Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low-Cost Approaches WW05 Kazuya Yamamoto1, Hiroshi Okazaki2, Kenjiro Nishikawa3 1Mitsubishi Electric Corp., 2NTT DoCoMo R&D Labs, 3Kagoshima University Yamamoto.Kazuya@bk.MitsubishiElectric.co.jp Slide 1 of 190 Thermal management of electronics: Measurement and the limits of GaN‐ on‐diamond electronics Professor Martin Kuball University of Bristol Martin.Kuball@bristol.ac.uk WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 1 of 60 Outline GaN electronics Thermal management challenges Thermal materials and device characterization Ultra‐high power electronics: GaN‐on‐ Diamond HEMTs Conclusions WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 2 of 60 Microwave GaN Electronics www.fujitsu.com You can buy this already (GaN‐on‐SiC, GaN‐on‐Si) !!! WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 3 of 60 Thermal management Electronic Warfare Challenges: New material and device design; how to measure channel temperature of a device ? Directed Energy Systems Radar Systems WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 4 of 60 Performance and reliability Impact ionization Bulk trap generation Interface trap generation Killat et al., Compound Semiconductor Jan/Feb 2013 Device degradation is temperature and electric field accelerated. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 5 of 60 Electric field & temperature Source field plate Passivation Gate Drain AlGaN Channel GaN Electric fields can be ‘shaped’ using e.g. field plates, T‐shaped gate, slanted gate i.e., electric field driven device degradation can be limited. Nucleation layer 4H SiC Temperature is the main factor at present limiting the reliability ie determining the maximum possible power density. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 6 of 60 GaN HEMT thermal history Worked example 20W 10GHz solid‐state module GaAs based Cell size determined by operating frequency module PRESENTLY 4x reduction in size 5W 4.47W 0.5dB manifold loss GaN HEMT Vds=28V (5W/mm) 1 x 1mm transistor Zopt=50ohm // 0.2pF GaAs pHEMT Vds=8V (0.9W/mm) 4 x 1.38mm transistor Zopt=4.5ohm // 1.1pF GaN based module GaAs (0.5 W/mK) thermal conductivity much lower than GaN‐on‐SiC (1.6; 4.5 W/mK). Is this the limit WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 7 of 60 Channel temperature Device degradation determined buy temperature Rate of device failure exp(‐Ea/kBT), with Ea = activation energy T = channel temperature (or temperature at specific location inside channel). Gate metal diffusion J. A. del Alamo et al., IEEE IEDM 2004 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 8 of 60 Why waste with experiment I can do a thermal simulation … this saves me a lot of money and time Heater width ? Heat has to traverse interfaces Thermal conductivity of material and variation through layer(s) WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 9 of 60 IR Thermography Basic principle: Measures intensity of thermal IR radiation Measured intensity: with σ Stefan‐Boltzmann constant Intensity T2 T 2 > T 1 T1 10 m Wavelength 1 m Often 3‐5 m or 8‐10 m spectral window is used. Fast, but diffraction‐limited spatial resolution of >3‐10 µm. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 10 of 60 IR Thermography Limited lateral resolution: IR: 3‐10 m Typical ‘no’ depth resolution (for uncoated devices): often 2‐5 m IR measures a temperature average which is often not easy to define. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 11 of 60 Electrical Methods Basic principle: Quantifies changes in IV curve with temperature rise, e.g., a change in saturation current. E.g. Kuzmik et al, IEEE Trans. Electron Dev.48 1496 (2002); McAllister et al., J. Vac. Sci. Techn. 24, 624 (2006); Simms, IEEE Trans. Electron Dev. 55, 478 (2008). Advantage: Uses electrical test equipment standard in most laboratories; measures however average temperature over whole device. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 12 of 60 Raman thermography Based on that vibrations of ‘atoms’ (phonons) of materials are temperature dependent off on Spatial resolution 0.5‐0.7 µm. Temperature resolution < 2‐5 C. Time resolution 10 ns. Easy to use. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 13 of 60 The Raman method Photons of a laser ‘hit’ device Scattered light contains three different wavelengths Photon energy change Light of same photon energy as laser Light of increased photon energy (photon absorbs a vibration of ‘atoms’) Light of reduced photon energy (photon created a vibration of ‘atoms’) Raman probes typically this light M. Kuball, Surf. Interface Anal. 31, 987 (2001). WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 14 of 60 A typical Raman spectrum SiC GaN GaN GaN SiC Temperature in ‘all’ different material layers in a device can be probed simultaneously: GaN, SiC temperature in GaN/SiC HEMT, AlGaAs, GaAs temperature in GaAs pHEMT, … WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 15 of 60 Raman vs IR thermography 160 120 Raman IR 3D sim. gate Temperature [oC] AlGaN/GaN HEMT on SiC drain source Traditionally used IR underestimates temperature by 40% 80 M. Kuball et al., CSICS 2007 -10 -5 0 5 10 Position [m] Spatial resolution: Raman 0.5‐0.7 μm; IR 7 μm. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 16 of 60 Importance of interfaces 180 100 80 60 40 20 0 GaN SiC Thermal Boundary Resistance (TBR) GaN 0.1 SiC 1 10 100 1000 Depth (m) CS-6 5 URI-4 4 30% CS-2 CS-5 URI-2 3 20% URI-3b CS-1 1 URI-1 100 Manoi et al., IEEE Electron Dev. Lett 31, 1395 (2010). CS-4 URI-3a 2 CS-3 150 Measurement accuracy 200 10% 5% Extra channel temperature rise o Temperature rise ( C) 120 TBR x10-8 (m2K/W) × 140 US, Europe, Japan 6 Thermal resistance of interface 160 250 Interface temperature (oC) WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 17 of 60 How to obtain time resolution ON Pulsed laser source OFF Best ever reported IR time resolution Time resolution: 10 ns. Spatial resolution: 0.5 μm. Order of magnitude faster and more accurate than other thermography techniques. Riedel et al., IEEE Electron Dev. Lett. 29, 416 (2008); Kuball et al. IEEE Electron Dev. Lett. 28, 86 – 89 (2007). WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 18 of 60 Time‐resolved Raman S 100 m Ability to trace temperature with 10ns time resolution D S G 25 µm D S Temperature splits Identical increase Time‐dependent thermal cross‐talk Manoi et al, Solid State Elecronics 57, 14 (2011). Slide 19 of 60 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Can one improve resolution ? (a) Source Gate (b) Drain Source Gate Solid Immersion Lens (SIL) Drain GaN θ SiC n=2.6 R SIL air (c) Axial: 6 widefield confocal Designed SIL 4 0.4 SiC SIL limit 0.2 2 0 0.0 1 N.A. drain source o Axial resolution [m] 8 heat 0.6 Temperature [ C] Lateral: widefield confocal Lateral resolution [m] 10 180 160 gate Upper GaN (Exp.) Lower GaN (Exp.) Average GaN (Sim.) 2 2 3 Lateral position [m] J.W. Pomeroy et al., J. Appl. Phys. 118, 144501 (2015) 4 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 20 of 60 DC vs RF temperatures Junction temperature 256°C 225°C 1E+8 175°C 1E+7 MTTF [hrs] 1E+6 RF test 1E+5 Ea=1.92eV Ea=1.82eV 1E+4 1E+3 DC test 1E+2 1E+1 18 19 20 21 22 23 24 25 26 27 28 1/KT [1/eV] Activation energy determined in DC and RF lifetime test similar. J.W. Pomeroy, Microelectronics Reliability 55, 2505 (2015). WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 21 of 60 However, for >100V 5.9 W/mm average 100V J.W. Pomeroy et al. ROCS 2015. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 22 of 60 How to do it in real life ? top Measure Raman from bottom Can be performed on‐wafer or in‐package. Only condition is that the semiconductor of the device is optically visible. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 23 of 60 Temperature measured For optically transparant materials – Average of temperature in small volume Gate 2 m Ga 0.75 m N C Si 0.75 m For optically non‐ transparant materials – Small ‘surface’ area Temperature average over 0.75 m x 0.75 m x 50nm (for GaAs) It is well defined over which area an average of temperature is measured and should be compared a subsequent thermal simulation !!! WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 24 of 60 Simulation to aid experiment Pomeroy et al, IMS 2012 If there is a T‐gate or field plate, we consider this by using thermal simulation, as those ‘screen’ the hot spot. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 25 of 60 Methodology developed … 180 Laser 160 light Step 2 – We fit in thermal simulation any resistances at internal chip interfaces. Step 1 – We fit in thermal simulation the gradient in temperature through SiC to extract SiC thermal conductivity. Ungated device Ohmic 140 contact 120 o Temperature rise ( C) Ohmic contact AlGaN GaN 100 Depth scan 80 60 40 20 Extract key material parameters 0 GaN 0.1 SiC 1 10 source 100 gate 1000 drain Depth (m) HEMT GaN SiC depth Step 3 – We measure temperature in active device region and determine Edge of Chip (EoC) temperature (temperature 500 m away from device). Step 4 – We fit thermal simulation to HEMT and EoC temperature; EoC temperature considers die attach. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 26 of 60 Thermal simulation 2D drift‐diffusion model 3D finite element thermal model AlGaN e.g., 20VDS, ‐3VGS GaN 3.75µm drain source 20nm Gate Gate F.P. edge F.P. edge Joule heating map Calibrated 3D thermal model including die and package This approach combines the advantage of accurate Pdiss profile (drift‐diffusion) with 3D finite element, e.g. large models J. W. Pomeroy et al., Microelectron. Reliab., (55)12, 2505 (2015). WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 27 of 60 Good repeatability ≤ ±5°C WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 28 of 60 GaN‐on‐Diamond HEMT Qorvo & Bristol, CSICS 2013; Pomeroy et al, Appl. Phys. Lett. 104, 083513 (2014). A 3× power density increase was achieved by the DARPA NJTT program WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 29 of 60 Thermal conductivities Single crystalline diamond Enriched Diamond Natural diamond J.R. Olson, Phys. Rev. Lett. 70, 14 (1993). SiC thermal conductivity: 4.8 W/cmK WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 30 of 60 Single‐crystalline diamond Not a realistic option to use for semiconductor technology WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 31 of 60 Poly‐crystalline diamond Grain size Defects phonon‐phonon WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 32 of 60 The role of interfaces WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 33 of 60 Importance of interfaces Source Drain Gate GaN interface Diamond ¼ model Raman probe 1 µm Ungated Transistor 2 finger transistor Polycrystalline diamond properties as well as interfaces need to be optimized. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 34 of 60 Diamond thermal conductivity GaN contact map mesa Diamond interface contact Polycrystalline Diamond F.E. model of ungated HEMT ¼ cross section Fit finite element model by adjusting two parameters: Diamond thermal + GaN/diamond conductivity interface TBReff Pomeroy et al, CSICS 2014. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost 35 Approaches Slide 35 of 60 Diamond thermal properties TBReff: Effective thermal boundary resistance Effective thermal conductivity: Weighted average, influenced by grain size. Substrate T.C. [W/mK] GaN Heat flux 160 W/mK thin dielectric Increasing thermal conductivity along growth direction 100 µm polycrystalline diamond TBReff ‐8 ×10 [m2K/W] GaN‐on‐SiC 420 2‐5 (~2.5 typical) GaN‐on‐di 1200 (effective) 2.7±0.3 O.W. Kiiding et al.Diamond Relt.Mater., 3 (1994) 1178 REMINDER: Bulk diamond: 2000‐3000 W/mK (Pomeroy et al., CSICS 2013) WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 36 of 60 Near nucleation diamond Thermal conductivity impacted by grain size Grain evolution can be controlled by manipulating the chemistry of the diamond growth. J.Anaya et al. Acta Materiala 103, 141 (2016); Appl. Phys. Lett. 106, 223101 (2015). WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 37 of 60 Near nucleation diamond WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 38 of 60 Diamond beyond nucleation Z 3‐heaters X Thermal conductivity. determined by 3‐omega technique. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 39 of 60 Diamond beyond nucleation R.Baranyai et al. APEX 9, 061302 (2016) WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 40 of 60 Modeling of diamond properties (i) thermal resistance between grains, (ii) shortening in the phonon mean free path due the reduced size of the grains S. Bhattacharyya et al. Appl. Phys. Lett. 79, 1441 (2001) Grain Boundary thickness Callaway , , , , , 1 J.Anaya et al., Acta Materiala 103, 141 (2016). , , Kapitza‐like WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 41 of 60 GaN‐diamond interface WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 42 of 60 Transient Thermoreflectance Wafer mapping & fast wafer screening Varied thickness 28 nm to 100 nm Presently we develop this into a commercial equipment. Sun et al. IEEE Electron Dev. Lett. (accepted) 2016; Appl. Phys. Lett. 106, 111906 (2015); Pomeroy et al. IEEE Electron Dev. Lett. 35, 1007 (2014). Slide 43 of 60 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Validation of the technique Normalized signal Exp. Sim. surface temp. Sim. total refl. 0 200 400 600 800 t (ns) Measured transient is consistent on wafers w/ and w/o gold transducer Precautions have been taken to ensure that the measured signal represents the surface temperature transient. • Different UV powers result in identical transients. H. Sun et al, CSMantech 2015 • Thermo-optic simulation further supports data. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 44 of 60 GaN‐diamond interface Dielectric seeding layer needs to be optimized. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 45 of 60 GaN‐diamond interface 2000 sub (W/m-K) 1500 H. Sun et al, CSMantech 2015 200 °C MW diamond 100 1000 50 • HF diamond 500 60 Transistor peak temperature rise SiC • 50 40 30 20 10 When substrate thermal conductivity is low, TBReff is not the major factor. However, TBReff limits heat removal for high thermal conductivity substrates. 2 TBReff (m K/GW) WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 46 of 60 Wafer screening Rare imperfections can be screened out Typically good homogeneity Fast wafer‐mapping of the GaN‐on‐ Diamond thermal resistance WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 47 of 60 Thermoreflectance for devices D Materials: Field Plate Passivation S G •Two metallization levels: 1.Drain contact ( ) 2.Source field plate ( ) GaN •GaN ( ) SiC or Diamond D 45 μm 6 μm 6 μm G Field plate S S Martin Horcajo et al., CS Mantech 2016 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 48 of 60 GaN‐diamond HEMT design Decreasing thermal resistance associated with the carrier enables a further Increase in power density to ~3X J. W. Pomeroy et al, CSICS 2014 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 49 of 60 GaN layer optimization Source TBReff d Drain Gate 1 µm Low TBR A 1µm‐thick GaN buffer is optimal for the range TBReff values expected d [m] 1.0 Measured TBR Peak Temperature rise [Norm.] Diamond 1.2 e.g. 4 Finger HEMT Vary GaN buffer thickness (d) and TBReff Heat spreading GaN 0.8 Thin GaN 0.75 1 1.5 2 2.5 3 0.6 0 1 2 -8 3 J. W. Pomeroy et al, CSICS 2014 4 2 TBR x10 [Wm /K] WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 50 of 60 Raman experimental validation 260 GaN‐on‐SiC 240 GaN Diamond S G D GaN 220 OR decreasing interface thermal resistance We validate all thermal simulations with Raman thermography measurements. 180 diamond W/mK 710 1400 710 o Current GaN‐on‐diamond Increasing diamond thermal conductivity Temperature rise [ C] 200 Measurement: GaN Diamond Simulation: diamond Rinterface 160 o 2 Cm /W -8 2.7x10 -8 2.7x10 0 140 120 100 80 60 40 20 0 0.1 1 Depth [m] 10 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches 100 Slide 51 of 60 GaN‐diamond stability WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 52 of 60 Coefficient of thermal exansion Diamond WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 53 of 60 Stress in GaN layer Diamond Determined using Raman spectroscopy Year 2011 With nitride transition layers 0.8 2012 2013 Without nitride transition layers 565.0 566.0 0.2 0.0 -0.2 566.5 567.0 Stress free Hot filament Wafer Microwave E2 peak (cm-1) 0.4 565.5 Tensile Stress (GPa) 0.6 567.5 Sun et al, CS Mantech 2016 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 54 of 60 How to test for stability ? WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 55 of 60 High mechanical stability GaN Cantilever to appy force diamond We try to cause fracture at the GaN-diamond interface D. Liu et al, Appl. Phys. Lett., 107, 251902 (2015). WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 56 of 60 Estimation of interface strength GaN layer fractures as >3 GPa; GaN/diamond interface fracture strength is much greater. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 57 of 60 Thermomechanical stability Stress in GaN‐on‐ diamond induced by heating (CTE mismatch) Fracture at > 3GPa Good thermo-mechanical stability in the areas studied ie no change in the TBR after annealing. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 58 of 60 Conclusions GaN electronics main challenge at present is its heat sinking. Raman thermography offers the opportunity to quantify channel temperature, and to identify and optimize thermal bottlenecks such as interfaces. Raman thermography enables 0.5µm spatial and 10ns time resolution thermal imaging in 3D, which can be improved further using solid immersion lenses (SILs). GaN‐on‐Diamond HEMTs enable 3x improvement in power density, however, require optimization in diamond thermal properties near the interface and of the interface itself. Transient thermoreflectance can be used for wafer mapping of thermal interfaces (before device fabrication) and for device thermal analysis. High mechanical & thermo‐mechanical stability of GaN‐diamond interface was demonstrated. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 59 of 60 Acknowledgment Research Professor & Research Fellows Prof. Michael J. Uren Dr. James W. Pomeroy Dr. Dong Liu Postdoctoral Researchers Dr. Julian Anaya Dr. Roland Baranyai Dr. Tommaso Brazzini Dr. Indranil Chatterjee Dr. Sara Horcajo Dr. Huarui Sun Dr. Serge Karbojan PhD students: Peter Butler Callum Middleton Bahar Oner Alexander Pooth Maire Power Ben Rakauskas Will Waller Yan Zhou and others WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 60 of 60 • • • • — — — — — • • • • • • • 5 • • — — • • • • Æ — — — — • • • Æ Æ Æ • • – – – – • ° • • • – – • • ° • • ° • — • • • • ° • — — • — — • – • – • • • — — Æ • • • • — — • — • 9 9 9 9 ITEM UNITS Bulk Thermal Conductivity W/m-K CTE ppm/⁰C Modulus @ RT Gpa AuSn 57 16 59 Ag Epoxy 23 38 3.0 Mfr A 50 26 6.7 Sintered Ag Mfr B 100 28 10 Mfr C 180 22 26 • — • • • — — • • • • • • — — — — — — • • • • • • Pout(dBm) Drain Efficiency (%) 20 16 Gain, Return Loss (dB) 12 8 4 0 -4 -8 -12 S11 S21 S22 -16 -20 1.0 1.1 1.2 1.3 1.4 1.5 1.6 61.0 80 60.5 70 60.0 6 60 59.5 50 59.0 40 58.5 30 58.0 Output Power (dBm) 20 57.5 Drain Efficiency (%) 10 57.0 1.10 1.15 1.20 1.25 1.30 1.35 1.40 1.45 Drain Efficiency (%) Frequency (GHz) 0 1.50 Ω 20 10 5 0 -5 -10 S(2,1) S(1,1) S(2,2) -15 -20 2.5 2.7 2.9 3.1 3.3 3.5 Frequency (GHz) 800 100 700 90 600 80 500 70 400 60 300 50 Drain Efficiency (%) 2.3 Output Power (W) Gain, Return Loss (dB) 15 200 40 Output Power 100 30 Drain Efficiency 0 20 2.6 2.7 2.8 2.9 Frequency (GHz) 3.0 3.1 3.2 20 10 5 0 -5 -10 S11 S21 -15 S22 -20 4000 4200 4400 4600 4800 5000 5200 5400 5600 5800 6000 6200 6400 6600 6800 7000 51.0 100 50.5 90 50.0 80 49.5 70 49.0 60 48.5 50 48.0 Output Power (dBm) 40 47.5 Drain Efficiency (%) 30 47.0 20 5.2 5.3 5.4 5.5 5.6 Frequency (GHz) 5.7 5.8 5.9 Drain Efficiency (%) Frequency (MHz) Output Power (dBm) Gain, Return Loss (dB) 15 16 12 Gain, Return Loss (dB) 8 4 0 -4 -8 -12 S21 S11 S22 -16 -20 8.2 8.4 8.6 8.8 9.0 9.2 9.4 9.6 9.8 10.0 10.2 10.4 10.6 Output Power (dBm), Drain Efficiency (%) Frequency (GHz) 55 50 45 40 Drain Efficiency 35 Output Power 30 8.8 8.9 9.0 9.1 9.2 9.3 9.4 9.5 9.6 9.7 Frequency (GHz) 20 10 5 0 -5 -10 S21 S11 S22 -15 -20 3.0 3.1 3.2 3.3 3.4 3.5 3.6 3.7 Output Power and Efficiency vs. Frequency Frequency (GHz) 100 100 90 90 80 80 70 70 60 60 50 50 40 20 3.05 40 Output Power (W) 30 30 Efficiency (%) 3.10 3.15 3.20 3.25 3.30 3.35 Frequency (GHz) 3.40 3.45 3.50 20 3.55 Drain Efficiency (%) 2.9 Output Power (W) Gain and Return Loss (dB) 15 25 20 S21 10 S11 5 S22 0 -5 -10 -15 -20 1.85 1.9 1.95 2 2.05 2.1 2.15 2.2 2.25 2.3 2.35 2.4 Frequency (GHz) 57.0 75.0 56.5 70.0 56.0 65.0 55.5 60.0 55.0 55.0 Psat (dBm) 54.5 54.0 2.10 50.0 Efficiency (%) 2.11 2.12 2.13 2.14 2.15 Frequency (GHz) 2.16 2.17 45.0 2.18 Efficiency (%) 1.8 Output Power (dBm) Gain, Return Loss (dB) 15 • • • • • • Cost effective GaN HEMT developments with appropriate thermal transfer Kazutaka Inoue Sumitomo Electric Industries, Ltd. inoue‐kazutaka@sei.co.jp Wednesday, October, 5th, 2016 Slide 1 of 44 Outline 1. Fundamentals 2. Thermal Study of Substrate 3. Thermal Design (GaN for Radar) 4. Thermal Design (GaN for Base Station) 5. Summary Slide 2 of 44 History of Sumitomo GaN HEMTs ｰ2000 2003 2004 2005 Fujitsu, Fujitsu Laboratories Development Start 2006 2007 2008 Eudyna Devices 2009 2010 - Sumitomo Electric Device Innovations Engineering Sample Release Mass-production #1 Volume leader in RF high power GaN We strive for Performance Quality Reliability Cost Capacity 2013 GaN Market Share above 5W below 4GHz Source: ABI Research 2014 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 3 of 44 Microwave Device Products 【Mobile Base Station】 【Satellite】 【Automotive Radar】 GaAs MMIC GaN HEMT GaAs FET 【Radio Link】 GaAs MMIC 【Radar】 Gulf Control Air Traffic Control Marine GaN HEMT for Radar WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 4 of 44 Why GaN ? Johnson's Figure of Merit Johnson’s FoM = ft x BV = vs /2π x BF BV = BF × Lg Material Si GaAs SiC GaN Bang Gap Energy (eV) 1.1 1.4 3.2 3.4 Critical Breakdown Field (MV/cm) 0.3 0.4 3.0 3.0 1.5 0.5 4.9 1.5 600 1500 2.0 2.7 Thermal Conductance (W/cm/K) Mobility (cm2/V/s) Saturated Velocity (*107 cm/s) 1300 6000 1.0 1.3 GaN GaAs, InP Si, SiGe fT = vs / ( 2πLg ) WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 5 of 44 Cost Effective Design, Focused on Thermal Transfer Chip cost is determined by ... ‐ Wafer process cost and yield ‐ Maximum channel temperature design Maximum channel temperature is determined by... ‐ Reliability of the device technology ‐ Thermal conductance of material (on‐SiC, on‐Si) ‐ Chip pattern layout ‐ Operating condition (Thermal dissipation ↔ Efficiency) WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 6 of 44 Yield of GaN HEMT Wafer Process Psat= 52.2 dBm = 0.15 Gp= 16.6 dB = 0.15 Rth = 1.2 C/W = 0.02 BTS GaN HEMT 2.7GHz 160W, n = 9000 pcs, 181 lots, = lot average GaN HEMT wafer process has been refined, through over 10‐years mass production. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 7 of 44 Reliability of GaN HEMT DC‐HTOL test RF‐HTOL test (Vds=60V , Tch=250, 275, 300, 315degC) (Vds=55V, Tch=270, 290, 310degC, P4dB) K.Osawa et.al, ” Over 74% Efficiency, L-Band 200 W GaN-HEMT for Space Applications ,” EuMC2016. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 8 of 44 Material vs. Thermal Design Parameters, related to thermal design ; ‐ Thermal conductance of substrate material (SiC:4.9 Si:1.5 [W/cm・K] ) ‐ Substrate thickness (Typical thickness is ∼100 μm ) ‐ Gate to gate pitch (Wide pitch improves the degree of heat spreading) Gate Drain Source AlGaN GaN Gate Drain HEMT Layers∼1μm Substrate Thickness ∼100 μm Heat Spreading Slide 9 of 44 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Thermal Design vs. Efficiency Thermal Dissipation ＝ Pdc ‐ Pout + Pin ＝ Pdc × ( 1− Power‐Added‐Efficiency ) DC supply power (Pdc) RF input power (Pin) Power Amp. Thermal dissipation RF output power (Pout) Slide 10 of 44 Outline 1. Fundamentals 2. Thermal Study of Substrate 3. Thermal Design (GaN for Radar) 4. Thermal Design (GaN for Base Station) 5. Summary Slide 11 of 44 Case Study of GaN on‐Si ∼RF Performance S G RF loss(low gain) D Large parasitic C AlGaN 2DEG Low resistivity region Silicon Diffusion of the Ga & Al into the silicon substrate GaN HEMT on‐Si realize comparable performance to GaN on SiC. I. Makabe et al.,“ Improvement of RF performance of GaN‐HEMT on silicon substrate,”Proc. International Workshop on Nitride Semiconductors, No. TuEP12, Wrocław, Poland（Aug. 2014） Lg=0.6μm, f=2GHz Vds=50V, Idq=0A WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 12 of 44 Case Study of GaN on‐Si ∼Thermal Design (Sub. Thickness) 1. Simulated Thermal resistance is not comparable, even in 30 μm thickness. ・・・・・ 2. Excessive thinning causes chip warping @ GaAs MESFET Sub. Thickness : 20 μm Sub. Thickness : 30 μm GaN HEMT on‐Si or on‐SiC (Wg=36 mm) AuSn Chip (flat) AuSn Cu‐based Package Chip (warped) Package Base Package Base Net improvement is limited by chip warping ! Small deviation Large deviation Slide 13 of 44 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Case Study of GaN on‐Si ∼Thermal Design (Gate Pitch) Thermal Conductance SiC:4.9 Si:1.5 [W/cm・K] Chip (on‐SiC) Thermal Resistance Package Base Chip (SiC) Chip (on‐Si) AuSn Package Base Chip (Si) AuSn + PKG on‐SiC AuSn + PKG on‐Si Thermally equivalent pitch of GaN on‐Si is 1.6 times of GaN on‐SiC. SiC substrate cost used to be several times higher to the wafer process cost. In such cost structure, GaN on‐Si was a reasonable solution. It should be noted the 1.6 times chip size increases “net wafer process cost” of GaN on Si The power density of GaN HEMT has been increasing. It used to be ∼5 W/mm, but now reaches to ∼10 W/mm. Higher power density favors better thermal conductance material. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 14 of 44 Case Study of GaN on‐Si ∼Thermal Design (Chip Stretch) Two options to realize thermally equivalent GaN on‐Si Option.1 (Lateral Stretch) Option.2 (Vertical Stretch) on‐SiC 300μm on‐Si ‐ Larger PKG is required for on‐Si chip. ‐ Larger unit finger increases gate resistance, which degrades the gain by more than 1dB. Gain Effect 480μm PKG for on‐SiC ≧1dB PKG for on‐Si ? 300μm Judgement 480μm Unit Length Total cost strongly depends on PKG price. Trade‐off issue arises (vs.gain). GaN on‐SiC is the best solution at present, judging from both cost and RF property. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 15 of 44 In addition... ∼SiC Quality Improvement 10 years ago (Surface Defects ) SiC quality was one of the issues to be solved. The current SiC has realized smooth surface, and it contributes to the drastic improvement of the GaN on SiC cost structure. Current SiC (Drastically Improved ) Abnormal epitaxial growth Defect on substrate GaN SiC WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 16 of 44 Prospect of Substrate Material Power density [W/mm] Power density of GaN HEMT have been increased, and will be continued. The benchmark of GaN on Si vs. SiC indicates the availability of GaN on diamond. ? 2005 Higher power density favors better thermal conductance material. 2016 on‐Si ‐ RF‐loss elimination required ‐ Thermal limitation ‐ No more cost effective on‐SiC ‐ Expensive & many defects ‐ Chip selection needed ‐ Most promising 2025 ‐‐ ‐ Will be most promising ‐ Thermal design may be critical @ >20W/mm ‐ Competitive @>20W/mm ‐ Quality & cost required Slide 17 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches of 44 on‐diamond ‐‐ ‐ Similar to early stage SiC ? Outline 1. Fundamentals 2. Thermal Study of Substrate 3. Thermal Design (GaN for Radar) 4. Thermal Design (GaN for Base Station) 5. Summary Slide 18 of 44 Output Power vs. Frequency Pout Satellite CW operation 1KW Case.2 BTS Backed-off operation 100W Case.1 Radar Pulsed operation Si 10W 1W GaN Si/GaAs 1GHz GaAs/GaN GaAs 10GHz Frequency 100GHz WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 19 of 44 Satellite Application (CW‐operation, as a reference) K.Osawa et.al, ” Over 74% Efficiency, L-Band 200 W GaN-HEMT for Space Applications ,” EuMC2016. Gate width Chip size ： 24 mm × 2 chip ： 6.0 ×0.86 mm (5.16mm2) Design point of 100W GaN HEMT die Designed for CW and saturated operation . (Most severe thermal requirement) WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 20 of 44 Case.1 Rader (Pulsed operation) Rader System Object CW operation Channel Temp. Pulsed operation ON Device OFF Tch image Thermal design of pulsed operation chip differs from CW one, and the estimation of the transitional Tch is important. The following slides explain the detail. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 21 of 44 Channel Temperature Simulation using Transient Thermal Resistance RF/Vg ON OFF T‐ (Rth Approximation) Rth() dTch peak1 -(Rth Approximation) During the pulse‐ON, channel temperature rises due to the heat generation. At the pulse‐OFF, the channel temperature falls down by heat dissipation, which is expressed as reverse behavior of heat generation. With combining these 2 lines, the channel temperature falls. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 22 of 44 Channel Temperature Simulation using Transient Thermal Resistance RF/Vg ON Rth(T+)-Rth(T) OFF T‐ Rth(2T+)-Rth(2T) (Rth Approximation) Rth() dTch peak1 dTch peak 1 = Pd*Rth() -(Rth Approximation) dTch peak3 dTch peak2 dTch peak 2 = Pd*{Rth(T+)-Rth(T)+Rth()} = Pd* {Rth(T+)-Rth(T)}+dTch peak 1 dTch peak 3 = Pd*{Rth(2T+)-Rth(2T)+Rth(T+)-Rth(T)+Rth()} = Pd*{Rth(2T+)-Rth(2T)}+dTch peak 2 The peak Tch at the 1st pulse is calculated as shown in above. The 2nd peak Tch is also calculated, combining these 3 values. The total Tch rise is summation of the delta Tch at the each on‐state condition. Slide 23 of 44 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Channel Temperature Measurement & Analysis 150 150 Shutter Speed 1msec 140 140 = Averaging Time 130 120 120 T ch [degC] T ch [d e g C ] Simulated Tch 130 Simulated IR response 110 100 90 110 100 80 80 70 70 60 60 50 50 -4 -2 0 2 Time [msec] 4 6 Simulated IR response 90 Measured IR Tch -4 -2 0 2 4 6 Time [msec] The analysis performed more than 10 years ago. We only had an IR system with rather slow shutter speed. But we found the measured Tch from IR agreed with the simulated Tch curve, by taking the shutter speed into consideration. Thus, we concluded that the mentioned simulation in previous slide is reliable. And the compact kW‐class GaN device was developed by utilizing this analysis. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 24 of 44 kW‐Class GaN HEMT Pallet Amplifier for Radar VGG VDD RF out RF in VDD VGG Size : 58.5 mm X 40.0 mm X 8.0 mm • Input/Output matched to 50 ohm • Includes RC Bias Circuit • Cu base E.Mitani et.al, ” A kW-class AlGaN/GaN HEMT pallet amplifier for S-band high power application ,” EuMIC2007. Slide 25 of 44 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches kW‐Class GaN for Radar RF Performance IDD(DC)=2.0A, Pulse Width 200sec, Duty 10% 59.4dBm 60.0dBm VDD=80 V 60 90 f=2.9 GHz 70 55.0% 55 60 50 40 30 50 13.8dB 20 80 Output Power [dBm] 80 Drain Efficiency [%] Gain [dB] Output Power [dBm] 90 70 f=3.2 GHz 55 60 49.5% 50 40 30 50 14.1dB 20 10 0 45 35 40 45 Input Power [dBm] Output Power Drain Efficiency 50 Gain Drain Efficiency [%] Gain [dB] VDD=65 V 60 10 0 45 35 40 45 Input Power [dBm] Output Power Drain Efficiency 50 Gain Compact 1 kW class GaN HEMT for radar application has successfully demonstrated, by optimizing for pulsed operation. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 26 of 44 Up‐to‐date Achievement X‐band 300 W GaN HEMT for Rader K.Kikuchi et.al, ” An 8.5–10.0 GHz 310 W GaN HEMT for radar applications”, 2014 IEEE MTT-S Int. Microwave Symposium Digest, 2014. ： ： ： ： 55.2 dBm 14.5 55 Pout 14.0 GL 45 14.1 dB 13.0 40 35 13.5 42% PAE 12.5 30 55.2 dBm (333 W) 56 25 20 54 52 Gain 15 10 50 Pout 48 10.2 dB Gain (dB) 50 58 15.0 GL (dB) Pout (dBm), PAE (%) 60 24.0 mm × 17.4 mm 14.4 mm × 2chip (＞10 watt/mm) 5.4 mm ×0.7 mm Lg=0.4μm, Via-Hole Output Power (dBm) PKG size Gate width Chip size Chip outline 5 0 46 12.0 32 34 36 38 40 42 44 46 35 40 45 50 55 60 65 70 Input Power (dBm) Vds (V) Pulse Width = 100 sec, Duty = 10%, Frequency = 9.0 GHz Slide 27 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches of 44 30 Up‐to‐date Achievement X‐band 300 W GaN HEMT for Radar 400 Vds = 65 V, Idq = 0.80 A Pulse Width = 100 sec, Duty = 10% 350 Output Power (W) This work (Maximum) Frequency 8.5 – 10.0 (GHz) (9.0) Output Power 310 (W) (333) Power Gain 10.0 (dB) (10.2) CW 300 Pulse 250 This work 333 W 200 150 100 50 0 7 8 9 10 11 Frequency (GHz) WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches 12 Slide 28 of 44 Prospect of Cost Effective GaN for Rader applications Using GaN to replace the klystron ‐ High reliability, Maintenance free ‐ Compact, Easy operation ‐> Reduction of OPEX GaN output power is not as large as klystron, but it allows long pulses. Pulse compression can be used. Cost effective GaN HEMT will penetrate into the radar applications. Filter Linear FM Pulse Frequency Dependent Delay Filter http://www.river.go.jp/xbandradar/ Compressed Pulse WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 29 of 44 Outline 1. Fundamentals 2. Thermal Study of Substrate 3. Thermal Design (GaN for Radar) 4. Thermal Design (GaN for Base Station) 5. Summary Slide 30 of 44 Case.2 BTS Application (Backed‐off Operation) Drain Efficiency Power Probability BTS trend → Smaller → Higher efficiency W‐CDMA WiMAX LTE, LTE‐adv 5G 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 year Normalized Output Power Both efficiency improvements of device itself & efficiency enhancement circuit techniques are required, at backed‐off power region. Slide 31 of 44 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Operation Class (F, inverse‐F) D Id G Inverse Class-F Class-F Vd Ids Ids Vds Vds Id, Vd Id, Vd S 0 90 180 270 360 ωt Load Impedance for Harmonics Even : 0 (Short) Odd : ∞ (Open) Peak Value of Voltage, Current Voltage: 2x Vds(DC) Current: 3.14x Average Ids 0 90 180 270 360 ωt Even : ∞ (Open) Odd : 0 (Short) Voltage: 3.14x Vds(DC) Current: 2x Average Ids WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 32 of 44 Combination of Inv. Class‐F & GaN HEMT Property Ids On-R 1.6 1.4 Vgs=+2V 1.2 Ids [A] Inverse-F -F 1.0 0.8 Vgs=+0V 0.6 0.4 Rds Vgs=-1V 0.2 0 0 50 Vds 100 150 200 Vds [V] Inverse Class‐F requirement are, ‐ Low Ron (Steep Knee) reduces RF‐loss ‐ High breakdown voltage contributes to keep good pinch off GaN HEMT is the best technology for these requirements at present. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 33 of 44 Circuit of Inverse Class‐F 100W GaN HEMT Design Concept Circuit : Harmonic tune by simple(=low cost) LC circuit (LPF) GaN chip : Size minimization by referring thermal loss Capacitor (r=150) Input 100W GaN Chip Output Product: EGN21C105I2D Vds=50V, f=2.14GHz 100 50 90 48 80 46 70 44 60 42 50 40 40 38 30 36 20 Drain Efficiency [%] 52 Pout [dBm] Inductor • Bonding Wire • Transmission Line on Alumina sub. 18 20 22 24 26 28 30 32 34 36 38 Pin [dBm] Slide 34 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches of 44 Id Theory of Doherty Amplifier (a) Backed‐off Zopt Peak Amp. OFF /4 Power Matching Open Class-C Main Amp. /4 Zopt/2 2xZopt Vd Class-B or AB 2xZopt Efficiency Matching (b) Peak Power Peak Amp. ON /4 Class-C Main Amp. Class-B or AB Zopt /4 depend on both main and peak amp's eff. depend on main amp's eff. Zopt/2 B A Efficiency Zopt b a Zopt 6dB BO Psat WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Pout Slide 35 of 44 GaN HEMT for Doherty PA a b gm Rds Rs c Drain Ids Rd Cds a) Ron DC IV-Characteristics RF Signal Loss b) Rds Source c) Cds Vds Key of High Frequency Performance 70W-class Device for 2.6GHz Base Stations 70W-class Device Vds Imax Wg Cds Si LDMOS-FET 28V 26A 120mm 31pF GaN HEMT (This work) 50V 11A 18mm 3.8pF H. Deguchi et.al, "A 33W GaN HEMT Doherty Amplifier with 55% Drain Efficiency for 2.6GHz Base Stations”, 2009 IEEE MTT-S Int. Microwave Symposium Digest, pp.1273-1276, 2009. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 36 of 44 RF Performance of Doherty PA Pout Pout(Pulse) Drain Efficiency 55 100 50 45 80 40 70 35 Effi. 30 60 50 25 40 20 30 15 20 Gain 10 Cds-reduced Device 90 Pout Drain Efficiency [%] D rain E fficiency [% ] Output P ower [dB m ], Gain [dB ] Vds = 50V, Idq-m = 200mA Idq-p : Class-C Bias f = 2.6GHz, CW Pulse Duty = 10% (6s/60s) Gain Gain(Pulse) Conventional Device Freq.=2.57GHz, Vds=50V 10 5 0 20 22 24 26 28 30 32 34 36 38 40 42 Input Power [dBm] H. Deguchi et.al, "A 33W GaN HEMT Doherty Amplifier with 55% Drain Efficiency for 2.6GHz Base Stations”, 2009 IEEE MTT-S Int. Microwave Symposium Digest, pp.1273-1276, 2009. Slide 37 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches of 44 Example of Envelope Tracking (ET) Envelope Tracking Operation Fixed Vds Supply Vds Vds Conventional Operation Time Time Output Signal Output Signal F.Yamaki et.al, ” A 65 % drain efficiency GaN HEMT with 200 W peak power for 20 V to 65 V envelope tracking base station amplifier ,” 2011 IEEE MTT-S Int. Microwave Symposium Digest, 2011. Modulated Vds Supply GaN HEMT for Envelope‐Tracking 1. Cds reduction & Vds dependence reduction 2. BVdsx ≧300 V , for 20‐65V ET‐operation Imax =680mA/mm, Cds=0.15pF/mm, BVdsx≧300V WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 38 of 44 Optimum Thermal Design for BTS 2. Efficiency Boosting Output Power [W] General use Depend on efficiency Specified for BTS 1. Thermal Design ∼2005 2006 2009 2011 Circuit Technique Class‐AB Higher Class Doherty w/Class‐F ‐1 ET Drain Eff. @8dB‐B.O. 34% 45% 55% 68% Thermal Dissipation Ref. 83% 68% 48% 2 1 Chip Size [mm2] Slide 39 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches of 44 RRH* BTS‐system and GaN HEMT * Remote Radio Head Base Station RRH RRH RRH Optic Fiber（0.1-20km) Less than20kg/20L →Easy to Settle Control Unit CAPEX (Capital Expenditure) ‐> Cost reduction of GaN itself has been progressed, and small & light weight PA, utilizing GaN high efficiency, contributes to BTS setting cost. OPEX（Operating Expenditure） ‐> Higher efficiency PA realizes lower power consumption. Slide 40 WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches of 44 Microwave Device Market & SEI GaN HEMT Shipment Total $1600M (@2013) Si‐LDMOS GaN HEMT WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 41 of 44 Outline 1. Fundamentals 2. Thermal Study of Substrate 3. Thermal Design (GaN for Radar) 4. Thermal Design (GaN for Base Station) 5. Summary Slide 42 of 44 Summary (1/2) GaN HEMTs have already realized high quality, uniformity and reliability for infrastructure RF power applications. Thus, the cost reduction has been strongly required. The adequate thermal transfer design is one of the solutions. GaN on‐SiC has been proved the best material to satisfy both the cost and thermal requirements. The focus design on pulsed operation is effective for compact radar devices, and the efficiency boosting in backed‐off region have reduced the chip size of base station PAs. Quality SiC crystal, Epi‐growth uniformity, Small process deviation Reliability Stringent qualification, Ruggedness Cost Adequate thermal design, utilizing SiC material property GaN HEMT to various applications (Satellite, Radar, Base station) Slide 43 of 44 Summary (2/2) GaN HEMT has already been adopted in several markets. For further market penetration, continuous cost down efforts are essential. Price 90W GaN Chip x2 (Not specified for BTS) 24 x 17.4 mm 210W GaN Chip (Specified for BTS) 21 x 13.2 mm 4’’ Wafer 4 inch 2 inch Year 3 inch Slide 44 of 44 Cost effective approaches for European GaAs & GaN Power solutions Guillaume CALLET guillaume.callet@ums‐gaas.com WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches Slide 1 Outlilne What can be the reducing cost drivers for GaN on SiC solution ? Presentation of UMS Overview of GaN technologies Thermal analysis for the development of packaged GaN solution UMS Products & Foundry Solutions Conclusions Slide 2 Outlilne Presentation of UMS III-V company with 20 year experience in semiconductor (especially GaAs HEMT) Overview of GaN technologies Thermal analysis for the development of packaged GaN solution UMS Products & Foundry Solutions Conclusions Slide 3 UMS at a glance Founded in 1996 by gathering Thales and AIRBUS Defense and Space GmbH activities European source of RF MMIC solutions, GaAs and GaN foundry services 2 industrial facilities in Ulm (Germany) & Villebon (France) 400 people Villebon facility Ulm facility Slide 4 Outlilne Presentation of UMS Overview of GaN technologies Thermal analysis for the development of packaged GaN solution UMS Products & Foundry Solutions Conclusions Slide 5 UMS GaN technologies Power by die (W) Status GH50 Power bar 50 New Development GH50 GH15 Released Released Design kit available Design kit available To be released 2017 Targeted Performances 0.50µm 0.25µm 0.15µm Power 5 W/mm 4 W/mm 3.5 W/mm 50 V 30 V 20V >200 V >120 V MTF 1e6 / 200°C 1e6 / 200°C 1e6 / 200°C Domain of frequencies Up to 7 GHz Up to 20 GHz Up to 40GHz Breakdown voltage (Vbds) GH25 MMIC GH25 Gate length Operating Vds 100 GH50 GH25 GH15 10 5 10 15 20 25 30 35 40 Target Frequency (GHz) Slide 6 Outlilne Presentation of UMS Overview of GaN technologies Thermal analysis for the development of packaged GaN solution Various Packaging solutions Thermal management of QFN oPCB / Glue / Die Coating UMS Products & Foundry Solutions Conclusions Slide 7 Packaging offer for power Main constraints for GaN packaging ‐ related to costs: ‐ Frequency band ‐ Thermal management Slide 8 Thermal management analysis Different solutions investigated: Plastic molded QFN Package solutions: Flange Ceramic metal SMD SMD QFN Ceramic metal SMD Stack variation PCB variations Interlayer stack Flange package – Mo / Diamond Tab – CuMo Coin Slide 9 Broad band packages 1: Plastic molded QFN 2: Enhanced flange package 3: Ceramic metal SMD 4: Flange package Package + PCB insertion losses (dB) Package + PCB return losses (dB) 4 2 3 1 2 3 4 1 SMD QFN package is excellent Flange packages are competitive up to 20GHz Package optimization still ongoing to achieve VSWR <1.5:1 at 20GHz Slide 10 Part of the matching can be integrated into the die Package Thermal analysis Target: Package GH50 & GH25 products C1 R1 GaN /SiC R4 Thermal flow Die attach Leadframe C2 R2 Solder R5 PCB C3 R3 Analysis must be carried out on different samples: Power‐bares MMICs Finite element simulations performed on different 3 mains package family – ANSYS Slide 11 MMIC Packaging: Modeling assumption Analysis performed on ANSYS: Symmetry : ¼ of the device is meshed GaN/SiC interface Layer with low thermal conductivity (TBR) Joule heating Block heater along drain edge of gate foot (≈ 1.5 µm) Boundary condition Fixed temperature on backside of full assembly Meshing Specific methodology for power bars ≈ 1 000 000 nodes 2nd stage Pdiss@14GHz & 34dBm ≈ 3.8 W/transistor 8 transistors 30.4W Pdiss = 3.16W/mm Test Vehicle GH25 MMIC: 1st stage : 4x8x125µm 2nd stage : 8x8x150µm 1st stage Pdiss@14GHz & 34dBm ≈ 2 W/transistor 8 transistors 16W Pdiss = 2W/mm Slide 12 Flange Package Tc = 80°C / Pdiss : 2nd stage 3.16 W/mm (30.4 W) / 1st stage 2 W/mm (16 W) Cases Flange KYO Flange KYO + Tab Mo Flange + Tab Diamond GaN / TBR SiC – 100 µm AuSn – 25 µm GaN / TBR – 2 µm SiC – 100 µm AuSn – 25 µm GaN / TBR SiC – 100 µm AuSn – 25 µm Mo – 150 µm CVD/Cu – 300 µm AuSn – 25 µm AuSn – 25 µm Stack CuMoCu – 1.4 mm CuMoCu – 1.4 mm Thermal paste – 50 µm Thermal paste – 50 µm (°C/W) Tj _max (°C) Thermal paste – 50 µm Tcase Tcase Rth_tot CuMoCu – 1.4 mm Tcase 3.75 3.84 3.24 194 196.8 178 Slide 13 Ceramic metal SMD Tc = 80°C / Pdiss : 2nd stage 3.16 W/mm (30.4 W) / 1st stage 2 W/mm (16 W) Cases SMD & PCB coin SMD & PCB vias PCB Coin Footprint GaN/TBR SiC – 100 µm AuSn – 25 µm GaN/TBR SiC – 100 µm AuSn – 25 µm CuW – 200 µm SnPb – 100 µm Stack Insulator – 1.3 mm Cu –1.3 mm CuW – 200 µm SnPb – 100 µm Cu – 18 µm Insulator – 203 µm CF3350 ‐ 50 µm Cu – 1 mm Al 6061 ‐ 3 mm Tcase Rth_tot (°C/W) Tj _max (°C) Tcase 5.06 4.69 233.8 222.7 PCB Via Footprint Slide 14 Plastic Package SMD QFN Tc = 80°C / Pdiss : 2nd stage 3.16 W/mm (30.4 W) / 1st stage 2 W/mm (16 W) Cases SMD & PCB coin SMD & PCB vias GaN/TBR GaN/TBR SiC – 100 µm Ag – 15 µm C194 – 200 µm SnPb – 100 µm Cu – 18 µm SiC – 100 µm Ag – 15 µm C194 – 200 µm SnPb – 100 µm Stack Insulator – 1.3 mm Cu –1.3 mm Insulator – 203 µm CF3350 ‐ 50 µm 2 different Ag based glues evaluated: a) 20 W.m‐1.K‐1 b) 70 W.m‐1.K‐1 Cu – 1 mm Al 6061 ‐ 3 mm Tcase Tcase Rth_tot (°C/W) Tj _max (°C) 4.97a 4.72b 4.56a 4.32b 231a 224b 219a 211b Slide 15 Synthesis Case study Packaging Remark Rth (°C/W) Bandwidth Cost 1 Flange KYO Hermetic 3.75 Up to 6 GHz 2 Flange KYO + Mo TAB Hermetic 3.84 Up to 6 GHz 3 Flange KYO + Diamond TAB Hermetic 3.25 Up to 6 GHz 4 SMD + PCB coin Hermetic 5.06 Up to 14 GHz 5 SMD + PCB via Hermetic 4.69 Up to 14 GHz 6 SMD QFN + PCB coin Die attach : 20 W.m-1.K-1 4.97 > 20 GHz Die attach : 70 W.m-1.K-1 4.72 > 20 GHz 7 SMD QFN + PCB via Die attach : 20 W.m-1.K-1 4.56 > 20 GHz Die attach : 70 W.m-1.K-1 4.32 > 20 GHz Slide 16 Approach power bar assembly Temperature Gradient 8x8x400µm / Tref = 75°C / P = 2W/mm/ CW Tj_Peak=211,9°C Slide 17 Die Attach Impact 8x8x400 / 7x7 QFN, CW, P=2W/mm, Tcase=75°C, Cond glue = 40W/m.K T(°C) delta T (°C) Rth(°C/W) GaN / TBR substrat SiC colle puce (40W/mK) Leadframe BrasureSnPb CI/glue Drain Al 217.9 196.5 178.3 162.7 150.0 126.3 96.1 21.4 18.2 15.6 12.7 23.7 30.3 21.1 0.418 0.355 0.304 0.248 0.463 0.591 0.411 contribution (%) 15.0 12.7 10.9 8.9 16.6 21.2 14.7 Total 75 142.9 2.792 100.0 8x8x400 / 7x7 QFN, CW, P=2W/mm, Tcase=75°C, Cond glue = 20W/m.K GaN / TBR substrat SiC colle puce (20W/mK) Leadframe BrasureSnPb CI/glue Drain Al T(°C) 230.9 208.9 190.5 161.6 149.5 126.2 96.1 delta T (°C) 22.0 18.4 28.9 12.1 23.3 30.1 21.1 Rth(°C/W) 0.430 0.360 0.565 0.235 0.455 0.588 0.411 contribution (%) 14.1 11.8 18.6 7.7 14.9 19.3 13.5 Total 75 155.9 3.046 100.0 Slide 18 GH50 Power Bar / Transient 8x8x400 / 7x7 QFN / Tref = 75°C / P = 2W/mm Slide 19 Conclusion on assembly SMD QFN solutions are broadband and lower cost solution Applied to GH25 and GH50 Analysis shows that thermal management strongly depends on the PCB • Coin offers very good thermal dissipation / difficult to implement to series • Thermal glue can reduce by roughly 0.2 W.m-1.K-1 As demonstrated GaN on SiC packaged in QFN can also operate in CW • Important care must be take to the functioning conditions Higher Integration MMIC $$$ Internally‐Matched Q‐MMIC Internally Matched DISCRETE General Purpose L S C X Ku Freq Slide 20 Outlilne Presentation of UMS Overview of GaN technologies Thermal analysis for the development of packaged GaN solution UMS Products & Foundry Solutions Foundry Presentation o Market addressed are not only military o Foundry partner contribution Quasi-MMIC solution development o Concept: Combination of GaN & GaAs o Cost reduce o Solutions Conclusions Slide 21 UMS Foundry Solutions Slide 22 1W High Power Amplifier 37‐ 40GHz PPH15X‐20 Power detector inside, Gain Control Advanced concept / PPH15X-20 / QFN / CHA6194-QXG 6194 Application Point to point Point to Multipoint High linearity HPA Specific features: GaAs pHEMT process Power detector dynamic 30dB Low AM/AM, AMP/PM, QFN 5x6 RF bandwidth: 37-40GHz Linear Gain: 20dB Power at 1dB comp.: 30dBm Sat. Power : 31dBm RL>13dB Consumption: 6V, 0.8 A Slide 23 1W High Power Amplifier 37‐ 40GHz PPH15X‐20 Power detector inside, Gain Control Vd=6V Idq=0.8A 34 32 Pout (dBm), Gain(dB) 30 28 26 Pout@1dB gain comp Pout@4dB gain comp Linear Gain 24 22 20 18 36 37 38 39 40 41 Freq (GHz) Slide 24 Foundry Portfolio QFN Assembly offer for all processes Slide 25 GH50 & GH25 Foundry key figures Overall repartition Nbre of wafer / year 200 150 GH50 100 Defense Telecom Space 50 GH25 0 2011 2012 2013 2014 2015 R&D 2016 GH25: Since 2010 more than 130 wafers processed in the frame of 80 different projects GH50: Production launched and transferred to 4’’ More 150 wafers to be manufactured in both GH50 & GH25 in 2016 Design Kit including EM Stack and DRC for ADS2016 Slide 26 DK available for GH25‐10 GH25: MMIC process: Qualified & Open in Foundry • • • • • • • • 250nm gate length On 4‐inch SiC wafer Frequency range: DC – 20 GHz Vds = 30V as standard Recommended Operating Value Idss = 850 mA/mm as average value Very High breakdown voltage: Vbds > 120V Power density: 4W/mm @ 10GHz in CW Mode Design Kit available for ADS2009‐2016 & MWO – – – – NL model for Hot FET (electrothermal) L FET model for noise NL model for Cold‐FET NL model for diodes (Scalable models for passives and active elements) • New features: DRC & Stack EM available for ADS2016 Slide 27 ANSYS/ADS Thermal analysis Transistor GH25 8x125 d’UMS / Vds=25V Ids=25mA/mm ADS2013 Slide 28 TNO – HPA GH25‐10 design THERMAL SIMULATION In courtesy of TNO Input stage 1x10x275um, Pdiss=8.2 W Output stage 4x10x275, Pdiss=23.3 W Matching networks, Pdiss=7.2W ( = Pdiss_total - (input+output stage) ) Tj Input stage Tj Output stage MMIC bottom Package bottom GaNS‐3 Slide 29 UMS product Solutions Slide 30 Product CHK015A‐QBA DC ‐ 6 GHz Packaged Transistor Application Radar & Communications General Purpose Transistor Specific feature Wide-band Low parasitic Plastic Package Low thermal resistance RF bandwidth: Linear Gain: Output Power: Drain Efficiency: PAE: Package: DC – 6GHz 14dB @ 6GHz > 15W > 70% 50% @ 6GHz DFN 3x4 Slide 31 25W / X-band / QFN Sampling Now In sampling / GH25 Based / Die product to be packaged V‐ Application Defence / Space In STG1 STG2 Out Main Performances RF bandwidth: 8.5 – 10.5GHz Gain: 30dB Pout : 25W PAE_associated: 45% Consumption: V+ Specific feature High efficiency High power Die / 15W version available in QFN 30V, 0.8A Slide 32 Quasi MMIC Concept Use of UMS proprietary Passive MMIC Technology: Q-MMIC is close to MMIC size Input Matched Circuit + $ + GaN Power + Bar $ $ $ + Fast design & fabrication Output Matched Circuit 50Ω 50Ω $ Fast design & fabrication ☺ Customize your Q‐MMIC ☺ Slide 33 Why “Quasi MMIC”? • Integration – Close to MMIC size Example of 50W C‐ band HPA 6mm² • Cost • Flexibility – Short dev. cycle times (passive MMICs) GH50/25: 14Weeks ICT URLC: 5 Weeks 7.6mm² 6 MMIC / Quasi‐MMIC Cost – Close to hybrid solutions 3.6mm² 5 4 3 2 1 10 15 20 MMIC Size (mm²) Applicable to QFN for low cost development 25 Slide 34 New Product 100W / L‐band / QFN In development / GH50 based / Internally‐Matched / Q‐MMIC Application Radar / Dual Use Main Performances Specific feature @ 1.3GHz Peak Pout=110W With PAE=57% Gain = 14dB RF bandwidth: 1.2 – 1.4GHz Linear Gain: 15dB Output Power: 100W Gain @ 50W 10dB PAE : 55% Package: DFN 7x7 “High Performance Plastic Packaged 100W L‐Band Quasi‐MMIC HPA” D. Bouw et al. ‐ EuMC06‐05 50W / C‐band / QFN Slide 35 New Product In development / GH25 based / Internally‐Matched / Q‐MMIC Application Radar & Communications Main Performances Specific feature High PAE Low parasitic Power Plastic Package Low thermal resistance RF bandwidth: 5.2 – 5.9GHz 5.9-6.9GHz Linear Gain: 14dB 13dB Output Power: 50W 50W Gain @ 50W 10dB 9dB PAE : 47% 45% Package: DFN 7x7 DFN 7x7 Slide 36 C‐band DPA Description Advanced Development Symmetric DPA Architecture / GH25 based / Q‐MMIC / QFN packaging Application Communications / 5G Main Objectives Configuration 1 stage => for investigations purpose Single & dual input Linearity / Main & Peak synchronization … Key Performance Frequency band : 5.6 – 6.6GHz Peak Output Power: 15W PAE @ 6dB OBO: > 35% Slide 37 C-band DPA Characterizations PAE vs Freq / 6 & 10dB OBO / 4 Boards / T=25°C Biasing conditions: Vd = 30V Id_q_main = 60mA / Vgs0_peak = -7V 6dB OBO 10dB OBO Slide 38 Outline Presentation of UMS Overview of GaN technologies Thermal analysis for the development of packaged GaN solution UMS Products & Foundry Solutions Conclusions Slide 39 Conclusions QFN Solution validated for power technologies up to 40GHz – including GaN MMIC in QFN for GH25 validated with enhanced glue + PCB solutions • High efficiency solutions already available Q‐MMIC in QFN used for GH50 & GH25 • Challenging solutions evaluated Release of PPH15X‐20 for applications up to 40GHz Foundry access allowing more and more QFN assembly Tools available for ADS allow more accurate thermal analysis Association with URLC (passive process) available in Foundry Indicators allow to identify the cost decrease (Industrial Manufacturing cycle time reducing, …) Final passivation is under development and should be available to improve the robustness versus humidity for GH50/GH25/URLC Slide 40 Aknowledgment • TNO for their participation • Colleagues for their support & works – – – – – – M. Camiade P‐F. Alleaume L. Brunel M. Feron E. Leclerc J‐P. Viaud Slide 41

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

Download PDF

advertising