w_wdata[35:0]
X
[2]
X
3
afi_rps_n[1:0]
afi_wps_n[1:0]
3
afi_addr[41:0]
X
1
0
[3]
3
2
X
[4]
afi_wdata_valid
afi_wdata[35:0]
X
afi_bws_n[3:0]
X
X
X
0
mem_k
mem_rps_n
mem_wps_n
[5]
X
mem_a[20:0]
0X 1
X
mem_d[17:0]
[6]
mem_bws_n[1:0]
X
0
X
Notes for the above Figure:
1.
Controller receives write command.
2.
Controller receives write data.
3.
Controller issues two write commands to PHY.
4.
Controller sends write data to PHY.
5.
PHY issues two write commands to memory.
6.
PHY sends write data to memory.
16.4 RLDRAM II Timing Diagrams
This topic contains timing diagrams for UniPHY-based external memory interface IP for
RLDRAM protocols.
External Memory Interface Handbook Volume 3: Reference Material
460
16 Timing Diagrams for UniPHY IP
The following figures present timing diagrams, based on a Stratix III device:
Figure 195. Half-Rate RLDRAM II Read
afi_clk
avl_ready
[1]
avl_read_req
avl_size[2:0]
X
2
X
avl_addr[22:0]
X
0
X
avl_rdata_valid
afi_cs_n[1:0]
3
[6]
X
X
avl_rdata[71:0]
1
[2]
3
afi_we_n[1:0]
3
afi_ref_n[1:0]
3
afi_ba[5:0]
X
0
X
afi_addr[37:0]
X
0
X
afi_rdata_en_full
afi_rdata_en
afi_rdata_valid
[5]
X
afi_rdata[71:0]
X
mem_ck
mem_cs_n
[3]
mem_we_n
mem_ref_n
mem_ba[2:0]
X
0
X
mem_a[18:0]
X
0
X
mem_qk
[4]
mem_dq[17:0]
External Memory Interface Handbook Volume 3: Reference Material
461
16 Timing Diagrams for UniPHY IP
Notes for the above Figure:
1.
Controller receives read command.
2.
Controller issues read command to PHY.
3.
PHY issues read command to memory.
4.
PHY receives read data from memory.
5.
Controller receives read data from PHY.
6.
User logic receives read data from controller.
External Memory Interface Handbook Volume 3: Reference Material
462
16 Timing Diagrams for UniPHY IP
Figure 196. Half-Rate RLDRAM II Write
afi_clk
avl_ready
[1]
avl_write_req
avl_size[2:0]
X
2
X
avl_addr[22:0]
X
2
X
avl_wdata[71:0]
X
[2]
X
afi_cs_n[1:0]
3
1
3
afi_we_n[1:0]
3
1
3
[3]
3
afi_ref_n[1:0]
afi_ba[5:0]
X
0
X
afi_addr[37:0]
X
0
X
afi_wdata_valid[1:0]
0
afi_wdata[71:0]
X
afi_dm[3:0]
3
3
0
[5]
X
0
3
mem_ck
mem_cs_n
mem_we_n
[4]
mem_ref_n
mem_ba[2:0]
X
0
X
mem_a[18:0]
X
0
X
mem_dk
[6]
mem_dq[17:0]
mem_dm
External Memory Interface Handbook Volume 3: Reference Material
463
16 Timing Diagrams for UniPHY IP
Notes for the above Figure:
1.
Controller receives write command.
2.
Controller receives write data.
3.
Controller issues write command to PHY.
4.
PHY issues write command to memory.
5.
Controller sends write data to PHY.
6.
PHY sends write data to memory.
External Memory Interface Handbook Volume 3: Reference Material
464
16 Timing Diagrams for UniPHY IP
Figure 197. Full-Rate RLDRAM II Read
afi_clk
avl_ready
[1]
avl_read_req
avl_size[2:0]
X
2
X
avl_addr[23:0]
X
0
X
avl_rdata_valid
X
avl_rdata[35:0]
X
afi_cs_n
[6]
[2]
afi_we_n
afi_ref_n
afi_ba[2:0]
X
0
X
afi_addr[19:0]
X
0
X
afi_rdata_en_full
afi_rdata_en
afi_rdata_valid
X
afi_rdata[35:0]
X
[5]
mem_ck
[3]
mem_cs_n
mem_we_n
mem_ref_n
mem_ba[2:0]
X
0
X
mem_a[19:0]
X
0
X
mem_qk
[4]
mem_dq[17:0]
External Memory Interface Handbook Volume 3: Reference Material
465
16 Timing Diagrams for UniPHY IP
Notes for the above Figure:
1.
Controller receives read command.
2.
Controller issues read command to PHY.
3.
PHY issues read command to memory.
4.
PHY receives read data from memory.
5.
Controller receives read data from PHY.
6.
User logic receives read data from controller.
Figure 198. Full-Rate RLDRAM II Write
afi_clk
avl_ready
[1]
avl_write_req
avl_size[2:0]
X
2
X
avl_addr[23:0]
X
0
X
avl_wdata[35:0]
X
[2]
X
afi_cs_n
afi_we_n
[3]
afi_ref_n
afi_ba[2:0]
X
0
X
afi_addr[19:0]
X
0
X
afi_wdata_valid
afi_wdata[35:0]
X
afi_dm[1:0]
3
X
0
[5]
3
mem_ck
mem_cs_n
mem_we_n
[4]
mem_ref_n
mem_ba[2:0]
X
0
X
mem_a[19:0]
X
0
X
mem_dk
mem_dq[17:0]
mem_dm
External Memory Interface Handbook Volume 3: Reference Material
466
[6]
16 Timing Diagrams for UniPHY IP
Notes for the above Figure:
1.
Controller receives write command.
2.
Controller receives write data.
3.
Controller issues write command to PHY.
4.
PHY issues write command to memory.
5.
Controller sends write data to PHY.
6.
PHY sends write data to memory.
16.5 LPDDR2 Timing Diagrams
This topic contains timing diagrams for UniPHY-based external memory interface IP for
LPDDR2 protocols.
External Memory Interface Handbook Volume 3: Reference Material
467
16 Timing Diagrams for UniPHY IP
Figure 199. Half-Rate LPDDR2 Read
[8]
[1]
afi_clk
avl_ready
avl_read_req
avl_size[2:0]
X
1
X
avl_addr[22:0]
X
1
X
avl_burstbegin
avl_rdata_valid
ABA...
avl_rdata[31:0]
afi_addr[39:0]
X
afi_cs_n[1:0]
X
X
2
X
afi_rdata_en_full[1:0]
0
afi_rdata_en[1:0]
0
X
1
X
3
0
3
0
0
afi_rdata_valid[1:0]
X
afi_rdata[31:0]
3
0
AB
mem_ck
mem_cs_n
mem_ca[9:0]
X
X
X
mem_dqs
mem_dq[7:0]
[2]
[3] [4]
Notes for the above Figure:
1.
Controller receives read command.
2.
Controller issues activate command to PHY.
3.
PHY issues activate command to memory.
4.
Controller issues read command to PHY.
5.
PHY issues read command to memory.
External Memory Interface Handbook Volume 3: Reference Material
468
[5]
[6]
[7]
16 Timing Diagrams for UniPHY IP
6.
PHY receives read data from memory.
7.
Controller receives read data from PHY.
8.
User logic receives read data from controller.
Figure 200. Half-Rate LPDDR2 Write
[1] [2]
[3]
[5]
afi_clk
avl_ready
avl_write_req
avl_size[2:0]
1
avl_addr[22:0]
000000
avl_burstbegin
avl_wdata[31:0]
avl_be[7:0] X
afi_cs_n[1:0] 3
afi_addr[39:0] X
ABAB...
FF
X
2
0... X
3
1
0...
X
afi_dqs_burst[1:0] 0
2
afi_wdata_valid[1:0] 0
afi_wdata[31:0] X
3
0
3
0
ABABABAB
afi_dm[3:0] X
F
0
afi_wlat[5:0] 03
mem_ck
mem_cs_n
mem_ca[9:0] X
X
X
mem_dq[7:0]
mem_dqs
mem_dm
[4]
[6]
[7]
External Memory Interface Handbook Volume 3: Reference Material
469
[8]
16 Timing Diagrams for UniPHY IP
Notes for the above Figure:
1.
Controller receives write command.
2.
Controller receives write data.
3.
Controller issues activate command to PHY.
4.
PHY issues activate command to memory.
5.
Controller issues write command to PHY.
6.
PHY issues write command to memory.
7.
Controller sends write data to PHY.
8.
PHY sends write data to memory.
External Memory Interface Handbook Volume 3: Reference Material
470
16 Timing Diagrams for UniPHY IP
Figure 201. Full-Rate LPDDR2 Read
[1]
[2]
[8]
[4]
afi_clk
avl_ready
avl_read_req
avl_size[2:0] X
X
1
avl_addr[24:0] X
X
avl_burstbegin
avl_rdata_valid
X
avl_rdata[63:0] X
afi_cs_n
afi_addr[19:0] X
X
X
afi_rdata_en_full
afi_rdata_en
afi_rdata_valid
afi_rdata[63:0] X
X
mem_ck
mem_cs_n
mem_ca[9:0] X
X
X
0
mem_dqs[3:0]
mem_dq[31:0]
AAA...
[3]
[5]
[6]
[7]
External Memory Interface Handbook Volume 3: Reference Material
471
16 Timing Diagrams for UniPHY IP
Notes for the above Figure:
1.
Controller receives read command.
2.
Controller issues activate command to PHY.
3.
PHY issues activate command to memory.
4.
Controller issues read command to PHY.
5.
PHY issues read command to memory.
6.
PHY receives read data from memory.
7.
Controller receives read data from PHY.
8.
User logic receives read data from controller.
External Memory Interface Handbook Volume 3: Reference Material
472
16 Timing Diagrams for UniPHY IP
Figure 202. Full-Rate LPDDR2 Write
[1] [2]
[3]
[5]
afi_clk
avl_ready
avl_write_req
avl_size[2:0] X
1 X
avl_addr[24:0] X
1 X
avl_burstbegin
avl_wdata[63:0] X
AB... X
avl_be[7:0] X
FF X
afi_cs_n
afi_addr[19:0] X
0...
X
0...
X
F
afi_dqs_burst[3:0] 0
F
0
ABABABABAB...
X
afi_wdata_valid[3:0] 0
afi_wdata[63:0] X
X
afi_dm[7:0] F
0
FF
00
X
afi_wlat[5:0] 00
mem_ck
mem_cs_n
X
mem_ca[9:0] X
X
mem_dqs[3:0]
0 F 0 F 0F 0 F 0
mem_dq[31:0]
ABABABAB
0
mem_dm[3:0] F
[4]
[6]
F
[7]
External Memory Interface Handbook Volume 3: Reference Material
473
[8]
16 Timing Diagrams for UniPHY IP
Notes for the above Figure:
1.
Controller receives write command.
2.
Controller receives write data.
3.
Controller issues activate command to PHY.
4.
PHY issues activate command to memory.
5.
Controller issues write command to PHY.
6.
PHY issues write command to memory.
7.
Controller sends write data to PHY.
8.
PHY sends write data to memory.
16.6 RLDRAM 3 Timing Diagrams
This topic contains timing diagrams for UniPHY-based external memory interface IP for
RLDRAM 3 protocols.
External Memory Interface Handbook Volume 3: Reference Material
474
16 Timing Diagrams for UniPHY IP
Figure 203. Quarter-Rate RLDRAM 3 Read
[4]
afi_clk
afi_addr
afi_rdata_en_full
afi_rdata_valid
afi_we_n
afi_ba
afi_cs_n
[1]
afi_wdata
afi_rdata_en
afi_rst_n
afi_rlat
afi_rdata
afi_wdata_valid
afi_dm
mem_ck
mem_ck_n
global_reset
mem_cs_n
[2]
mem_we_n
mem_ba
mem_a
mem_qk
mem_qk_n
mem_dq
[3]
mem_dk
mem_dk_n
mem_ref_n
mem_dm
000c8000c8000c8000c8
3
7
0
0
f
7
3
0
zzzz...
c8c8c8c8c8c...
9c63
8421
0000
8
f
c
c8c8c8c8c8c8c8c8c8c8c8c8c8c8c8c8c8c8
3
7
0
f
00
00
0000
0
000c8
Notes for the above Figure:
1. Controller issues read command to PHY.
2. PHY issues read command to memory.
3. PHY receives data from memory.
4. Controller receives read data from PHY.
External Memory Interface Handbook Volume 3: Reference Material
475
16 Timing Diagrams for UniPHY IP
Figure 204. Quarter-Rate RLDRAM 3 Write
afi_clk
afi_addr
afi_rdata_en_full
afi_rdata_valid
afi_we_n
[1]
afi_ba
afi_cs_n
afi_wdata
afi_rdata_en
afi_rst_n
afi_rlat
afi_rdata
afi_wdata_valid 00
[2]
afi_dm
mem_ck
mem_ck_n
global_reset
mem_cs_n
mem_we_n
[3]
mem_ba
mem_a
mem_qk 3
mem_qk_n 0
mem_dq
[4]
mem_dk 3
mem_dk_n 0
mem_ref_n
mem_dm
000e9000e9000e9000e9
9
e
8421
9c63
9
e
e9e9e9e9e9e9e9e9e9e9e9e9e9e9e9e9e9e9
3c
f
0000
f
03
00
3 6 c
9 0
1
2 4
8
000e9
0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0
3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3
0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0
3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3
Notes for the above Figure:
1. Controller issues write command to PHY.
2. Data ready from controller for PHY.
3. PHY issues write command to memory.
4. PHY sends read data to memory.
External Memory Interface Handbook Volume 3: Reference Material
476
16 Timing Diagrams for UniPHY IP
Figure 205. Half-rate RLDRAM 3 Read
afi_clk
afi_addr
afi_rdata_en_full
afi_rdata_valid
afi_we_n
afi_ba
afi_cs_n
[1]
afi_wdata
afi_rdata_en
afi_rst_n
afi_rlat
afi_rdata
afi_wdata_valid
afi_dm
mem_ck
mem_ck_n
global_reset_n
mem_cs_n
[2]
mem_we_n
mem_a
mem_ba
mem_qk
mem_qk_n
mem_dq
[3]
mem_dk
mem_dk_n
mem_ref_n
mem_dm
[4]
0... 0000800008
0 3
1
0
3
00 21
63
2
3 0
9... 888888888888888888
0 3
1
3
00
0
3
00
3
0
xxxxxxxxx88... 888888888888888888
0
00
00009
00008
0
1
2
3
3 0 3 0 3 0 3 0 3 0 3 0 3 0
0 3 0 3 0 3 0 3 0 3 0 3 0 3
3
0
0 3 0 3 0 3 0 3 0 3 0 3 0
3 0 3 0 3 0 3 0 3 0 3 0 3
6
0
3 0 3 0 3 0 3 0 3 0 3 0 3 0 3 0
0 3 0 3 0 3 0 3 0 3 0 3 0 3 0 3
3 0 3
0 3 0
0 3 0 3 0 3 0 3 0 3 0 3 0
3 0 3 0 3 0 3 0 3 0 3 0 3
3 0 3 0
0 3 0 3
3 0 3 0
0 3 0 3
Notes for the above Figure:
1.
Controller issues read command to PHY.
2.
PHY issues read command to memory.
3.
PHY receives data from memory.
4.
Controller receives read data from PHY.
External Memory Interface Handbook Volume 3: Reference Material
477
3 0
0 3
3 0
0 3
16 Timing Diagrams for UniPHY IP
Figure 206. Half-Rate RLDRAM 3 Write
afi_clk
afi_addr 0000700007
afi_rdata_en_full 0
afi_rdata_valid 0
afi_we_n 3 1
3
[1]
afi_ba 21 63
00
3
afi_cs_n 3 1
afi_wdata 777777777777777777
afi_rdata_en 0
afi_rst_n 3
afi_rlat 00
afi_rdata fffffffffxxxxxxxxx
afi_wdata_valid 0
c
[2]
afi_dm 00
mem_ck
mem_ck_n
global_reset_n
mem_cs_n
mem_we_n
[3]
00007
mem_a 00008
mem_ba 0
1
mem_qk 3 0 3 0 3 0 3
mem_qk_n 0 3 0 3 0 3 0
mem_dq
[4]
mem_dk 3 0 3 0 3 0 3
mem_dk_n 0 3 0 3 0 3 0
mem_ref_n
mem_dm
local_cal_success
0
2
0
3
3
3 0
0 3
3
0
6
0
3
3
0
0
0
3
3
0
0
3
3
0
0
3
3
0
0
3
3
0
0
3
3
0
0
3
3
0
0
3
3 0
0 3
3
0
0
3
3
0
0
3
3
0
0
3
3
0
0
3
3
0
0
3
3
0
0
3
3
0
0
3
3
0
Notes for the above Figure:
1. Controller issues write command to PHY.
2. Data ready from controller for PHY.
3. PHY issues write command to memory.
4. PHY sends read data to memory.
16.7 Document Revision History
Date
Version
Changes
May 2017
2017.05.08
Rebranded as Intel.
October 2016
2016.10.31
Maintenance release.
May 2016
2016.05.02
Maintenance release.
November 2015
2015.11.02
Maintenance release.
May 2015
2015.05.04
Maintenance release.
continued...
External Memory Interface Handbook Volume 3: Reference Material
478
16 Timing Diagrams for UniPHY IP
Date
Version
Changes
December 2014
2014.12.15
Maintenance release.
August 2014
2014.08.15
Maintenance release.
December 2013
2013.12.16
Updated timing diagrams for LPDDR2.
November 2012
2.1
•
•
Added timing diagrams for RLDRAM 3.
Changed chapter number from 10 to 12.
June 2012
2.0
•
•
Added timing diagrams for LPDDR2.
Added Feedback icon.
November 2011
1.1
•
Consolidated timing diagrams from 11.0 DDR2 and DDR3 SDRAM
Controller with UniPHY User Guide, QDR II and QDR II+ SRAM Controller
with UniPHY User Guide, and RLDRAM II Controller with UniPHY IP User
Guide.
Added Read and Write diagrams for DDR3 quarter-rate.
•
External Memory Interface Handbook Volume 3: Reference Material
479
17 External Memory Interface Debug Toolkit
17 External Memory Interface Debug Toolkit
The EMIF Toolkit lets you run your own traffic patterns, diagnose and debug calibration
problems, and produce margining reports for your external memory interface.
The toolkit is compatible with UniPHY-based external memory interfaces that use the
Nios II-based sequencer, with toolkit communication enabled, and with Arria 10 EMIF
IP. Toolkit communication is on by default in versions 10.1 and 11.0 of UniPHY IP; for
version 11.1 and later, toolkit communication is on whenever debugging is enabled on
the Diagnostics tab of the IP core interface.
The EMIF Toolkit can communicate with several different memory interfaces on the
same device, but can communicate with only one memory device at a time.
Note:
The EMIF Debug Toolkit does not support MAX 10 devices.
17.1 User Interface
The EMIF toolkit provides a graphical user interface for communication with
connections.
All functions provided in the toolkit are also available directly from the quartus_sh
TCL shell, through the external_memif_toolkit TCL package. The availablity of
TCL support allows you to create scripts to run automatically from TCL. You can find
information about specific TCL commands by running help -pkg
external_memif_toolkit from the quartus_sh TCL shell.
If you want, you can begin interacting with the toolkit through the GUI, and later
automate your workflow by creating TCL scripts. The toolkit GUI records a history of
the commands that you run. You can see the command history on the History tab in
the toolkit GUI.
17.1.1 Communication
Communication between the EMIF Toolkit and external memory interface connections
varies, depending on the connection type and version. In versions 10.1 and 11.0 of
the EMIF IP, communication is achieved using direct communication to the Nios IIbased sequencer. In version 11.1 and later, communication is achieved using a JTAG
Avalon-MM master attached to the sequencer bus.
The following figure shows the structure of UniPHY-based IP version 11.1 and later,
with JTAG Avalon-MM master attached to sequencer bus masters.
Intel Corporation. All rights reserved. Intel, the Intel logo, Altera, Arria, Cyclone, Enpirion, MAX, Nios, Quartus
and Stratix words and logos are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other
countries. Intel warrants performance of its FPGA and semiconductor products to current specifications in
accordance with Intel's standard warranty, but reserves the right to make changes to any products and services
at any time without notice. Intel assumes no responsibility or liability arising out of the application or use of any
information, product, or service described herein except as expressly agreed to in writing by Intel. Intel
customers are advised to obtain the latest version of device specifications before relying on any published
information and before placing orders for products or services.
*Other names and brands may be claimed as the property of others.
ISO
9001:2008
Registered
17 External Memory Interface Debug Toolkit
Figure 207. UniPHY IP Version 11.1 and Later, with JTAG Avalon-MM Master
JTAG Avalon
Master
(new)
Combined
ROM/RAM
( Variable
Size )
EMIF Toolkit
Bridge
Debug Bus
Sequencer Bus
Avalon -MM
Avalon-MM
NIOS II
SCC
PHY
AFI
Tracking
Register
File
Sequencer Managers
17.1.2 Calibration and Report Generation
The EMIF Toolkit uses calibration differently, depending on the version of external
memory interface in use. For versions 10.1 and 11.0 interfaces, the EMIF Toolkit
causes the memory interface to calibrate several times, to produce the data from
which the toolkit generates its reports. In version 11.1 and later, report data is
generated during calibration, without need to repeat calibration. For version 11.1 and
later, generated reports reflect the result of the previous calibration, without need to
recalibrate unless you choose to do so.
17.2 Setup and Use
Before using the EMIF Toolkit, you should compile your design and program the target
device with the resulting SRAM Object File (. sof). For designs compiled in the Quartus
II software version 12.0 or earlier, debugging information is contained in the JTAG
Debugging Information file (.jdi); however, for designs compiled in the Quartus II
software version 12.1 or later, all debugging information resides in the .sof file.
You can run the toolkit using all your project files, or using only the Quartus Prime
Project File (.qpf), Quartus Prime Settings File (.qsf), and .sof file; the .jdi file is also
required for designs compiled prior to version 12.1. To ensure that all debugging
information is correctly synchronized for designs compiled prior to version 12.1,
ensure that the .sof and .jdi files that you used are generated during the same run of
the Quartus Prime Assembler.
After you have programmed the target device, you can run the EMIF Toolkit and open
your project. You can then use the toolkit to create connections to the external
memory interface.
External Memory Interface Handbook Volume 3: Reference Material
481
17 External Memory Interface Debug Toolkit
17.2.1 General Workflow
To use the EMIF Toolkit, you must link your compiled project to a device, and create a
communication channel to the connection that you want to examine.
17.2.2 Linking the Project to a Device
1.
To launch the toolkit, select External Memory Interface Toolkit from the Tools
menu in the Quartus Prime software.
2.
After you have launched the toolkit, open your project and click the Initialize
connections task in the Tasks window, to initialize a list of all known
connections.
3.
To link your project to a specific device on specific hardware, perform the following
steps:
a.
Click the Link Project to Device task in the Tasks window.
b.
Select the desired hardware from the Hardware dropdown menu in the Link
Project to Device dialog box.
c.
Select the desired device on the hardware from the Device dropdown menu in
the Link Project to Device dialog box.
d.
Select the correct Link file type, depending on the version of software in
which your design was compiled:
•
If your design was compiled in the Quartus II software version 12.0 or
earlier, select JDI as the Link file type, verify that the .jdi file is correct
for your .sof file, and click Ok.
•
If your design was compiled in the Quartus II software version 12.1 or
later, or the Quartus Prime software, select SOF as the Link file type,
verify that the .sof file is correct for your programmed device, and click
Ok.
Figure 208. Link Project to Device Dialog Box
When you link your project to the device, the toolkit verifies all connections on the
device against the information in the JDI or SOF file, as appropriate. If the toolkit
detects any mismatch between the JDI file and the device connections, an error
message is displayed.
For designs compiled using the Quartus II software version 12.1 or later, or the
Quartus Prime software, the SOF file contains a design hash to ensure the SOF file
used to program the device matches the SOF file specified for linking to a project. If
the hash does not match, an error message appears.
External Memory Interface Handbook Volume 3: Reference Material
482
17 External Memory Interface Debug Toolkit
If the toolkit successfully verifies all connections, it then attempts to determine the
connection type for each connection. Connections of a known type are listed in the
Linked Connections report, and are available for the toolkit to use.
17.2.3 Establishing Communication to Connections
After you have completed linking the project, you can establish communication to the
connections.
1.
2.
In the Tasks window,
•
Click Establish Memory Interface Connection to create a connection to the
external memory interface.
•
Click Establish Efficiency Monitor Connection to create a connection to the
efficiency monitor.
•
Click Establish Traffic Generator Connection to create a connection to the
Traffic Generator 2.0 (if enabled, Arria 10 only).
To create a communication channel to a connection, select the desired connection
from the displayed pulldown menu of connections, and click Ok. The toolkit
establishes a communication channel to the connection, creates a report folder for
the connection, and creates a folder of tasks for the connection.
Note: By default, the connection and the reports and tasks folders are named
according to the hierarchy path of the connection. If you want, you can
specify a different name for the connection and its folders.
3. You can run any of the tasks in the folder for the connection; any resulting reports
appear in the reports folder for the connection.
17.2.4 Selecting an Active Interface
If you have more than one external memory interface in an Arria 10 I/O column, you
can select one instance as the active interface for debugging.
1. To select one of multiple EMIF instances in an Arria 10 I/O column, use the Set
Active Interface dialog box.
2.
If you want to generate reports for the new active interface, you must first
recalibrate the interface.
External Memory Interface Handbook Volume 3: Reference Material
483
17 External Memory Interface Debug Toolkit
17.2.5 Reports
The toolkit can generate a variety of reports, including summary, calibration, and
margining reports for external memory interface connections. To generate a supported
type of report for a connection, you run the associated task in the tasks folder for that
connection.
Summary Report
The Summary Report provides an overview of the memory interface; it consists of the
following tables:
•
Summary table. Provides a high-level summary of calibration results. This table
lists details about the connection, IP version, IP protocol, and basic calibration
results, including calibration failures. This table also lists the estimated average
read and write data valid windows, and the calibrated read and write latencies.
•
Interface Details table. Provides details about the parameterization of the memory
IP. This table allows you to verify that the parameters in use match the actual
memory device in use.
•
Groups Masked from Calibration table. Lists any groups that were masked from
calibration when calibration occurred. Masked groups are ignored during
calibration.
Note: This table applies only to UniPHY-based interfaces; it is not applicable to
Arria 10 EMIF.
•
Ranks Masked from Calibration tables (DDR2 and DDR3 only). Lists any ranks that
were masked from calibration when calibration occurred. Masked ranks are
ignored during calibration.
Calibration Report (UniPHY)
The Calibration Report provides detailed information about the margins observed
before and after calibration, and the settings applied to the memory interface during
calibration; it consists of the following tables:
•
Per DQS Group Calibration table: Lists calibration results for each group. If a
group fails calibration, this table also lists the reason for the failure.
Note: If a group fails calibration, the calibration routine skips all remaining groups.
You can deactivate this behaviour by running the Enable Calibration for
All Groups On Failure command in the toolkit.
•
DQ Pin Margins Observed Before Calibration table: Lists the DQ pin margins
observed before calibration occurs. You can refer to this table to see the per-bit
skews resulting from the specific silicon and board that you are using.
•
DQS Group Margins Observed During Calibration table: Lists the DQS group
margins observed during calibration.
•
DQ Pin Settings After Calibration and DQS Group Settings After Calibration table:
Lists the settings made to all dynamically controllable parts of the memory
interface as a result of calibration. You can refer to this table to see the
modifications made by the calibration algorithm.
External Memory Interface Handbook Volume 3: Reference Material
484
17 External Memory Interface Debug Toolkit
Calibration Report (Arria 10 EMIF)
The Calibration Report provides detailed information about the margins observed
during calibration, and the settings applied to the memory interface during calibration;
it consists of the following tables:
•
Calibration Status Per Group table: Lists the pass/fail status per group.
•
DQ Pin Margins Observed During Calibration table: Lists the DQ read/write
margins and calibrated delay settings. These are the expected margins after
calibration, based on calibration data patterns. This table also contains DM/DBI
margins, if applicable.
•
DQS Pin Margins Observed During Calibration table: Lists the DQS margins
observed during calibration.
•
FIFO Settings table: Lists the VFIFO and LFIFO settings made during calibration.
•
Latency Observed During Calibration table: Lists the calibrated read/write latency.
•
Address/Command Margins Observed During Calibration table: Lists the margins
on calibrated A/C pins, for protocols that support Address/Command calibration.
Margin Report
The Margin Report lists the post-calibration margins for each DQ and data mask pin,
keeping all other pin settings constant; it consists of the following tables:
Note:
•
DQ Pin Post Calibration Margins table. Lists the margin data in tabular format.
•
Read Data Valid Windows report. Shows read data valid windows in graphical
format.
•
Write Data Valid Windows report. Shows write data valid windows in graphical
format.
The Margin Report applies only to UniPHY-based interfaces; it is not applicable to Arria
10 EMIF. For Arria 10 EMIF, the Calibration Report provides equivalent information.
17.3 Operational Considerations
Some features and considerations are of interest in particular situations.
Specifying a Particular JDI File
Correct operation of the EMIF Toolkit depends on the correct JDI being used when
linking the project to the device. The JDI file is produced by the Quartus Prime
Assembler, and contains a list of all system level debug nodes and their heirarchy path
names. If the default .jdi file name is incorrect for your project, you must specify the
correct .jdi file. The .jdi file is supplied during the link-project-to-device step, where
the revision_name.jdi file in the project directory is used by default. To supply an
alternative .jdi file, click on the ellipse then select the correct .jdi file.
External Memory Interface Handbook Volume 3: Reference Material
485
17 External Memory Interface Debug Toolkit
PLL Status
When connecting to DDR-based external memory interface connections, the PLL status
appears in the Establish Connection dialog box when the IP is generated to use the
CSR controller port, allowing you to immediately see whether the PLL status is locked.
If the PLL is not locked, no communication can occur until the PLL becomes locked and
the memory interface reset is deasserted.
When you are linking your project to a device, an error message will occur if the
toolkit detects that a JTAG Avalon-MM master has no clock running. You can run the
Reindex Connections task to have the toolkit rescan for connections and update the
status and type of found connections in the Linked Connections report.
Margining Reports
The EMIF Toolkit can display margining information showing the post-calibration datavalid windows for reads and writes. Margining information is determined by
individually modifying the input and output delay chains for each data and strobe/
clock pin to determine the working region. The toolkit can display margining data in
both tabular and hierarchial formats.
Group Masks
To aid in debugging your external memory interface, the EMIF Toolkit allows you to
mask individual groups and ranks from calibration. Masked groups and ranks are
skipped during the calibration process, meaning that only unmasked groups and ranks
are included in calibration. Subsequent mask operations overwrite any previous
masks.
Using with Arria 10 Devices in a PHY-Only Configuration
If you want to use the Debug Toolkit with an Arria 10 EMIF IP in a PHY-only
configuration, you must connect a debug clock and a reset signal to the Arria 10
external memory interface debug component. Intel recommends using the AFI clock
and AFI reset for this purpose.
Note:
For information about calibration stages in UniPHY-based interfaces, refer to UniPHY
Calibration Stages in the Functional Description - UniPHY chapter. For information
about calibration stages in Arria 10 EMIF IP, refer to Arria 10 EMIF Calibration, in the
Functional Description - Arria 10 EMIF chapter.
Related Links
•
Calibration Stages on page 59
The calibration process begins when the PHY reset signal deasserts and the PLL
and DLL lock.
•
Functional Description—UniPHY on page 13
•
Functional Description—Intel Arria 10 EMIF IP on page 144
Intel Arria 10 devices can interface with external memory devices clocking at
frequencies of up to 1.3 GHz. The external memory interface IP component for
Arria 10 devices provides a single parameter editor for creating external
memory interfaces, regardless of memory protocol.
External Memory Interface Handbook Volume 3: Reference Material
486
17 External Memory Interface Debug Toolkit
17.4 Troubleshooting
In the event of calibration failure, refer to the following figure to assist in
troubleshooting your design. Calibration results and failing stages are available
through the external memory interface toolkit.
Figure 209. Debugging Tips
Calibration Failure
Failing stage of group
Stage 2
Write Leveling
Stage 1
Read Calibration (VFIFO)
Guaranteed Read
Failure
No working DQSen
phase found
Address/Command
Skew Issue
Check
Address/Command
Clock Phase
Check
Address/Command
Board Skew and
Delays
Ensure board parameters
are correct and include
complete path to the
memory device
VFIFO Center
(Per-bit read deskew failure)
Add or subtract delay
to DQS to center DQS
within DQ window
Ensure board parameters
are correct and include
complete path to the
memory device
DQ/DQS
Centering Issue
Check DQS enable
calibration margin
Ensure DQ and DQS
are matched on the
board
No first working
write leveling
phase found
No last working write
leveling phase found
Verify the write leveling
window to ensure all
groups are within the
maximum write
leveling range
Increase the D6
delay on CLK
Stage 3
Read Calibration (LFIFO)
Write leveling copy
failure
Verify the write leveling
window to ensure all
groups are within the
maximum write
leveling range
Increase the D6 delay
on DQ and DQS for the
groups that cannot
write level
LFIFO Tuning
Failure is unexpected here.
If a failure occurs at this
point, it is probably
Stage 2 (Write Leveling)
that is failing.
Contact
Altera
Stage 4 Write
Calibration
Write Center
(Per-bit write deskew) failure
Mask the failing DQ
group and rerun
calibration
Check margins to see
if failure is all bits in
group or only some bits
Verify board traces
and solder joints
Check DQ-to-DQS
phase
Check DQ-to-DQS
phase; it should
be edge aligned.
17.5 Debug Report for Arria V and Cyclone V SoC Devices
The External Memory Interface Debug Toolkit and EMIF On-Chip Debug Port do not
work with Arria V and Cyclone V SoC devices. Debugging information for Arria V and
Cyclone V SoC devices is available by enabling a debug output report, which contains
similar information.
17.5.1 Enabling the Debug Report for Arria V and Cyclone V SoC Devices
To enable a debug report for Arria V or Cyclone V SoC devices, perform the following
steps:
1. Open the <design_name>/hps_isw_handoff/sequencer_defines.h file in a
text editor.
2.
In the sequencer_defines.h file, locate the following line: #define
RUNTIME_CAL_REPORT 0
3.
Change #define RUNTIME_CAL_REPORT 0 to #define RUNTIME_CAL_REPORT
1, and save the file.
4. Generate the board support package (BSP) with semihosting enabled, or with
UART output.
The system will now generate the debugging report as part of the calibration process.
17.5.2 Determining the Failing Calibration Stage for a Cyclone V or Arria V
HPS SDRAM Controller
To determine the failing calibration stage, you must turn on the debug output report
by setting the RUNTIME_CAL_REPORT option to 1 in the sequencer_defines.h file,
located in the hps_isw_handoff directory.
External Memory Interface Handbook Volume 3: Reference Material
487
17 External Memory Interface Debug Toolkit
If calibration fails, the following statements are printed in the debug output report:
SEQ.C:
SEQ.C:
SEQ.C:
SEQ.C:
Calibration Failed
Error Stage : <Num>
Error Substage: <Num>
Error Group : <Num>
To determine the stage and sub-stage, open the sequencer.h file in the
hps_isw_handoff directory and look for the calibration defines:
/* calibration stages */
#define CAL_STAGE_NIL 0
#define CAL_STAGE_VFIFO 1
#define CAL_STAGE_WLEVEL 2
#define CAL_STAGE_LFIFO 3
#define CAL_STAGE_WRITES 4
#define CAL_STAGE_FULLTEST 5
#define CAL_STAGE_REFRESH 6
#define CAL_STAGE_CAL_SKIPPED 7
#define CAL_STAGE_CAL_ABORTED 8
#define CAL_STAGE_VFIFO_AFTER_WRITES 9
/* calibration substages */
#define CAL_SUBSTAGE_NIL 0
#define CAL_SUBSTAGE_GUARANTEED_READ 1
#define CAL_SUBSTAGE_DQS_EN_PHASE 2
#define CAL_SUBSTAGE_VFIFO_CENTER 3
#define CAL_SUBSTAGE_WORKING_DELAY 1
#define CAL_SUBSTAGE_LAST_WORKING_DELAY 2
#define CAL_SUBSTAGE_WLEVEL_COPY 3
#define CAL_SUBSTAGE_WRITES_CENTER 1
#define CAL_SUBSTAGE_READ_LATENCY 1
#define CAL_SUBSTAGE_REFRESH 1
For details about the stages of calibration, refer to Calibration Stages in Functional
Description - UniPHY.
Related Links
Calibration Stages on page 59
The calibration process begins when the PHY reset signal deasserts and the PLL
and DLL lock.
17.6 On-Chip Debug Port for UniPHY-based EMIF IP
The EMIF On-Chip Debug Port allows user logic to access the same calibration data
used by the EMIF Toolkit, and allows user logic to send commands to the sequencer.
You can use the EMIF On-Chip Debug Port to access calibration data for your design
and to send commands to the sequencer just as the EMIF Toolkit would. The following
information is available:
•
Pass/fail status for each DQS group
•
Read and write data valid windows for each group
In addition, user logic can request the following commands from the sequencer:
•
Destructive recalibration of all groups
•
Masking of groups and ranks
•
Generation of per-DQ pin margining data as part of calibration
External Memory Interface Handbook Volume 3: Reference Material
488
17 External Memory Interface Debug Toolkit
The user logic communicates through an Avalon-MM slave interface as shown below.
Figure 210. User Logic Access
User logic
Avalon
Slave
Altera
Memory Interface
17.6.1 Access Protocol
The On-Chip Debug Port provides access to calibration data through an Avalon-MM
slave interface. To send a command to the sequencer, user logic sends a command
code to the command space in sequencer memory. The sequencer polls the command
space for new commands after each group completes calibration, and continuously
after overall calibration has completed.
The communication protocol to send commands from user logic to the sequencer uses
a multistep handshake with a data structure as shown below, and an algorithm as
shown in the figure which follows.
typedef struct_debug_data_struct {
...
// Command interaction
alt_u32 requested_command;
alt_u32 command_status;
alt_u32 command_parameters[COMMAND_PARAM_WORDS];...
}
To send a command to the sequencer, user logic must first poll the command_status
word for a value of TCLDBG_TX_STATUS_CMD_READY, which indicates that the
sequencer is ready to accept commands. When the sequencer is ready to accept
commands, user logic must write the command parameters into
command_parameters, and then write the command code into
requested_command.
The sequencer detects the command code and replaces command_status with
TCLDBG_TX_STATUS_CMD_EXE, to indicate that it is processing the command. When
the sequencer has finished running the command, it sets command_status to
TCLDBG_TX_STATUS_RESPONSE_READY to indicate that the result of the command is
available to be read. (If the sequencer rejects the requested command as illegal, it
sets command_status to TCLDBG_TX_STATUS_ILLEGAL_CMD.)
User logic acknowledges completion of the command by writing
TCLDBG_CMD_RESPONSE_ACK to requested_command. The sequencer responds by
setting command_status back to STATUS_CMD_READY. (If an illegal command is
received, it must be cleared using CMD_RESPONSE_ACK.)
External Memory Interface Handbook Volume 3: Reference Material
489
17 External Memory Interface Debug Toolkit
Figure 211. Debugging Algorithm Flowchart
Read
Command_status
command_status ==
CMD_READY ?
No
Yes
Write command
payload
Write command code
Read command_status
command_status ==
RESPONSE_READY ?
No
Yes
Write
RESPONSE_ACK code
End
17.6.2 Command Codes Reference
The following table lists the supported command codes for the On-Chip Debug Port.
Table 136.
Supported Command Codes
Command
Parameters
Description
TCLDBG_RUN_MEM_CALIBRATE
None
Runs the calibration routine.
TCLDBG_MARK_ALL_DQS_GROUPS_AS_VALID
None
Marks all groups as valid for calibration.
TCLDBG_MARK_GROUP_AS_SKIP
Group to skip
Mark the specified group to be skipped by
calibration.
continued...
External Memory Interface Handbook Volume 3: Reference Material
490
17 External Memory Interface Debug Toolkit
Command
Parameters
Description
TCLDBG_MARK_ALL_RANKS_AS_VALID
None
Mark all ranks as valid for calibration
TCLDBG_MARK_RANK_AS_SKIP
Rank to skip
Mark the specified rank to be skipped by
calibration.
TCLDBG_ENABLE_MARGIN_REPORT
None
Enables generation of the margin report.
17.6.3 Header Files
The external memory interface IP generates header files which identify the debug data
structures and memory locations used with the EMIF On-Chip Debug Port. You should
refer to these header files for information required for use with your core user logic. It
is highly recommended to use a software component (such as a Nios II processor) to
access the calibration debug data.
The header files are unique to your IP parameterization and version, therefore you
must ensure that you are referring to the correct version of header for your design.
The names of the header files are: core_debug.h and core_debug_defines.h. The
header files reside in <design_name>/<design_name>_s0_software.
17.6.4 Generating IP With the Debug Port
The following steps summarize the procedure for implementing your IP with the EMIF
On-Chip Debug Port enabled.
1.
Start the Quartus Prime software and generate a new external memory interface.
For QDR II and RLDRAM II protocols, ensure that sequencer optimization is set to
Performance (for Nios II-based sequencer).
2.
On the Diagnostics tab of the parameter editor, turn on Enable EMIF On-Chip
Debug Port.
3.
Ensure that the EMIF On-Chip Debug Port interface type is set to Avalon-MM
Slave.
4.
Click Finish to generate your IP.
5.
Find the Avalon interface in the top-level generated file. Connect this interface to
your debug component.
input
input
output
input
input
output
input
output
wire
wire
wire
wire
wire
wire
wire
wire
[19:0] seq_debug_addr,
seq_debug_read_req,
[31:0] seq_debug_rdata,
seq_debug_write_req,
[31:0] seq_debug_wdata,
seq_debug_waitrequest,
[3:0] seq_debug_be,
seq_debug_rdata_valid
//
//
//
//
//
//
//
//
seq_debug.address
.read
.readdata
.write
.writedata
.waitrequest
.byteenable
.readdatavalid
External Memory Interface Handbook Volume 3: Reference Material
491
17 External Memory Interface Debug Toolkit
If you are using UniPHY-based IP with the hard memory controller, also connect
the seq_debug_clk and seq_debug_reset_in signals to clock and
asynchronous reset signals that control your debug logic.
6.
Find the core_debug.h and core_debug_defines.h header files in
<design_name>/<design_name>_s0_software and include these files in your
debug component code.
7. Write your debug component using the supported command codes, to read and
write to the Avalon-MM interface.
The debug data structure resides at the memory address SEQ_CORE_DEBUG_BASE,
which is defined in the core_debug_defines.h header file.
17.6.5 Example C Code for Accessing Debug Data
A typical use of the EMIF On-Chip Debug Port might be to recalibrate the external
memory interface, and then access the reports directly using the
summary_report_ptr, cal_report_ptr, and margin_report_ptr pointers,
which are part of the debug data structure.
The following code sample illustrates:
/*
* DDR3 UniPHY sequencer core access example
*/
#include
#include
#include
#include
<stdio.h>
<unistd.h>
<io.h>
"core_debug_defines.h"
int send_command(volatile debug_data_t* debug_data_ptr, int command, int args[],
int num_args)
{
volatile int i, response;
// Wait until command_status is ready
do {
response = IORD_32DIRECT(&(debug_data_ptr->command_status), 0);
} while(response != TCLDBG_TX_STATUS_CMD_READY);
// Load arguments
if(num_args > COMMAND_PARAM_WORDS)
{
// Too many arguments
return 0;
}
for(i = 0; i < num_args; i++)
{
IOWR_32DIRECT(&(debug_data_ptr->command_parameters[i]), 0, args[i]);
}
// Send command code
IOWR_32DIRECT(&(debug_data_ptr->requested_command), 0, command);
// Wait for acknowledgment
do {
response = IORD_32DIRECT(&(debug_data_ptr->command_status), 0);
} while(response != TCLDBG_TX_STATUS_RESPOSE_READY && response !=
TCLDBG_TX_STATUS_ILLEGAL_CMD);
// Acknowledge response
IOWR_32DIRECT(&(debug_data_ptr->requested_command), 0, TCLDBG_CMD_RESPONSE_ACK);
// Return 1 on success, 0 on illegal command
return (response != TCLDBG_TX_STATUS_ILLEGAL_CMD);
}
int main()
External Memory Interface Handbook Volume 3: Reference Material
492
17 External Memory Interface Debug Toolkit
{
volatile debug_data_t* my_debug_data_ptr;
volatile debug_summary_report_t* my_summary_report_ptr;
volatile debug_cal_report_t* my_cal_report_ptr;
volatile debug_margin_report_t* my_margin_report_ptr;
volatile debug_cal_observed_dq_margins_t* cal_observed_dq_margins_ptr;
int i, j, size;
int args[COMMAND_PARAM_WORDS];
// Initialize pointers to the debug reports
my_debug_data_ptr = (debug_data_t*)SEQ_CORE_DEBUG_BASE;
my_summary_report_ptr = (debug_summary_report_t*)
(IORD_32DIRECT(&(my_debug_data_ptr->summary_report_ptr), 0));
my_cal_report_ptr = (debug_cal_report_t*)(IORD_32DIRECT(&(my_debug_data_ptr>cal_report_ptr), 0));
my_margin_report_ptr = (debug_margin_report_t*)
(IORD_32DIRECT(&(my_debug_data_ptr->margin_report_ptr), 0));
// Activate all groups and ranks
send_command(my_debug_data_ptr, TCLDBG_MARK_ALL_DQS_GROUPS_AS_VALID, 0, 0);
send_command(my_debug_data_ptr, TCLDBG_MARK_ALL_RANKS_AS_VALID, 0, 0);
send_command(my_debug_data_ptr, TCLDBG_ENABLE_MARGIN_REPORT, 0, 0);
// Mask group 4
args[0] = 4;
send_command(my_debug_data_ptr, TCLDBG_MARK_GROUP_AS_SKIP, args, 1);
send_command(my_debug_data_ptr, TCLDBG_RUN_MEM_CALIBRATE, 0, 0);
// SUMMARY
printf("SUMMARY REPORT\n");
printf("mem_address_width: %u\n", IORD_32DIRECT(&(my_summary_report_ptr>mem_address_width), 0));
printf("mem_bank_width: %u\n", IORD_32DIRECT(&(my_summary_report_ptr>mem_bank_width), 0));
// etc...
// CAL REPORT
printf("CALIBRATION REPORT\n");
// DQ read margins
for(i = 0; i < RW_MGR_MEM_DATA_WIDTH; i++)
{
cal_observed_dq_margins_ptr = &(my_cal_report_ptr->cal_dq_in_margins[i]);
printf("0x%x DQ %d Read Margin (taps): -%d : %d\n", (unsigned
int)cal_observed_dq_margins_ptr, i,
IORD_32DIRECT(&(cal_observed_dq_margins_ptr->left_edge), 0),
IORD_32DIRECT(&(cal_observed_dq_margins_ptr->right_edge), 0));
}
// etc...
return 0;
}
17.7 On-Chip Debug Port for Arria 10 EMIF IP
The EMIF On-Chip Debug Port allows user logic to access the same calibration data
used by the EMIF Toolkit, and allows user logic to send commands to the sequencer.
You can use the EMIF On-Chip Debug Port to access calibration data for your design
and to send commands to the sequencer just as the EMIF Toolkit would. The following
information is available:
•
Pass/fail status for each DQS group
•
Read and write data valid windows for each group
External Memory Interface Handbook Volume 3: Reference Material
493
17 External Memory Interface Debug Toolkit
In addition, user logic can request the following commands from the sequencer:
•
Destructive recalibration of all groups
•
Masking of groups and ranks
•
Generation of per-DQ pin margining data as part of calibration
The user logic communicates through an Avalon-MM slave interface as shown below.
Figure 212. User Logic Access
User logic
Avalon
Slave
Altera
Memory Interface
17.7.1 Access Protocol
The On-Chip Debug Port provides access to calibration data through an Avalon-MM
slave interface. To send a command to the sequencer, user logic sends a command
code to the command space in sequencer memory. The sequencer polls the command
space for new commands after each group completes calibration, and continuously
after overall calibration has completed.
The communication protocol to send commands from user logic to the sequencer uses
a multistep handshake with a data structure as shown below, and an algorithm as
shown in the figure which follows.
typedef struct_debug_data_struct {
...
// Command interaction
alt_u32 requested_command;
alt_u32 command_status;
alt_u32 command_parameters[COMMAND_PARAM_WORDS];...
}
To send a command to the sequencer, user logic must first poll the command_status
word for a value of TCLDBG_TX_STATUS_CMD_READY, which indicates that the
sequencer is ready to accept commands. When the sequencer is ready to accept
commands, user logic must write the command parameters into
command_parameters, and then write the command code into
requested_command.
The sequencer detects the command code and replaces command_status with
TCLDBG_TX_STATUS_CMD_EXE, to indicate that it is processing the command. When
the sequencer has finished running the command, it sets command_status to
TCLDBG_TX_STATUS_RESPONSE_READY to indicate that the result of the command is
available to be read. (If the sequencer rejects the requested command as illegal, it
sets command_status to TCLDBG_TX_STATUS_ILLEGAL_CMD.)
External Memory Interface Handbook Volume 3: Reference Material
494
17 External Memory Interface Debug Toolkit
User logic acknowledges completion of the command by writing
TCLDBG_CMD_RESPONSE_ACK to requested_command. The sequencer responds by
setting command_status back to STATUS_CMD_READY. (If an illegal command is
received, it must be cleared using CMD_RESPONSE_ACK.)
Figure 213. Debugging Algorithm Flowchart
Read
Command_status
command_status ==
CMD_READY ?
No
Yes
Write command
payload
Write command code
Read command_status
command_status ==
RESPONSE_READY ?
No
Yes
Write
RESPONSE_ACK code
End
17.7.2 EMIF On-Chip Debug Port
In Arria 10 and later families, access to on-chip debug is provided through software
running on a Nios processor connected to the external memory interface.
If you enable the Use Soft Nios Processor for On-Chip Debug option, the
system instantiates a soft Nios processor, and software files are provided as part of
the EMIF IP.
External Memory Interface Handbook Volume 3: Reference Material
495
17 External Memory Interface Debug Toolkit
Instructions on how to use the software are available in the following file: :
<variation_name>/altera_emif_arch_nf_<version number>/<synth|
sim>/<variation_name>_altera_emif_arch_nf_<version
number>_<unique ID>_readme.txt.
17.7.3 On-Die Termination Calibration
The Calibrate Termination feature lets you determine the optimal On-Die
Termination and Output Drive Strength settings for your memory interface, for
Arria 10 and later families.
The Calibrate Termination function runs calibration with all available termination
settings and selects the optimal settings based on the calibration margins.
The Calibrate Termination feature is available for DDR3, DDR4, and RLDRAM 3
protocols, on Arria 10 devices.
17.7.4 Eye Diagram
The Generate Eye Diagram feature allows you to create read and write eye
diagrams for each pin in your memory interface, for Arria 10 and later families.
The Generate Eye Diagram feature uses calibration data patterns to determine
margins at each Vref setting on both the FPGA pins and the memory device pins. A full
calibration is done for each Vref setting. Other settings, such as DQ delay chains, will
change for each calibration. At the end of a Generate Eye Diagram command, a
default calibration is run to restore original behavior
The Generate Eye Diagram feature is available for DDR4 and QDR-IV protocols, on
Arria 10 devices.
17.8 Driver Margining for Arria 10 EMIF IP
The Driver Margining feature lets you measure margins on your memory interface
using a driver with arbitrary traffic patterns.
Margins measured with this feature may differ from margins measured during
calibration, because of different traffic patterns. Driver margining is not available if
ECC is enabled.
To use driver margining, ensure that the following signals on the driver are connected
to In-System Sources/Probes:
•
Reset_n: An active low reset signal
•
Pass: A signal which indicates that the driver test has completed successfully. No
further memory transactions must be sent after this signal is asserted.
•
Fail: A signal which indicates that the driver test has failed. No further memory
transactions must be sent after this signal is asserted.
•
PNF (Pass Not Fail): An array of signals that indicate the pass/fail status of
individual bits of a data burst. The PNF should be arranged such that each bit
index corresponds to (Bit of burst * DQ width) + (DQ pin). A 1
indicates pass, 0 indicates fail. If the PNF width exceeds the capacity of one InSystem Probe, specify them in PNF[1] and PNF[2]; otherwise, leave them blank.
External Memory Interface Handbook Volume 3: Reference Material
496
17 External Memory Interface Debug Toolkit
If you are using the example design for EMIF, the In-System Sources/Probes can be
enabled by adding the following line to your .qsf file:
set_global_assignment -name VERILOG_MACRO
"ALTERA_EMIF_ENABLE_ISSP=1"
17.8.1 Determining Margin
The Driver Margining feature lets you measure margins on your Arria 10 EMIF IP
interface using a driver with arbitrary traffic patterns.
The Driver Margining feature is available only for DDR3 and DDR4 interfaces on Arria
10 devices, when ECC is not enabled.
1. Establish a connection to the desired interface and ensure that it has calibrated
successfully.
2.
Select Driver Margining from the Commands folder under the target interface
connection.
3. Select the appropriate In-System Sources/Probes using the drop-down menus.
4.
If required, set additional options in the Advanced Options section:
•
Specify Traffic Generator 2.0 to allow margining on a per-rank basis.
Otherwise, margining is performed on all ranks together.
•
Step size specifies the granularity of the driver margining process. Larger
step sizes allow faster margining but reduced accuracy. It is recommended to
omit this setting.
•
Adjust delays after margining causes delay settings to be adjusted to the
center of the window based on driver margining results.
•
The Margin Read, Write, Write DM, and DBI checkboxes allow you to
control which settings are tested during driver margining. You can uncheck
boxes to allow driver margining to complete more quickly.
5. Click OK to run the tests.
The toolkit measures margins for DQ read/write and DM. The process may take
several minutes, depending on the margin size and the duration of the driver
tests. The test results are available in the Margin Report.
17.9 Read Setting and Apply Setting Commands for Arria 10 EMIF
IP
The Read Setting command allows you to read calibration settings directly from the
EMIF PHY. The Apply Setting command allows you to write calibration settings, to
override existing settings for testing purposes.
17.9.1 Reading or Applying Calibration Settings
The Read Setting and Apply Setting commands let you read and write calibration
settings directly.
The Read Setting and Apply Setting commands are available only for DDR3 and
DDR4 interfaces on Arria 10 devices.
External Memory Interface Handbook Volume 3: Reference Material
497
17 External Memory Interface Debug Toolkit
1.
Establish a connection to the desired interface.
2.
Select Read Setting or Apply Setting from the Settings folder under the target
interface connection.
3. Select the desired setting type.
4. Select the desired rank shadow register to modify. (Leave this field blank if it is
not applicable).
5.
Select the index of the pin or group to modify.
6.
(For the Apply Setting command) Enter the new value to apply to the desired
location.
7.
Click OK .
The setting is read (or applied) using a read/write at the address indicated in the
Tcl command window. You can perform similar transactions using the On-Chip
Debug Port.
17.10 Traffic Generator 2.0
The Traffic Generator 2.0 lets you emulate traffic to the external memory, and helps
you test, debug, and understand the performance of your external memory interface
on hardware in a standalone fashion, without having to incorporate your entire design.
The Traffic Generator 2.0 lets you customize data patterns being written to the
memory, address locations accessed in the memory, and the order of write and read
transactions. You can use the traffic generator code with any FPGA architecture and
memory protocol.
17.10.1 Configuring the Traffic Generator 2.0
The traffic generator replaces user logic to generate traffic to the external memory.
You must incorporate the traffic generator design into the EMIF IP design during IP
generation.
When you generate the example design in the parameter editor, the traffic generator
module and EMIF IP are generated together. If you have an example design with the
Traffic Generator 2.0 enabled, you can configure the traffic pattern using the EMIF
Debug Toolkit.
External Memory Interface Handbook Volume 3: Reference Material
498
17 External Memory Interface Debug Toolkit
Figure 214. Traffic Generator 2.0 Generated with EMIF IP in Example Design Mode
Generating the External Memory Interface
1.
Select the FPGA and Memory parameters.
2.
On the Diagnostics tab, configure the following parameters:
3.
a.
Select Use Configurable Avalon Traffic Generator 2.0.
b.
Configure the desired traffic pattern, by specifying traffic patterns to be
bypassed. The traffic pattern not bypassed is issued to the memory
immediately after completion of calibration. You can choose to bypass any of
the following traffic patterns:
•
Bypass the default traffic pattern Specifies not to use the default
traffic patterns from the traffic generator. The default patterns include
single read/write, byte-enabled read/write, and block read/write.
•
Bypass the user-configured traffic stage. Specifies to skip the stage
that uses the user-defined test bench file to configure the traffic generator
in simulation.
•
Bypass the traffic generator repeated-writes/repeated-reads test
pattern. Bypasses the traffic generator's repeat test stage, which causes
every write and read to be repeated several times.
•
Bypass the traffic generator stress pattern. Bypasses a test stage
intended to stress-test signal integrity and memory interface calibration.
•
Export Traffic Generator 2.0 configuration interface. Instantiates a
port for traffic generator configuration. Use this port if the traffic generator
is to be configured by user logic.
Click Generate Example Design to generate the EMIF IP, including the Traffic
Generator 2.0 design, with the traffic pattern that you have configured.
Note: If you click the Generate HDL option instead, the Traffic Generator 2.0
design is not included in the generated IP.
External Memory Interface Handbook Volume 3: Reference Material
499
17 External Memory Interface Debug Toolkit
Figure 215. Enabling the Traffic Generator 2.0 in the Parameter Editor
17.10.2 Running the Traffic Generator 2.0
You can use the EMIF Debug Toolkit to configure the traffic generator infrastructure to
send custom traffic patterns to the memory.
1. Launch the EMIF Debug Toolkit by selecting Tools ➤ System Debugging Tools
➤ External Memory Interface Toolkit.
2.
3.
After you launch the toolkit, you must establish the following connections, before
running the custom traffic generator:
•
Initialize Connections
•
Link Project to Device
•
Connections
—
Create Memory Interface Connection
—
Create Traffic Generator Connection
Launch the Traffic Generator by selecting Traffic Generator ➤ Settings ➤ Run
Custom Traffic Pattern.
17.10.3 Understanding the Custom Traffic Generator User Interface
The Custom Traffic Generator interface lets you configure data patterns, the bytes to
be enabled, the addressing mode, and the order in which traffic is organized.
External Memory Interface Handbook Volume 3: Reference Material
500
17 External Memory Interface Debug Toolkit
The interface has three tabs:
•
Data tab
•
Address tab
•
Loops tab
Data Tab
The Data tab is divided into Data Pins and Data Mask Pins sections.
Figure 216. Data Tab
The Data Pins section helps with customizing the patterns selected for the data pins.
You can choose between two options for Data Mode:
•
PRBS The default write data to all the data pins.
•
Fixed Pattern Lets you specify a pattern to be written to the memory.
Select the All Pins option when you want to write the same data pattern to all the
data pins. If data must be individually assigned to the data pins, you must enter the
data value for each individual pin. The width of the data entered is based on the
AVL_TO_DQ_WIDTH_RATIO, which is based on the ratio of the memory clock to the
user clock.
External Memory Interface Handbook Volume 3: Reference Material
501
17 External Memory Interface Debug Toolkit
All data bytes are enabled by default; the Data Mask Pins section lets you disable
any of the bytes if you want to. To disable data bytes individually, select Test data
mask.
You can choose between two options for Data Mode:
•
PRBS Specifies the PRBS pattern to enable or disable data bytes. A 1 denotes a
data byte enabled, while a 0 denotes a data byte being masked or disabled.
•
Fixed Pattern Lets you enable or disable individual bytes. You can apply byte
enables to all pins or to individual bytes. A 1 denotes a data byte enabled, while a
0 denotes a data byte being masked or disabled.
Address Tab
The Address tab lets you configure sequential, random, or random sequential (where
the initial start address is random, but sequential thereafter) addressing schemes. The
Address tab is divided into Address Mode and Address Configuration sections.
Figure 217. Address Tab
The Address Mode section lets you specify the pattern of addresses generated to
access the memory. You can choose between three address modes :
•
Sequential Each address is incremented by the Sequential address increment
value that you specify. You also specify the Start Address from which the
increments begin.
•
Random Each address is generated randomly. (You set the number of random
addresses on the Loops tab.)
•
Random Sequential Each address is generated randomly, and then incremented
sequentially. You specify the number of sequential increments in the Number of
sequential addresses field.
External Memory Interface Handbook Volume 3: Reference Material
502
17 External Memory Interface Debug Toolkit
The Address Configuration section contains the settings with which you configure
the address mode that you chose in the Address Mode section. The following settings
are available:
•
Start address Specifies the starting address for Sequential Address Mode. The
maximum address value that can be reached is 1FF_FFFF. (The Traffic Generator
2.0 will accept higher values, but wraps back to 0 after the maximum value has
been reached.) The Start address setting applies only to Sequential Address
Mode.
•
Number of sequential addresses Specifies the number of sequential addresses
generated after the first random address generated. This setting applies only in
Random sequential mode.
•
Sequential address increment Specifies the size of increment between each
address in the Sequential address mode and Random sequential address
mode.
•
Return to start address Specifies that the address value generated return back
to the value entered in the Start Address field, after a block of transactions to
the memory has completed. This setting applies only to Sequential address
mode.
•
Address masking Masking provides additional options for exploring certain
specific address spaces in memory:
—
Disabled does not enable masking, and increments address based on the
selected Address Mode.
—
Fixed cycling allows you to restrict the addressing to a specific row or a
specific bank, which you can specify in the corresponding Mask Value field.
External Memory Interface Handbook Volume 3: Reference Material
503
17 External Memory Interface Debug Toolkit
Loops Tab
The Loops tab lets you order the transactions to the memory as desired. A unit size of
transactions to the memory is defined as a block; a block includes a set of write
transaction(s) immediately followed by a set of read transaction(s).
Figure 218. Loops Tab
External Memory Interface Handbook Volume 3: Reference Material
504
17 External Memory Interface Debug Toolkit
The Loops tab provides the following configuration options:
•
Loops Specifies the number of blocks of transactions to be sent to the memory.
This option helps to extend the range of addresses that the controller can access.
The address range is incremented as each loop is executed, unless you specify
Return to Start Address, which causes each loop to begin from the same start
address. The range of supported values for the Loops option is from 1 to 4095.
•
Writes per block Specifies the size of the block (that is, the number of
consecutive write operations that can be issued in a single block). The range of
values for this option is as follows:
•
—
When address masking is disabled, the number of writes per block supported
is 1 to 4094.
—
When address masking is enabled, the maximum number of writes issued
inside a block is 255.
Reads per block Specifies the number of consecutive read operations that can be
issued in a single block, immediately following the consecutive writes issued. The
number of reads per block should be identical to the number of writes per block,
because data mismatches can occur when the two values are not identical. The
range of values for this option is as follows:
—
When address masking is disabled, the number of reads per block supported is
1 to 4094.
—
When address masking is enabled, the maximum number of reads issued
inside a block is 255.
•
Write repeats Specifies the number of times each write command is issued in
repetition to the same address. A maximum number of 255 repeat write
transactions can be issued. The repeat writes are issued immediately after the first
write command has been issued.
•
Read repeats Specifies the number of times each read command is issued in
repetition to the same address. A maximum number of 255 repeat read
transactions can be issued. The repeat reads are issued immediately after the first
read command has been issued.
•
Avalon burst length Specifies the length of each Avalon burst. The value of this
field should be less than the Sequential address increment specified on the
Address tab. The number of write and read repeats default to 1 if the Avalon
burst length is greater than 1.
17.10.4 Applying the Traffic Generator 2.0
You can apply the Traffic Generator 2.0 to run stress tests, debug your hardware
platform for signal integrity problems, and to emulate actual memory transactions.
This topic presents some common applications where you can benefit by using the
Traffic Generator 2.0.
Testing Signal Integrity with PRBS Data Pattern
You can apply PRBS data to the data pins to help emulate an actual traffic pattern to
the memory interface. The traffic generator uses a PRBS7 data pattern as the default
traffic pattern on the data pins, and can support PRBS-15 and PRBS-31.
External Memory Interface Handbook Volume 3: Reference Material
505
17 External Memory Interface Debug Toolkit
Debugging and Monitoring an Address for Reliable Data Capture
You can send a single write followed by multiple reads to a specific address to help
debug and monitor a specific address for reliable data capture. You can do this with
the following settings:
•
Writes per block: 1
•
Reads per block: 1
•
Write repeats: 1
•
Read repeats: 1 to 255
Figure 219. Configuring the Loops Tab for a Single Write Followed by Multiple Reads
If you specify a Loops value greater than 1, every block of write and multiple read
transactions will follow the same pattern. If there is a specific address to which this
transaction must be issued, you should specify that address in the Start address field
on the Address tab, with the Sequential address mode selected.
Accessing Large Sections of Memory
The maximum number of unique addresses that can be written to in one block is
4094. Using the maximum Loops value of 4095, the address range that can be
supported in one test is equal to the number of loops multiplied by the number of
writes per block. Further address expansion can be achieved by changing the Start
address value appropriately and reissuing the tests.
To continue addressing sections of the memory beyond the address range that can be
specified in one set of toolkit configurations, you can incrementally access the next set
of addresses in the memory by changing the Start address value.
For example, in a memory where row address width is 15, bank address width is 3
and column address width is 10, the total number of address locations to be accessed
is: 2 (row address width) x (bank address width x 2 (column address width)). The
maximum number of address locations that can be accessed is limited by the width of
the internal address bus, which is 25 bits wide.
For the example described above, you must set the following values on the Address
tab:
•
Select the Sequential address mode.
•
Set the Start address to 0x00.
•
Ensure that you do not select Return to start addess.
•
Ensure that you disable address masking for rank, row, bank, and bank group.
External Memory Interface Handbook Volume 3: Reference Material
506
17 External Memory Interface Debug Toolkit
Figure 220. Address Configuration to Access the First Set of Addresses
You must also set the following values on the Loops tab:
•
Set Loops to the maximun value of 4095.
•
Set Writes per block to the maximum value of 4094.
•
Set Reads per block to the maximum value of 4094.
•
Set Write repeats to 1.
•
Set Read repeats to 1.
Figure 221. Loop Configuration to Access the First Set of Addresses
Each iteration can access a maximum of 4095 x 4094 locations (16,764,930 address
locations i.e. Address ranging from 000_0000’h to FF_D001’h). To access the next
4095 x 4094 locations, the same settings as above must be repeated, except for the
Start address value, whichmust be set to a hex value of 16,764,931 i.e. FF_D002.
The same process can be repeated to further access memory locations inside the
memory. The maximum value supported is 25’h 1FF_FFFF which is the equivalent of
33,554,432 locations inside the memory.
External Memory Interface Handbook Volume 3: Reference Material
507
17 External Memory Interface Debug Toolkit
Figure 222. Address Configuration to Access the Second Set of Addresses
17.11 The Traffic Generator 2.0 Report
The traffic generator report provides information about the configuration of the Traffic
Generator 2.0 and the result of the most recent run of traffic.
Understanding the Traffic Generator 2.0 Report
The traffic generator report contains the following information:
•
A Pass flag value of 1 indicates that the run completed with no errors.
•
A Fail flag value of 1 indicates that the run encountered one or more errors.
•
The Failure Count indicates the number of read transactions where data did not
match the expected value.
•
The First failure address indicates the address corresponding to the first data
mismatch.
•
The Version indicates the version number of the traffic generator.
•
The Number of data generators indicates the number of data pins at the
memory interface.
•
The Number of byte enable generators indicates the number of byte enable
and data mask pins at the memory interface.
•
The Rank Address, Bank address, and Bank group width values indicate the
number of bits in the Avalon address corresponding to each of those components.
•
The Data/Byte enable pattern length indicates the number of bits in the fixed
pattern used on each data/byte enable pin.
•
The PNF (pass not fail) value indicates the persistent pass/fail status for each bit
in the Avalon data. It is also presented on a per-memory-pin basis for each beat
within a memory burst.
•
Fail Expected Data is the data that was expected on the first failing transaction
(if applicable).
•
Fail Read Data is the data that was received on the first failing transaction (if
applicable).
External Memory Interface Handbook Volume 3: Reference Material
508
17 External Memory Interface Debug Toolkit
17.12 Example Tcl Script for Running the EMIF Debug Toolkit
If you want, you can run the EMIF Debug Toolkit using a Tcl script. The following
example Tcl script is applicable to all device families.
The following example Tcl script opens a file, runs the debug toolkit, and writes the
resulting calibration reports to a file.
You should adjust the variables in the script to match your design. You can then run
the script using the command quartus_sh -t example.tcl.
# Modify the following variables for your project
set project "ed_synth.qpf"
# Index of the programming cable. Can be listed using "get_hardware_names"
set hardware_index 1
# Index of the device on the specified cable. Can be listed using
"get_device_names"
set device_index 1
# SOF file containing the EMIF to debug
set sof "ed_synth.sof"
# Connection ID of the EMIF debug interface. Can be listed using
"get_connections"
set connection_id 2
# Output file
set report "toolkit.rpt"
# The following code opens a project and writes its calibration reports to a file.
project_open $project
load_package ::quartus::external_memif_toolkit
initialize_connections
set hardware_name [lindex [get_hardware_names] $hardware_index]
set device_name [lindex [get_device_names -hardware_name $hardware_name]
$device_index]
link_project_to_device -device_name $device_name -hardware_name $hardware_name sof_file $sof
establish_connection -id $connection_id
create_connection_report -id $connection_id -report_type summary
create_connection_report -id $connection_id -report_type calib
write_connection_target_report -id $connection_id -file $report
17.13 Calibration Adjustment Delay Step Sizes for Arria 10 Devices
Refer to the following tables for information on delay step sizes for calibration
adjustment.
17.13.1 Addressing
Each reconfigurable feature of the interface has an associated memory address;
however, this address is placement dependent. If Altera PHYLite for Parallel Interfaces
IP cores and the Arria 10 External Memory Interfaces IP cores share the same I/O
column, you must track the addresses of the interface lanes and the pins. Addressing
is done at the 32-bit word boundary, where avl_address[1:0] are always 00.
External Memory Interface Handbook Volume 3: Reference Material
509
17 External Memory Interface Debug Toolkit
Address Map
These points apply to the following table:
•
id[3:0] refers to the Interface ID parameter.
•
lane_addr[7:0] refers to the address of a given lane in an interface. The Fitter
sets this address value. You can query this in the Parameter Table Lookup
Operation Sequence as described in Address Lookup section of the Intel PHYLite
for Parallel Interfaces IP Core User Guide.
•
pin[4:0] refers to the physical location of the pin in a lane. You can use the
Fitter to automatically determine a pin location or you can manually set the pin
location through .qsf assignment. Refer to the Parameter Table Lookup Operation
Sequence as described in Address Lookup section of the Intel PHYLite for Parallel
Interfaces IP Core User Guide for more information.
Feature
Pin Output Phase
Avalon Address R/W
{id[3:0],
3'h4,lane_addr[7:
0],pin{4:0],8'D0}
Address CSR R
{id[3:0],
3'h4,lane_addr[7
:0],pin{4:0],
8'E8}
Control
Value
Field
Range
Phase
Value
12..0
Minimum Setting: Refer
to Table 137 on page
513
Maximum Setting: Refer
to Table 137 on page
513
Incremental Delay:
1/128th VCO clock period
Note: The pin output
phase switches
from the CSR
value to the
Avalon value after
the first Avalon
write. It is only
reset to the CSR
value on a reset
of the interface.
Reserved
31..13
Delay
Value
8..0
Reserved
11..9
Enable
12
Reserved
31..13
1
Pin PVT
Compensated
Input Delay
{id[3:0],
3'h4,lane_addr[7:
0],
4'hC,lgc_sel[1:0]
,pin_off[2:0],
4'h0}
Not supported
1
1
—
Minimum Setting: 0
Maximum Setting: 511
VCO clock periods
Incremental Delay:
1/256th VCO clock period
—
0 = Delay value is 0.
1 = Select delay value
from Avalon register
—
continued...
External Memory Interface Handbook Volume 3: Reference Material
510
17 External Memory Interface Debug Toolkit
Feature
Avalon Address R/W
•
•
Strobe PVT
compensated
input delay 2
Control
Field
Range
Not supported
Delay
Value
9..0
lgc_sel[1:0] =
2'b01
Reserved
11..10
Enable
12
Reserved
31..13
Phase
Value
12..0
Reserved
14..13
Enable
15
Reserved
31..16
1
1
Strobe enable
phase 2
{id[3:0],
3'h4,lane_addr[7:
0],
4'hC,lgc_sel[1:0]
,3'h7,4'h0}
•
Value
lgc_sel[1:0] is:
— 2'b01 for DQ
[5:0]
— 2'b10 for DQ
[11:6]
pin_off[2:0] :
— 3'h0: DQ [0],
DQ [6]
— 3’h1: DQ [1],
DQ [7]
— 3’h2: DQ [2],
DQ [8]
— 3’h3: DQ [3],
DQ [9]
— 3’h4: DQ [4],
DQ [10]
— 3’h5: DQ [5],
DQ [11]
{id[3:0],
3'h4,lane_addr[7:
0],
4'hC,lgc_sel[1:0]
,3'h6,4'h0}
•
Address CSR R
{id[3:0],
3'h4,lane_addr[7
:0],4'hC,9'h198}
lgc_sel[1:0] =
2'b01
1
1
Minimum Setting: 0
Maximum Setting: 1023
VCO clock periods
Incremental Delay:
1/256th VCO clock period
—
0 = Select delay value
from CSR register. The
CSR value is set through
the Capture Strobe
Phase Shift parameter
during IP core
instantiation.
1 = Select delay value
from Avalon register
—
Minimum Setting: Refer
to Table 137 on page
513
Maximum Setting: Refer
to Table 137 on page
513
Incremental Delay:
1/128th VCO clock period
—
0 = Select delay value
from CSR register
1 = Select delay value
from Avalon register
—
continued...
External Memory Interface Handbook Volume 3: Reference Material
511
17 External Memory Interface Debug Toolkit
Feature
Avalon Address R/W
Strobe enable
delay 2
{id[3:0],
3'h4,lane_addr[7:
0],4'hC,9'h008}
Address CSR R
{id[3:0],
3'h4,lane_addr[7
:0],4'hC,9'h1A8}
Control
Field
Range
Delay
Value
5..0
Reserved
14..6
Enable
15
Reserved
31..16
Delay
Value
6..0
Reserved
14..7
Enable
15
Reserved
31..16
VREF
Code
5..0
1
1
Read valid delay
2
{id[3:0],
3'h4,lane_addr[7:
0],4'hC,9'h00C}
{id[3:0],
3'h4,lane_addr[7
:0],4'hC,9'h1A4}
1
1
Internal VREF
Code
{id[3:0],
3'h4,lane_addr[7:
0],4'hC,9'h014}
Not supported
Value
Reserved
1
31..6
Minimum Setting: 0
external clock cycles
Maximum Setting: 63
external memory clock
cycles
Incremental Delay: 1
external memory clock
cycle
—
0 = Select delay value
from CSR register
1 = Select delay value
from Avalon register
—
Minimum Setting: 0
external clock cycles
Maximum Setting: 127
external memory clock
cycles
Incremental Delay: 1
external memory clock
cycle
—
0 = Select delay value
from CSR register
1 = Select delay value
from Avalon register
—
Refer to Calibrated VREF
Settings in the Intel
PHYLite for Parallel
Interfaces IP Core User
Guide.
9
—
1. Reserved bit ranges must be zero.
2. Modifying these values must be done on all lanes in a group.
Note:
For more information about performing various clocking and delay calculations,
depending on the interface frequency and rate, refer to
PHYLite_delay_calculations.xlsx.
17.13.2 Output and Strobe Enable Minimum and Maximum Phase Settings
When dynamically reconfiguring the interpolator phase settings, the values must be
kept within the ranges below to ensure proper operation of the circuitry.
External Memory Interface Handbook Volume 3: Reference Material
512
17 External Memory Interface Debug Toolkit
Table 137.
Output and Strobe Enable Minimum and Maximum Phase Settings
VCO
Multiplication
Factor
1
2
4
8
Core Rate
Minimum Interpolator Phase
Maximum
Interpolator Phase
Output
Bidirectional
Bidirectional with
OCT Enabled
Full
0x080
0x100
0x100
0xA80
Half
0x080
0x100
0x100
0xBC0
Quarter
0x080
0x100
0x100
0xA00
Full
0x080
0x100
0x180
0x1400
Half
0x080
0x100
0x180
0x1400
Quarter
0x080
0x100
0x180
0x1400
Full
0x080
0x100
0x280
0x1FFF
Half
0x080
0x100
0x280
0x1FFF
Quarter
0x080
0x100
0x280
0x1FFF
Full
0x080
0x100
0x480
0x1FFF
Half
0x080
0x100
0x480
0x1FFF
Quarter
0x080
0x100
0x480
0x1FFF
For more information about performing various clocking and delay calculations,
depending on the interface frequency and rate, refer to
PHYLite_delay_calculations.xlsx.
17.14 Using the EMIF Debug Toolkit with Arria 10 HPS Interfaces
The External Memory Interface Debug Toolkit is not directly compatible with Arria 10
HPS interfaces.
To debug your Arria 10 HPS interface using the EMIF Debug Toolkit, you should create
an identically parameterized, non-HPS version of your interface, and apply the EMIF
Debug Toolkit to that interface. When you finish debugging this non-HPS interface, you
can then apply any needed changes to your HPS interface, and continue your design
development.
17.15 Document Revision History
Date
May 2017
Version
2017.05.08
Changes
•
•
•
•
October 2016
2016.10.31
Added Using the EMIF Debug Toolkit with Arria 10 HPS Interfaces topic.
Added Calibration Adjustment Delay Step Sizes for Arria 10 Devices topic.
Replaced EMIF Configurable Traffic Generator 2.0 section with new Traffic
Generator 2.0 section.
Rebranded as Intel.
Maintenance release.
continued...
External Memory Interface Handbook Volume 3: Reference Material
513
17 External Memory Interface Debug Toolkit
Date
May 2016
Version
2016.05.02
Changes
•
•
•
•
November 2015
2015.11.02
•
•
•
•
•
•
•
•
May 2015
2015.05.04
•
•
•
•
December 2014
2014.12.15
•
•
Added additional option to step 1 of Establishing Communication to
Connections.
Added sentence to second bullet in Eye Diagram.
Expanded step 4 and added step 5, in Determining Margin.
Added Configuring the Traffic Generator 2.0 and The Traffic Generator 2.0
Report.
Changed title of Architecture section to User Interface.
Added sentence to Driver Margining section stating that driver margining
is not available if ECC is enabled.
Removed note that the memory map for Arria 10 On-Chip Debug would be
available in a future release.
Created separate On-Chip Debug sections for UniPHY-based EMIF IP and
Arria 10 EMIF IP.
Changed title of Driver Margining (Arria 10 only) section to Driver
Margining for Arria 10 EMIF IP.
Changed title of Read Setting and Apply Setting Commands (Arria 10
only) to Read Setting and Apply Setting Commands for Arria 10 EMIF IP.
Added section Example Tcl Script for Running the EMIF Debug Toolkit.
Changed instances of Quartus II to Quartus Prime.
Added Determining the Failing Calibration Stage for a Cyclone V or Arria V
HPS SDRAM Controller.
Changed occurrences of On-Chip Debug Toolkit to On-Chip Debug Port.
Added Driver Margining (Arria 10 only) and Determining Margin.
Added Read Setting and Apply Setting Commands (Arria 10 only) and
Reading or Applying Calibration Settings.
Added paragraph to step 5 of Generating IP With the Debug Port.
Added mention of seq_debug_clk and seq_debug_reset_in to step 5
of Generating IP With the Debug Port.
August 2014
2014.08.15
Maintenance release.
December 2013
2013.12.16
Maintenance release.
November 2012
2.2
•
•
•
August 2012
2.1
Added table of debugging tips.
June 2012
2.0
•
•
November 2011
1.0
Harvested 11.0 DDR2 and DDR3 SDRAM Controller with UniPHY EMIF Toolkit
content.
Changes to Setup and Use and General Workflow sections.
Added EMIF On-Chip Debug Toolkit section
Changed chapter number from 11 to 13.
Revised content for new UnIPHY EMIF Toolkit.
Added Feedback icon.
External Memory Interface Handbook Volume 3: Reference Material
514
18 Upgrading to UniPHY-based Controllers from ALTMEMPHY-based Controllers
18 Upgrading to UniPHY-based Controllers from
ALTMEMPHY-based Controllers
The following topics describe the process of upgrading to UniPHY from DDR2 or DDR3
SDRAM High-Performance Controller II with ALTMEMPHY Designs.
NOTE: Designs that do not use the AFI cannot be upgraded to UniPHY. If your design
uses non-AFI IP cores, Intel recommends that you start a new design with the UniPHY
IP core . In addition, Intel recommends that any new designs targeting Stratix III,
Stratix IV, or Stratix V use the UniPHY datapath.
To upgrade your ALTMEMPHY-based DDR2 or DDR3 SDRAM High-Performance
Controller II design to a DDR2 or DDR3 SDRAM controller with UniPHY IP core, you
must complete the tasks listed below:
1. Generating Equivalent Design
2. Replacing the ALTMEMPHY Datapath with UniPHY Datapath
3. Resolving Port Name Differences
4. Creating OCT Signals
5. Running Pin Assignments Script
6. Removing Obsolete Files
7. Simulating your Design
The following topics describe these tasks in detail.
Related Links
•
Generating Equivalent Design on page 516
Create a new DDR2 or DDR3 SDRAM controller with UniPHY IP core, by
following the steps in Implementing and Parameterizing Memory IP and apply
the following guidelines:
•
Replacing the ALTMEMPHY Datapath with UniPHY Datapath on page 516
To replace the ALTMEMPHY datapath with the UniPHY datapath, follow these
steps:
•
Resolving Port Name Differences on page 517
Several port names in the ALTMEMPHY datapath are different than in the
UniPHY datapath.
•
Creating OCT Signals on page 518
In ALTMEMPHY-based designs, the Quartus Prime Fitter creates the alt_oct
block outside the IP core and connects it to the oct_ctl_rs_value and
oct_ctl_rt_value signals.
•
Running Pin Assignments Script on page 518
Remap your design by running analysis and synthesis.
•
Removing Obsolete Files on page 519
Intel Corporation. All rights reserved. Intel, the Intel logo, Altera, Arria, Cyclone, Enpirion, MAX, Nios, Quartus
and Stratix words and logos are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other
countries. Intel warrants performance of its FPGA and semiconductor products to current specifications in
accordance with Intel's standard warranty, but reserves the right to make changes to any products and services
at any time without notice. Intel assumes no responsibility or liability arising out of the application or use of any
information, product, or service described herein except as expressly agreed to in writing by Intel. Intel
customers are advised to obtain the latest version of device specifications before relying on any published
information and before placing orders for products or services.
*Other names and brands may be claimed as the property of others.
ISO
9001:2008
Registered
18 Upgrading to UniPHY-based Controllers from ALTMEMPHY-based Controllers
After you upgrade the design, you may remove the unnecessary ALTMEMPHY
design files from your project.
•
Simulating your Design on page 519
You must use the UniPHY memory model to simulate your new design.
•
Creating OCT Signals on page 518
In ALTMEMPHY-based designs, the Quartus Prime Fitter creates the alt_oct
block outside the IP core and connects it to the oct_ctl_rs_value and
oct_ctl_rt_value signals.
18.1 Generating Equivalent Design
Create a new DDR2 or DDR3 SDRAM controller with UniPHY IP core, by following the
steps in Implementing and Parameterizing Memory IP and apply the following
guidelines:
•
Specify the same variation name as the ALTMEMPHY variation.
•
Specify a directory different than the ALTMEMPHY design directory to prevent files
from overwriting each other during generation.
To ease the migration process, ensure the UniPHY-based design you create is as
similar as possible to the existing ALTMEMPHY-based design. In particular, you should
ensure the following settings are the same in your UniPHY-based design:
Note:
•
PHY settings tab
•
FPGA speed grade
•
PLL reference clock
•
Memory clock frequency
•
There is no need to change the default Address and command clock phase
settings; however, if you have board skew effects in your ALTMEMPHY design,
enter the difference between that clock phase and the default clock phase into the
Address and command clock phase settings.
•
Memory Parameters tab—all parameters must match.
•
Memory Timing tab—all parameters must match.
•
Board settings tab—all parameters must match.
•
Controller settings tab—all parameters must match
In ALTMEMPHY-based designs you can turn off dynamic OCT. However, all UniPHYbased designs use dynamic parallel OCT and you cannot turn it off.
Related Links
Implementing and Parameterizing Memory IP
18.2 Replacing the ALTMEMPHY Datapath with UniPHY Datapath
To replace the ALTMEMPHY datapath with the UniPHY datapath, follow these steps:
External Memory Interface Handbook Volume 3: Reference Material
516
18 Upgrading to UniPHY-based Controllers from ALTMEMPHY-based Controllers
1.
In the Quartus Prime software, open the Assignment Editor, on the Assignments
menu click Assignment Editor.
2. Manually delete all of the assignments related to the external memory interface
pins, except for the location assignments if you are preserving the pinout. By
default, these pin names start with the mem prefix, though in your design they
may have a different name.
3.
Remove the old ALTMEMPHY .qip file from the project, as follows:
a.
On the Assignments menu click Settings.
b.
Specify the old .qip, and click Remove.
Your design now uses the UniPHY datapath.
18.3 Resolving Port Name Differences
Several port names in the ALTMEMPHY datapath are different than in the UniPHY
datapath. The different names may cause compilation errors.
This topic describes the changes you must make in the RTL for the entity that
instantiates the memory IP core. Each change applies to a specific port in the
ALTMEMPHY datapath. Unconnected ports require no changes.
In some instances, multiple ports in ALTMEMPHY-based designs are mapped to a
single port in UniPHY-based designs. If you use both ports in ALTMEMPHY-based
designs, assign a temporary signal to the common port and connect it to the original
wires. The following table shows the changes you must make.
Table 138.
Changes to ALTMEMPHY Port Names
ALTMEMPHY Port
Changes
aux_full_rate_clk
The UniPHY-based design does not generate this signal. You can generate it if you
require it.
aux_scan_clk
The UniPHY-based design does not generate this signal. You can generate it if you
require it.
aux_scan_clk_reset_n
The UniPHY-based design does not generate this signal. You can generate it if you
require it.
dll_reference_clk
The UniPHY-based design does not generate this signal. You can generate it if you
require it.
dqs_delay_ctrl_export
This signal is for DLL sharing between ALTMEMPHY instances and is not applicable
for UniPHY-based designs.
local_address
Rename to avl_addr.
local_be
Rename to avl_be.
local_burstbegin
Rename to avl_burstbegin.
local_rdata
Rename to avl_rdata.
local_rdata_valid
Rename to avl_rdata_valid.
local_read_req
Rename to avl_read_req.
local_ready
Rename to avl_ready.
local_size
Rename to avl_size.
continued...
External Memory Interface Handbook Volume 3: Reference Material
517
18 Upgrading to UniPHY-based Controllers from ALTMEMPHY-based Controllers
ALTMEMPHY Port
Changes
local_wdata
Rename to avl_wdata.
local_write_req
Rename to avl_write_req.
mem_addr
Rename to mem_a.
mem_clk
Rename to mem_ck.
mem_clk_n
Rename to mem_ck_n.
mem_dqsn
Rename to mem_dqs_n.
oct_ctl_rs_value
Remove from design (see “Creating OCT Signals”).
oct_ctl_rt_value
Remove from design (see “Creating OCT Signals”).
phy_clk
Rename to afi_clk.
reset_phy_clk_n
Rename to afi_reset_n.
local_refresh_ack
reset_request_n
The controller no longer exposes these signals to the top-level design, so
comment out these outputs. If you need it, bring the wire out from the highperformance II controller entity in <project_directory>/<variation
name>.v.
Related Links
Creating OCT Signals on page 518
In ALTMEMPHY-based designs, the Quartus Prime Fitter creates the alt_oct block
outside the IP core and connects it to the oct_ctl_rs_value and
oct_ctl_rt_value signals.
18.4 Creating OCT Signals
In ALTMEMPHY-based designs, the Quartus Prime Fitter creates the alt_oct block
outside the IP core and connects it to the oct_ctl_rs_value and
oct_ctl_rt_value signals.
In UniPHY-based designs, the OCT block is part of the IP core, so the design no longer
requires these two ports. Instead, the UniPHY-based design requires two additional
ports, oct_rup and oct_rdn (for Stratix III and Stratix IV devices), or oct_rzqin
(for Stratix V devices). You must create these ports in the instantiating entity as input
pins and connect to the UniPHY instance. Then route these pins to the top-level design
and connect to the OCT RUP and RDOWN resistors on the board.
For information on OCT control block sharing, refer to “The OCT Sharing Interface” in
this volume.
18.5 Running Pin Assignments Script
Remap your design by running analysis and synthesis.
When analysis and synthesis completes, run the pin assignments Tcl script and then
verify the new pin assignments in the Assignment Editor.
External Memory Interface Handbook Volume 3: Reference Material
518
18 Upgrading to UniPHY-based Controllers from ALTMEMPHY-based Controllers
18.6 Removing Obsolete Files
After you upgrade the design, you may remove the unnecessary ALTMEMPHY design
files from your project.
To identify these files, examine the original ALTMEMPHY-generated .qip file in any text
editor.
18.7 Simulating your Design
You must use the UniPHY memory model to simulate your new design.
To use the UniPHY memory model, follow these steps:
1.
Edit your instantiation of the UniPHY datapath to ensure the local_init_done,
local_cal_success, local_cal_fail, soft_reset_n, oct_rdn, oct_rup,
reset_phy_clk_n, and phy_clk signals are at the top-level entity so that an
instantiating testbench can refer to those signals.
2. To use the UniPHY testbench and memory model, generate the example design
when generating your IP instantiation.
3. Specify that your third-party simulator should use the UniPHY testbench and
memory model instead of the ALTMEMPHY memory model, as follows:
a.
On the Assignments menu, point to Settings and click the Project Settings
window.
b.
Select the Simulation tab, click Test Benches, click Edit, and replace the
ALTMEMPHY testbench files with the following files:
•
\<project directory>\<variation name>_example_design
\simulation\verilog\submodules
\altera_avalon_clock_source.sv or \<project directory>
\<variation name>_example_design\simulation\vhdl
\submodules\altera_avalon_clock_source.vhd
•
\<project directory>\<variation name>_example_design
\simulation\verilog\submodules
\altera_avalon_reset_source.sv or \<project directory>
\<variation name>_example_design\simulation\vhdl
\submodules\altera_avalon_reset_source.vhd
•
\<project directory>\<variation name>_example_design
\simulation\verilog\<variation name>_example_sim.v or
\uniphy\<variation name>_example_design\simulation\vhdl
\<variation name>_example_sim.vhd
•
\<project directory>\<variation name>_example_design
\simulation\verilog\submodules\verbosity_pkg.sv
External Memory Interface Handbook Volume 3: Reference Material
519
18 Upgrading to UniPHY-based Controllers from ALTMEMPHY-based Controllers
4.
•
\<project directory>\<variation name>_example_design
\simulation\verilog\submodules
\status_checker_no_ifdef_params.sv or \<project directory>
\<variation name>_example_design\simulation\vhdl
\submodules\status_checker_no_ifdef_params.sv
•
\<project directory>\<variation name>_example_design
\simulation\verilog\submodules
\alt_mem_if_common_ddr_mem_model_ddr3_mem_if_dm_pins_en_
mem_if_dqsn_en.sv or \<project directory>\<variation
name>_example_design\simulation\vhdl\submodules
\alt_mem_if_common_ddr_mem_model_ddr3_mem_if_dm_pins_en_
mem_if_dqsn_en.sv
•
\<project directory>\<variation name>_example_design
\simulation\verilog\submodules
\alt_mem_if_ddr3_mem_model_top_ddr3_mem_if_dm_pins_en_me
m_if_dqsn_en or \<project directory>\<variation
name>_example_design\simulation\vhdl\submodules
\alt_mem_if_ddr3_mem_model_top_ddr3_mem_if_dm_pins_en_me
m_if_dqsn_en
Open the <variation name>_example_sim.v file and find the UniPHYgenerated simulation example design module name below: <variation
name>_example_sim_e0.
5. Change the module name above to the name of your top-level design module.
6. Refer to the following table and update the listed port names of the example
design in the UniPHY-generated <variation name>_example_sim.v file.
Table 139.
Example Design Port Names
Example Design Name
New Name
pll_ref_clk
Rename to clock_source.
mem_a
Rename to mem_addr.
mem_ck
Rename to mem_clk.
mem_ck_n
Rename to mem_clk_n.
mem_dqs_n
Rename to mem_dqsn.
drv_status_pass
Rename to pnf.
afi_clk
Rename to phy_clk.
afi_reset_n
Rename to reset_phy_clk_n.
drv_status_fail
This signal is not available, so comment out this output.
afi_half_clk
This signal is not exposed to the top-level design, so comment out this output.
For more information about generating example simulation files, refer to Simulating
Memory IP, in volume 2 of the External Memory Interface Handbook.
18.8 Document Revision History
External Memory Interface Handbook Volume 3: Reference Material
520
18 Upgrading to UniPHY-based Controllers from ALTMEMPHY-based Controllers
Date
Version
Changes
May 2017
2017.05.08
Rebranded as Intel.
October 2016
2016.10.31
Maintenance release.
May 2016
2016.05.02
Maintenance release.
November 2015
2015.11.02
Changed instances of Quartus II to Quartus Prime.
May 2015
2015.05.04
Maintenance release.
December 2014
2014.12.15
Maintenance release.
August 2014
2014.08.15
Maintenance release.
December 2013
2013.12.16
Removed local_wdata_req from port names table.
November 2012
2.3
Changed chapter number from 12 to 14.
June 2012
2.2
Added Feedback icon.
November 2011
2.1
Revised Simulating your Design section.
External Memory Interface Handbook Volume 3: Reference Material
521

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project