I>ataQLiest

I>ataQLiest

I>ataQLiest

a company of

(408) 437-8000 Telex 171973 Fax (408) 437-0292

December 1992

Dear Client:

Enclosed you will find a new 1992 binder for your Memories Worldwide Service.

Please move the 1992 Market Statistics, Vendor Profiles, and Dataquest Perspectives into the new binder behind the appropriate tab. Please keep the remaining 1991 documents in the existing binder for historical purposes.

Attached you will find a checklist of what should be in the new 1992 binder after the documents have been filed.

You will soon be receiving your 1993 Dataquest binders for storage of all 1993 deliverables.

If you have any questions please contact me at (408) 437-8320.

Kimberlle C. Southern

Operations Coordinator

mi.

MEMORIES WORLDWIDE

Binder Checklist

SECTION TITLE

Title Page

Disclaimer

Market Trends fTab)

Market Statistics fTab)

Worldwide MOS Memory Forecast

MOS Memory Unit Shipments 90-91

Worldwode MOS Memory Market Share 89-91

Vendor Profiles (Tab)

MMRY-SEG-VP-9201 (Miaon Technology)

Perspectives fTab)

MMRY-SEG-DP-9203

MMRY-SEG-DP-9202

MMRY-SEG-DP-9201

COPYRIGHT MISSING

1991

08/31/92

08/24/92 ^

on mm

^

(^i\mi ^

Qi%l2AI92 ^^

04/06/92 '-^

03/30/92 ^

Dataquest

Dataquest Incorporated

1290 Ridder Park Drive

San Jose, CA 95131-2398

(408) 437-8000

Telex: 171973

Fax: (408) 437-0292

United Kingdom

Dataquest UK Limited

Roussel House,

Broadwater Park

Denham, Nr Uxbrldge,

Middx UB9 5HP

England

0895-835050

Telex: 266195

Fax 0895 835260-1-2

Japan

Dataquest Japan Limited

Shinkawa Sanko Buildiitg

2 PI

1-3-17 Shinkawa

Chuo-kuTokyo 104

Japan

011-81-3-5566-0411

Telex: 781-32768

Fax: 011-81-3-5566-0425

France

Dataquest Europe SA

Tour Galli6ni 2

36, avenue du G6n6ral-de-Gaulle

93175 Bagnolet Cedex

FraiKc

(1)48 97 31 00

Telei: 233 263

Fax: (01)48 97 34 00

Korea

Dataquest Korea

Dacheung Building Room 1105

648-23 Yoiksam-dong

Kangnam-gu, Seoul 135-80

Korea

011-82-2-552-2332

Fax: 011-82-2-552-2661

Germany

Dataquest GmbH

Kronstadter Strasse 9

8000 Munich 80

1!esx Germany

Oil 49 89 93 09 09 0

Fax: Oil 49 89 930 3277

Dataquest Incorporated

Ledgeway/Dataquest

The Corpcnate Center

550 Cochituate Road

Pramingham, MA 01701

(508) 370-5555

Fax: (508) 370-6262

The content of this' report represents our interpretation and analysis of infonnation generally available to the public or released by responsible individuals in the s u ^ e a companies, but is not guaranteed as to accuracy or cosq>leteness. It does not contain material provided to us in confidence by our clients.

This information is DM furnished in connection with a sale or ofifier to sell securities, or in connection with the solicitation of an offer to buy securities. This firm and its parent and/or tlwir officers, stockholders, or members of the families may, ftom time to time, have long or short position in the securities mentioned and may sell or buy such securities.

Printed in the United States of America. All rights reserved. No part of this publication may be reproduced, stored in retrieval systems, or transmitted, in any form or by any means—mechanical, electronk:, photocopyiiig, duplicatiiig, microfilming, videotape, or otherwise—without the prior permission of the publisher.

01991 Dataquest Incorporated

#

1/28/92

Worldwide DRAM Consumption by Application

(Billions of Bit)

1988

1989

1990 1991 1992

1993

DRAM Production

Merchant

Captive (est.)

482,028 706,912 1,053,609 1,599,111 2,578,920 3,948,151

0 70,943 154,262 237,884 325,365 450,284

Total

482,028 777,855 1,207,871 1,836,995 2,904,285 4,398,435

Consumption by Computers

42,280

60,211 90,317 116,326

10,496

21,960

31,744 47,520 68,608

531,086

75.1%

154,240

61,792

337,184

PC Total 178,118

425

1,818

Handheld 0

134

2,511

32,883

18,392

299,972

213,203

Desktop 171,955

81,877

PC Add-in Memory 42,050

200,800

132,000

471,204

7,188

8,781

50,155

267,776

254,144

349,760

447,816

732,447 1,239,624

17,554 74,832

20,750 50,784

90,984 147,096

603,159 966,912

405,080

179,523 393,612 571,314

267,113

807,475 1,185,632 1,818,672 2,849,328 4

76.6% 74.1% 70.5% 72.2%

% by Computers 77.6%

22,560 32,856 55,880

Consumption by Application

531,086 807,475 1,185,632 1,818,672 2,849,328 4

Computers 374,Oil n/a 130,544 224,100 403,380 629,273

Other DP n/a n/a 115,386 189,379 356,868 469,550

Others n/a

531,086 1,053,405 1,599,111 2,578,920 3,948,151

Total 374,011

Source: Dataquest (January 1992)

NotRiinoie

Worldwide MOS Memory

Market Share, 1989-1991

July 6, 1992

DataQuest

Memories Worldwide

MMRY-SEG-MS-9201

• i IJ UM}

Dataques

Market Statistics

Worldwide MOS Memory

Marlcet Sliare

July 6, 1992

i

^ DataQuest'

Source:

Dataquest

Marlcet Statistics

File behind the Market Statistics tab inside the binder labeled Memories Worldwide

Published by Dataquest Incorporated

The content of this report represents our interpretation and analysis of information generally available to the public or released by knowledgeable individuals in the subject industry, but is not guaranteed as to accuracy or completeness. It does not contain material provided to us in confidence by our clients.

Printed in the United States of America. All rights reserved. No part of this publication may be reproduced, stored in retrieval systems, or transmitted, in any form or by any means—mechanical, electronic, photocopying, duplicating, microfilming, videotape, or otherwise—without die prior permission of the publisher.

© 1992 Dataquest Incorporated

July 1992

i

Worldwide MOS Memory Market Share, 1989-1991

Table of Contents

P ^ e

Introduction 1

Segmentation 1

Definitions 1

Product Definitions 1

Regional Definitions 2

Line Item Definitions 2

Market Share Methodology 2

Notes on Market Share 3

Notes to Market Share Tables 3

Exchange Rates 3

Table Page

1-1 Each Company's Factory Revenue from Shipments of MOS Memory to the World

(Millions of U.S. Dollars) 4

1-2 Each Company's Factory Revenue from Shipments of MOS DRAMs to the World

(Millions of U.S. Dollars) 6

1-3 Each Company's Factory Revenue from Shipments of MOS SRAMs to the World

(Millions of U.S. Dollars) 7

1-4 Each Company's Factory Revenue from Shipments of MOS EPROMs to the World

(Millions of U.S. DoUars) 9

1-5 Each Company's Factory Revenue from Shipments of MOS Nonvolatile Memory to the

World (Millions of U.S. Dollars) 10

1-6 Each Company's Faaory Revenue from Shipments of Other MOS Memory to the World

(Millions of U.S. Dollars) 12

1-7 Each Company's Factory Revenue from Shipments of Bipolar Memory to the World

(Millions of U.S. Dollars) 13

2-1 Top 20 Companies' Faaory Revenue from Shipments of MOS Memory to the World

(Millions of U.S. Dollars) 14

2-2 Top 10 Companies' Faaory Revenue from Shipments of MOS DRAMs to the World

(Millions of U.S. Dollars) 15

2-3 Top 10 Companies' Faaory Revenue from Shipments of MOS SRAMs to the World

(Millions of U.S. Dollars) 16

2-4 Top 10 Companies' Faaory Revenue from Shipments of MOS EPROMs to the World

(Millions of U.S. Dollars) 17

2-5 Top 10 Companies' Faaory Revenue from Shipments of MOS Nonvolatile Memory to the

World (Millions of U.S. Dollars) 18

2-6 Top 10 Companies' Faaory Revenue from Shipments of Other MOS Memory to the

World (Millions of U.S. Dollars) 19

2-7 Top 10 Companies' Faaory Revenue from Shipments of Bipolar Memory to the World

(Millions of U.S. Dollars) 20

Note: All tables show estimated data.

Table Page

3-1 Each Company's Factory Revenue from Shipments of MOS Memory to North America

(Millions of U.S. DoUars) 21

3-2 Each Company's Factory Revenue from Shipments of MOS DRAMs to North America

(Millions of U.S. Dollars) 23

3-3 Each Company's Factory Revenue from Shipments of MOS SRAMs to North America

(Millions of U.S. Dollars) 24

3-4 Each Company's Faaory Revenue from Shipments of EPROMs to North America

(Millions of U.S. Dollars) 26

3-5 Each Company's Faaory Revenue from Shipments of MOS Nonvolatile Memory to

North America (Millions of U.S. Dollars) 27

3-6 Each Company's Factory Revenue from Shipments of Other MOS Memory to North

America (Millions of U.S. Dollars) 29

3-7 Each Company's Factory Revenue from Shipments of Bipolar Memory to North America

(Millions of U.S. Dollars) 30

4-1 Each Company's Faaory Revenue from Shipments of MOS Memory to Japan

(Millions of U.S. Dollars) 31

4-2 Each Company's Faaory Revenue from Shipments of MOS DRAMs to Japan

(Millions of U.S. Dollars) 33

4-3 Each Company's Faaory Revenue from Shipments of MOS SRAMs to Japan

(Millions of U.S. Dollars) 34

4-4 Each Company's Faaory Revenue from Shipments of MOS EPROMs to Japan

(Millions of U.S. Dollars) 36

4-5 Each Company's Faaory Revenue from Shipments of MOS Nonvolatile Memory to Japan

(Millions of U.S. Dollars) 37

4-6 Each Company's Faaory Revenue from Shipments of Other MOS Memory to Japan

(Millions of U.S. Dollars) 39

4-7 Each Company's Faaory Revenue from Shipments of Bipolar Memory to Japan

(Millions of U.S. Dollars) 40

5-1 Each Company's Faaory Revenue from Shipments of MOS Memory to Europe

(Millions of U.S. Dollars) 41

5-2 Each Company's Faaory Revenue from Shipments of MOS DRAMs to Europe

(Millions of U.S. Dollars) 43

5-3 Each Company's Faaory Revenue from Shipments of MOS SRAMs to Europe

(Millions of U.S. Dollars) 44

5-4 Each Company's Faaory Revenue from Shipments of MOS EPROMs to Europe

(Millions of U.S. DoUars) 46

5-5 Each Company's Faaory Revenue from Shipments of MOS Nonvolatile Memory to

Europe (Millions of U.S. Dollars) 47

5-6 Each Company's Faaory Revenue from Shipments of Other MOS Memory to Europe

(Millions of U.S. Dollars) 49

5-7 Each Company's Faaory Revenue from Shipments of Bipolar Memory to Europe

(Millions of U.S. Dollars) 50

6-1 Each Company's Faaory Revenue from Shipments of MOS Memory to Asia/Pacific-Rest of World (Millions of U.S. DoUars) 51

6-2 Each Company's Faaory Revenue from Shipments of MOS DRAMs to Asia/Pacific-Rest of World (MUlions of U.S. DoUars) 53

6-3 Each Company's Faaory Revenue from Shipments of MOS SRAMs to Asia/Pacific-Rest of World (MiUions of U.S. DoUars) 54

Note: All tables show estimated data.

I

Table Page

6-4 Each Company's Factory Revenue from Shipments of MOS EPROMs to Asia/Pacific-Rest of

World (Millions of U.S. Dollars) 56

6-5 Each Company's Factory Revenue from Shipments of MOS Nonvolatile Memory to

Asia/Pacific-Rest of World (Millions of U.S. Dollars) 57

6-6 Each Company's Factory Revenue from Shipments of Other MOS Memory to Asia/Pacific-

Rest of World (Millions of U.S. Dollars) 59

6-7 Each Company's Factory Revenue from Shipments of Bipolar Memory to Asia/Pacific-Rest of World (Millions of U.S. Dollars) 60

Note: All tables show estimiited data.

I

I

m

\Vorldwide MOS Memory

Market Share, 1989-1991

Introduction

This document contains detailed information on Dataquest's view of the MOS memory IC market. Included in this document are:

• 1989-1991 market share estimates

Analyses of the MOS memory market by company provide insight into high-technology markets and reinforce estimates of consumption, production, and company revenue.

Worldwide market share estimates combine data from many countries, each of which has a different and fluctuating exchange rate. Estimates of non-U.S. market consumption or revenue are based upon the average exchange rate for the given year. Refer to the section entitled "Exchange Rates" for more information regarding these average rates. As a rule. Dataquest's estimates are calculated in local currencies and then converted to U.S. dollars.

More detailed data on this market may be requested through Dataquest's client inquiry service. Qualitative analysis of these data is provided in the Dataqiiest Perspectives located in the binder of the same name.

Segmentation

This section outlines the market segments that are specific to this document. Dataquest's objective is to provide data along lines of segmentation that are logical, appropriate to the industry in question, and immediately useful to clients.

For a detailed explanation of Dataquest's market segmentation, refer to the Dataquest

Research and Forecast Methodology document located in the Source: Dataquest binder. For a complete listing of all market segments tracked by Dataquest, please refer to the Dataquest

High-Technology Guide: Segmentation and

Glossary.

Dataquest defines the MOS memory market as

DRAM, SRAM, EPROM, mask ROM, EEPROM, flash memory, and other MOS memory. MOS memory is defined as a MOS IC in which binary data are stored and electronically retrieved.

Merchant versus Captive Consumption: Dataquest includes all revenue, both merchant and captive, for semiconductor suppliers selling to the merchant market. The data exclude completely captive suppliers where devices are manufactured solely for the company's own use. A product that is used internally is valued at market price rather than at transfer or factory price.

Definitions

This section lists the definitions that are used by Dataquest to present the data in this document. Complete definitions for all terms associated with Dataquest's segmentation of the high-technology marketplace can be found in the Dataquest High-Technology Guide: Segmen-

tation and Glossary.

Product Definitions

DRAM: Includes dynamic RAM, multiport-DRAM

(M-DRAM), and video-DRAM (V-DRAM).

DRAMs have memory cells consisting of a single transistor, and require regular externally cyded memory cell refreshes. 'iTiese are volatile memories and addressing is multiplexed.

SRAM: Includes static RAM, multiport-SRAM

(M-SRAM), battery backed-up SRAM

(BB-SRAM), and pseudo-SRAM (P-SRAM).

SRAMs have memory cells consisting of a minimum of four transistors, except a P-SRAM, which has a memory cell consisting of a single transistor and is similar to a DRAM.

SRAMs do not require externally cyded memory cell refreshes. These are volatile memories and addressing is not multiplexed

(except in the case of P-SRAM).

EPROM: Includes erasable programmable readonly memory. Induded are ultraviolet EPROM

(UV EPROM) and one-time programmable read-only memory (OTPROM). EPROMs have

Memories Worldwide memory cells consisting of a single transistor, and do not require any memory cell refreshes.

These devices are nonvolatile memories.

Nonvolatile MOS Memory IC: Includes EPROM, mask ROM, EEPROM, and flash. Dataquest defines the mask ROM market as maskprogrammable read-only memory. Mask ROM is a form of memory that is programmed by the manufacturer to a user specification using a mask step. Mask ROM is programmed in hardware rather than software. These devices are considered nonvolatile memories. Dataquest defines the EEPROM market as electronically erasable programmable read-only memory. This market includes serial EEPROM (S-EEPROM), parallel EEPROM (P-EEPROM), and electronically alterable read-only memory (EAROM).

EEPROMs have memory cells consisting of a minimum of two transistors, and do not require memory cell refreshes. This market also includes nonvolatile RAM (NV-RAM), also known as shadow RAM. These semiconductor products are a combination of SRAM and

EEPROM technologies in each memory cell.

The EEPROM functions as a shadow backup for the SRAM when power is lost. Dataquest defines the flash market as a nonvolatile p r o d u a designated as flash EPROM/EEPROM fliat incorporates either 5V or 12V programming supplies and one-transistor (IT) or twotransistor (2T) memory cells with electrical programming and fast bulk/chip erase.

Regional Definitions

North America: Includes United States and

Canada

United States: Includes 48 contiguous states,

Washington, D.C., Alaska, Hawaii, and Puerto

Rico

Europe: Western Europe

Japan: Japan

Asia/Pacific-Rest of World: All other countries

l i n e Item Definitions

Factory revenue is defined as the amount of money received by a semiconductor vendor for its goods; revenue fi-om the sale of semiconductors sold either as finished goods, die, or wafers to another semiconductor vendor for resale is attributed to the semiconductor vendor who sells the product to a distributor or equipment manufacturer.

Other MOS Memory IC: Includes all other

MOS memory not already accounted for in the preceding categories. This category includes

MOS content addressable memory (CAM), MOS cache-tag RAM, MOS first-in/first-out memory

(FIFO), MOS last-in/first-out memory (UFO), and ferroelectroic memory.

Bipolar Memory: Includes bipolar digital semiconductor products in which binary data are stored and electronically retrieved. Included are

ECL or TTL random access memory (RAM), read-only memory (ROM), programmable ROM

(PROM), last-in/first-out (UFO) memory, and first-in/first-out (FIFO) memory; not included are products made with ixiixed bipolar CMOS

(that is, BiCMOS) with TTL or ECL outputs, which are classified as MOS.

Market Share Methodology

Dataquest utilizes both primary and secondary sources to produce market statistics data. In the fourth quarter of each year, Dataquest surveys all major participants within each industry. Selected companies are resurveyed during the first quarter of the following year to verify final annual results. This primary research is supplemented with additional primary research and secondary research to verify market size, shipment totals, and pricing information. Other sources of data utilized by Dataquest include:

• Information published by major industry participants

• Estimates made by knowledgeable and reliable industry spokespersons

Government data or trade association data

Published product literature and price lists

Interviews with knowledgeable manufacturers, distributors, and users

Relevant economic data

Information and data from online and/or

CD-ROM data banks

©1992 Dataquest Incorporated July-r-Reproduction Prohibited.

• End-user surveys

Worldwide MOS Memory Market Share, 1989-1991

• Articles in both the general and trade press

• Reports from financial analysts

Dataquest believes that the estimates presented in this document are the most accurate and meaningful statistics available.

Despite the care taken in gathering, analyzing, and categorizing the data in a meaningful way, careful attention must be paid to the definitions and assumptions used herein when interpreting the estimates presented in this document. Various companies, government agencies, and trade associations may use slightly different definitions of produa categories and regional groupings, or they may include different companies in their summaries. These differences should be kept in mind when making comparisons between data and numbers provided by Dataquest and those provided by other suppliers.

2. Gould AMI revenue from 1991 forward does not include foundry revenue.

3. Harris revenue includes GE Solid State revenue from 1989 forward.

4. Inmos revenue is included in SGS-

Thomson revenue from 1989 forward.

5. Macronix revenue is included under

Asia/Pacific Companies from 1991 forward.

6. Other North American Companies and

Other Asia/Pacific Companies revenue has been restated to reflect the fewer number of companies published in 1991.

7. Philips revenue includes Signetics revenue.

Exchange Rates

Dataquest uses an average annual exchange rate in converting revenue to U.S. dollar amounts. The following table outlines these rates for 1989 through 1991.

Notes on Market Share

In the process of conducting data collection and preparing market statistics information,

Dataquest will sometimes consolidate or revise the numbers of a particular company, model, series, or industry. In this section, any such changes contained within this document are outlined for your reference.

Japan CYenAJ.S.$)

France (FrancAJ.S.$)

Gennany (Deutsche

Mark/U.S.$)

United Kingdom

(U.S.$/Pound Sterling)

1989 1990 1991

138 144 136

6.39 5.44 5.64

1.88

1.62 1.66

1.50 1.79

1.77

Notes to Market Share Tables

1. GEC Plessey revenue includes MEDL and

Plessey revenue from 1990 forward.

©1992 Dataquest Incorporated July—^Reproduction Prohibited

Memories Worldwide

Table 1-1

Each Company's Factory Revenue from Shipments of MOS Memory to the World

(Millions of U.S. Dollars)

Total Market

1989

15,405

R e v e n u e

1 9 9 0

12,128

1 9 9 1

12,841

1 9 8 9

100.0

Market S h a r e ( % )

1 9 9 0

100.0

1 9 9 1

100.0

North American Companies

Advanced Micro Devices

AT&T

Atmel

Catalyst

Cypress Semiconductor

Dallas S e m i c o n d u a o r

Gould AMI

Harris

Honeywell

Integrated Device Technology

Intel

Int'l Microelectronic Prod.

ITT

Microchip Technology

Micron Technology

MOSel

Motorola

NCR

National Semiconductor

Performance S e m i c o n d u a o r

SEEQ Technology

Texas Instruments

Vitelic

VLSI Technology

WaferScale Integration

Xicor

Other North American Companies

3,651

258

433

17

10

94

25

37

2

158

1,095

66

23

28

87

50

13

47

31

149

10

395

20

407

8

132

16

40

2,977

253

13

54

35

166

14

14

24

2

132

371

8

0

60

286

31

395

4

137

19

33

741

64

8

27

68

18

3,298

270

4

78

32

186

21

11

23

0

128

395

6

10

57

455

75

412

3

112

18

33

738

85

0

23

91

32

24.5

2.1

.1

.4

.3

1.4

.1

.1

.2

.0

1.1

.3

3.3

.0

1.1

.2

.3

6.1

3.1

.1

.0

.5

2.4

.5

.1

.2

.6

.1

3.5

.6

3.2

.0

.9

.1

.3

5.7

.7

.0

.2

.7

.2

25.7

2.1

.0

.6

.2

1.4

.2

.1

.2

.0

1.0

3.1

.0

.1

.4

J a p a n e s e Companies

Fujitsu

Hitachi

Matsushita

Mitsubishi

NEC

NMB Semiconductor

Oki

Ricoh

9,678

1,188

1,396

362

1,117

1,594

127

441

28

7.095

913

1,224

265

745

1,233

96

350

26

7,141

909

1,330

217

762

1,242

60

380

8

2.8

.1

.1

.6

2.6

.1

2.6

.1

.9

.1

1.0

.1

.2

.2

.0

1.0

.3

7.1

.4

.1

.2

.6

.3

23.7

1.7

.1

.3

.2

62.8

7.7

9.1

2.3

7.3

10.3

.8

2.9

.2

58.5

7.5

10.1

2.2

6.1

10.2

.8

2.9

.2

55.6

7.1

10.4

1.7

5.9

9.7

.5

3.0

.1

CContinued)

©1992 Dataquest Incorf)orated July—Reproduction Prohibited

Worldwide MOS Memory Market Share, 1989-1991

Table 1-1 (Continued)

Each Company's Factory Revenue from Shipments of MOS Memory to the World

(Millions of U.S. Dollars)

R o h m

Sanyo

Seiko Epson

Sharp

Sony

Toshiba

Yamaha

O t h e r J a p a n e s e Companies

1989

5

118

137

434

215

1,681

0

835

R e v e n u e

1 9 9 0

13

86

55

454

204

1,431

0

0

1 9 9 1

28

82

37

476

183

1,425

2

0

1 9 8 9

.0

.8

Market S h a r e (%)

1 9 9 0

.1

1 9 9 1

.2

.7

.6

.9

2.8

1.4

10.9

.0

5.4

.5

3.7

1.7

11.8

.0

.0

.3

3.7

1.4

11.1

.0

.0

Eurojjean Companies

Eurosil

GEC Plessey

Matra MHS

MEDL

Philips

Plessey

SGS-Thomson

Siemens

Asia/Pacific Companies

Goldstar

Hualon Microelectronics Corp.

Hyundai

Macronjx

Samsung

Silicon Integrated Systems

United Microelectronics

W i n b o n d Electronics

NA - Not available

NM - Not meaningful

Source: Dataquest (July 1992)

716

0

0

31

7

60

3

239

376

1,360

8 2

NA

210

31

935

NA

102

NA

731

0

8

37

0

96

0

278

312

1,325

96

39

115

7

971

17

66

14

682

1

0

35

0

75

0

273

298

1,720

249

27

248

31

1,066

15

58

26

.5

NA

1.4

.2

4.6

.0

.0

.2

.0

.4

.0

1.6

2.4

8.8

6.1

NA

.7

NA

6.0

.0

.1

.3

.0

.8

.0

2.3

2.6

10.9

.8

.3

.9

.1

8.0

.1

.5

.1

5.3

.0

.0

.3

.0

.6

.0

2.1

2.3

13.4

1.9

.2

1.9

.2

8.3

.1

.5

.2

©1992 Dataquest Incorporated July—^Reproduction Prohibited

Memories Worldwide

Table 1-2

Each Company's Factory Revenue from Shipments of MOS DRAMs to the World

(Millions of U.S. Dollars)

Total Market

1989

9,104

Revenue

1990

6,525

1991

6,982

1989

100.0

Market Share (%)

1990

100.0

1991

100.0

North American Companies

Intel

Micron Technology

MOSel

Motorola

Texas Instruments

Vitelic

Japanese Companies

Fujitsu

Hitachi

Matsushita

Mitsubishi

NEC

NMB Semiconductor

Oki

Sanyo

Sharp

Sony

Toshiba

Other Japanese Con^anies

1,705

70

355

0

320

899

61

6,012

748

757

241

729

1,052

127

390

14

100

0

1,268

586

1,235

88

213

0

292

584

58

3,991

536

617

168

466

754

96

305

18

67

3

961

0

1,384

69

365

20

276

575

79

4,011

503

661

132

515

743

60

346

31

60

3

957

0

18.7

0.8

3.9

0.0

3.5

9.9

0.7

66.0

8.2

8.3

2.6

8.0

11.6

1.4

4.3

0.2

1.1

0.0

13.9

6.4

18.9

1.3

3-3

0.0

4.5

9.0

0.9

61.2

8.2

9.5

2.6

7.1

11.6

1.5

4.7

0.3

1.0

0.0

14.7

0.0

57.4

7.2

9.5

1.9

7.4

10,6

0.9

5.0

0.4

0.9

0.0

13.7

0.0

19.8

1.0

5.2

0.3

4.0

8.2

1.1

Eurojjean Companies

Siemens

361

361

298

298

287

287

4.0

4.0

4.6

4.6

4.1

4.1

Asia/Pacific Comfjanies

Goldstar

Hyundai

Samsung

NA - Not available

NM - Not meaningful

Source: Dataquest (July 1992)

1,026

61

160

805

1,001

85

77

839

1,300

228

186

886

11.3

0.7

1.8

8.8

15.3

1.3

1.2

12.9

18.6

3.3

2.7

12.7

©1992 Dataque$t Incorporated July—Reproduction Prohibited

Worldwide MOS Memory Market Share, 1989-1991

Table 1-3

Each Company's Factory Revenue from Shipments of MOS SRAMs to the World

(Millions of U.S. Dollars)

Total Market

1989

3,171

Revenue

1990

2,584

1991

2,576

1989

100.0

Market Share (%)

1990

100.0

1991

100.0

North American Companies

Advanced Micro Devices

AT&T

Atmel

Catalyst

Cypress Semiconductor

Harris

Honeywell

Integrated Device Technology

Intel

Micron Technology

MOSel

Motorola

NCR

National Semiconduaor

Performance Semiconductor

Texas Instmments

Vitelic

VLSI Technology

Other North American Companies

Japanese Companies

Fujitsu

Hitachi

Matsushita

Mitsubishi

NEC

Oki

Rohm

Sanyo

Seiko Epson

Sharp

Sony

Toshiba

Other Japanese Companies

536

52

5

1

7

106

31

2

107

1

40

17

80

1

26

16

2

5

7

30

2,246

276

429

25

247

263

23

5

82

137

76

208

226

249

524

33

5

1

4

124

18

2

71

15

73

12

99

1

22

19

2

6

7

10

1,756

237

406

22

183

237

20

7

55

55

94

193

247

0

551

16

4

1

0

125

17

0

45

15

90

36

132

2

15

18

4

6

0

25

1,742

261

449

20

151

241

14

9

37

37

109

172

242

0

16.9

1.6

0.2

0.0

0.2

3.3

1.0

0.1

3.4

0.0

0.5

0.1

0.2

0.2

0.9

70.8

8.7

13.5

0.8

7.8

8.3

0.7

0.2

2.6

4.3

2.4

6.6

7.1

7.9

1.3

0.5

2.5

0.0

0.8

0.7

0.1

2.7

0.6

2.8

0.5

3.8

0.0

20.3

1.3

0.2

0.0

0.2

4.8

0.9

0.7

0.1

0.2

0.3

0.4

68.0

9.2

15.7

0.9

7.1

9.2

0.8

0.3

2.1

2.1

3.6

7.5

9.6

0.0

67.6

10.1

17.4

0.8

5.9

9.4

0.5

0.3

1.4

1.4

4.2

6.7

9.4

0.0

(ContinuecO

5.1

0.1

0.6

0.7

0.2

0.2

0.0

1.0

4.9

0.7

0.0

1.7

0.6

3.5

1.4

21.4

0.6

0.2

0.0

0.0

©1992 Dataquest Incorporated July—^Reproduction Prohibited

Memories Worldwide

Table 1-3 (Continued)

Each Company's Factory Revenue £rom Shipments of MOS SRAMs to the World

(MlUions of U.S. Dollars)

European Companies

GEC Plessey

Matra MHS

MEDL

Philips

SGS-Thomson

1989

129

0

31

7

3

88

Revenue

1990

98

4

37

0

8

49

1991

82

0

35

0

9

38

1989

4.1

Market Share ("/o)

1990

1991

3.8

3.2

0.0

0.2

0.0

1.0

0.2

1.4

0.0

1.4

0.0

0.1

2.8

0.3

1.9

0.3

1.5

Asia/Pacific Companies

Goldstar

Huaion Microelectronics Corp.

Hyundai

Samsung

Silicon Integrated Systems

United Microelectronics

Winbond Electronics

NA * Not available

NM •> Not meaningful

Source: Dauquest Ouly 1992)

260

11

NA

49

100

NA

100

NA

206

6

10

30

92

2

64

2

201

17

10

48

93

2

22

9

8.2

0.3

NA

1.5

3.2

NA.

3.2

NA

8.0

0.2

0.4

1.2

3.6

0.1

2.5

0.1

7.8

0.7

0.4

1.9

3.6

0.1

0.9

0.3

©1992 Dataquest Incorporated July-HReproduction Prohibited

Worldwide MOS Memory Market Share, 1989-1991

Table 1-4

Each Company's Factory Revenue from Shipments of MOS EPROMs to the World

(Millions of U.S. Dollars)

Total IVIarket

1 9 8 9

NA

R e v e n u e

1 9 9 0

NA

1 9 9 1

1,358

1989

NA

Market S h a r e (%]

1 9 9 0

NA

1 9 9 1

100.0

North American Companies

Advanced Micro Devices

Atmel

Catalyst

Cypress S e m i c o n d u a o r

Intel

Microchip Technology

National Semiconductor

Texas Instruments

WaferScale Integration

Other North American Companies

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

774

225

30

2

37

205

33

81

136

23

2

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

57.0

16.6

2.2

0.1

2.7

15.1

2.4

6.0

10.0

1.7

0.1

J a p a n e s e Companies

Fujitsu

Hitachi

Mitsubishi

NEC

O k i

Sharp

Toshiba

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

367

86

59

67

81

3

3

68

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

27.0

6.3

4.3

4.9

6.0

0.2

0.2

5.0

European Comp)anies

Philips

SGS-Thomson

NA. " Not available

NM « Not meaningful

Source: Dataquest (July 1992)

NA

NA

NA

NA

NA '

NA

217

59

158

NA

NA

NA

NA

NA

NA

16.0

4.3

11.6

©1992 Dataquest Incorporated July—^Reproduction Prohibited

10

Memories 'Worldwide

Table 1-5

Each Company's Factory Revenue from Shipments of MOS Nonvolatile Memory to the World

(Millions of U.S. Dollars)

Total Market

1989

3,013

Revenue

1990

2,845

1991

3,071

1989

100.0

Market Share (%)

1990

100.0

1991

100.0

North American Companies

Advanced Micro Devices

AT&T

Atmel

Catalyst

Cypress Semiconduaor

Gould AMI

Harris

Intel

Int'l Microelectronic Prod.

ITT

Microchip Technology

MOSel

Motorola

National Semiconduaor

SEEQ Technology

Texas Instruments

VLSI Technology

WaferScale Integration

Xicor

Other North American Companies

1,309

203

8

46

24

22

25

1

362

17

10

94

0

7

106

40

194

16

28

87

19

1,090

209

8

53

31

21

14

1

268

8

0

60

6

4

115

33

155

1

27

68

8

1,195

237

0

77

32

37

11

0

311

6

10

57

8

4

97

33

159

0

23

91

2

43-4

6.7

0.3

1.5

0.8

0.7

0.8

0.0

12.0

0.6

0.3

3.1

0.0

0.2

3.5

1.3

6.4

0.5

0.9

2.9

0.6

38.9

7.7

0.0

2.5

1.0

1.2

0.4

0.0

10.1

0.2

0.3

1.9

0.3

0.1

3.2

1.1

5.2

0.0

0.7

3.0

0.1

0.3

0.0

2.1

0.2

0.1

4.0

1.2

5.4

0.0

0.9

2.4

0.3

38.3

7.3

0.3

1.9

1.1

0.7

0.5

0.0

9.4

Japanese Companies

Fujitsu

Hitachi

Matsushita

Mitsubishi

NEC

Oki

Ricoh

Rohm

Sanyo

Sharp

Sony

Toshiba

Yamaha

1,419

164

210

96

141

279

28

28

0

22

257

7

187

0

1,347

140

201

75

96

242

25

26

6

13

292

8

223

0

1,382

145

220

65

96

258

20

8

19

9

306

8

226

2

47.1

5.4

7.0

3.2

4.7

9.3

0.9

0.9

0.0

0.7

8.5

0.2

6.2

0.0

47.3

4.9

7.1

2.6

3.4

8.5

0.9

0.9

0.2

0.5

10.3

0.3

7.8

0.0

45.0

4.7

7.2

2.1

3.1

8.4

0.7

0.3

0.6

0.3

10.0

0.3

7.4

0.1

(Continued)

©1992 Dataquest Incorporated July—Reproduction Prohibited

World^Hde MOS Memory Market Share, 1989-1991

11

Table 1-5 (Continued)

Each Company's Factory Revenue from Shipments of MOS Nonvolatile Memory to the World

(Millions of U.S. Dollars)

European Companies

Eurosil

GEO Plessey

Philips

Plessey

SGS-Thomson

1989

211

0

0

57

3

151

Revenue

1990

290

0

4

88

0

198

1991

275

1

0

66

0

208

1989

7.0

0.0

0.0

1.9

0.1

5.0

Market Share (<>/o)

1990

10.2

0.0

0.1

1991

9.0

0.0

0.0

2.1

3.1

0.0

7.0

0.0

6.8

Asia/Pacific Companies

Goldstar

Hualon Microelectronics Corp.

Hyundai

Macronix

Samsung

Silicon Integrated Systems

United Microelectronics

Winbond Electronics

NA. - Not available

NM - Not meaningful

Source: Dataquest Quly 1992)

74

10

NA

1

31

30

NA

2

NA

118

5

29

8

7

40

15

2

12

219

4

17

14

31

87

13

36

17

2.5

0.3

NA

0.0

1.0

1.0

NA

0.1

NA

4.1

0.2

1.0

0.3

0.2

1.4

0.5

0.1

0.4

7.1

0.1

0.6

0.5

1.0

2.8

0.4

1.2

0.6

©1992 Dataquest Incorporated July—^Reproduction Prohibited

12

Memories Worldwide

Table 1-6

Each Company's Factory Revenue £rom Shipments of Other MOS Memory to the World

(Millions of U.S. Dollars)

Total Market

1989

117

Revenue

1990

174

1991

212

1989

100.0

Market Share (%)

1990

100.0

1991

100.0

North American Companies

Advanced Micro Devices

Cypress Semiconductor

Dallas Semiconductor

Harris

Integrated Device Technology

MOSel

NCR

Other North American Companies

Japanese Companies

Sanyo

Sharp

European Companies

SOS-Thomson

Siemens

NA - Not available

NM - Not meaningful

Source: DaUquest Ouly 1992)

101

3

21

10

5

51

3

7

1

1

0

1

15

0

15

128

11

21

14

5

61

. 1 3

3

0

1

0

1

45

31

14

168

17

24

21

6

83

11

1

5

6

5

1

38

27

11

86.3

2.6

17.9

8.5

4.3

43.6

2.6

6.0

0.9

0.9

0.0

0.9

12.8

0.0

12.8

73.6

6.3

12.1

8.0

2.9

35.1

7.5

1.7

0.0

0.6

0.0

0.6

25.9

17.8

8.0

79.2

8.0

11.3

9.9

2.8

39.2

5.2

0.5

2.4

2.8

2.4

0.5

17.9

12.7

5.2

©1992 Dataquest Incorporated July—Reproduction Prohibited

Table 1-7

Each Company's Factory Revenue from Shipments of Bipolar Memory to the World

(Millions of U.S. Dollars)

Total Market

1989

460

Revenue

1990

431

1991

356

1989

100.0

Market Share (%)

1990

100.0

1991

100.0

North American Companies

Advanced Micro Devices

Harris

Integrated Device Technology

Motorola

National Semiconductor

Raytheon

Texas Instruments

Other North American Companies

* 160

85

0

0

4

49

12

10

0

126

65

5

0

3

25

18

10

0

89

52

2

3

3

13

8

3

5

34.8

18.5

0.0

0.0

0.9

10.7

2.6

2.2

0.0

29.2

15.1

1.2

0.0

0.7

5.8

4.2

2.3

0.0

25.0

14.6

0.6

0.8

0.8

3.7

2.2

0.8

1.4

Japanese Companies

Fujitsu

Hitachi

NEC

European Companies

Philips

SGS-Thomson

NA. - Not available

MM - Not meaningful

Source: Dataquest (July 1992)

Worldwide MOS Memory Market Share, 1989-1991

253

135

97

21

47

47

0

259

144

95

20

46

45

1

231

113

99

19

36

36

0

55.0

29.3

21.1

4.6

10.2

10.2

0.0

60.1

33.4

22.0

4.6

10.7

10.4

0.2

13

64.9

31.7

27.8

5.3

10.1

10.1

0.0

©1992 Dataquest Incorporated July—^Reproduction Prohibited

14

Memories Worldwide

Table 2-1

Top 20 Companies' Factory Revenue from Shipments of MOS Memory to the World

(Millions of U.S. Dollars)

13

14

15

16

17

18

19

20

9

10

11

12

1991

Rank

1

2

3

4

7

8

5

6

1990

Rank

1

3

2

4

14

16

24

21

9

10

11

12

15

18

17

7

8

5

6

13

NA " Not available

NM - Not meaningful

Source: Dataquest (July 1992)

Toshiba

Hitachi

NEC

Samsung

Fujitsu

Mitsubishi

Texas Instruments

Sharp

Micron Technology

Motorola

Intel

Oki

Siemens

SGS-Thomson

Advanced Micro Devices

Goldstar

Hyundai

Matsushita

Cypress Semiconduaor

Sony

All Others

North American Companies

Japanese Companies

European Companies

Asia/Pacific Companies

Total Market

1990

Revenue

1,431

1,224

1,233

971

913

745

741

454

286

395

371

350

312

278

253

96

115

265

166

204

1325

2,977

7,095

731

1,325

12,128

1991

Revenue

1,425

1,330

1,242

1,066

909

762

738

476

455

412

395

380

298

273

270

249

248

217

186

183

1327

3,298

7,141

682

1,720

12,841

1991

Market

Share

(%)

11.1

10.4

9.7

8.3

7.1

5.9

5.7

3.7

3.5

3.2

3.1

3.0

2.3

2.1

2.1

1.9

1.9

1.7

1.4

1.4

10.3

25.7

55.6

5.3

13.4

100.0

Percent

Change

0

9

1

10

0

2

0

5

59

4

6

9

-4

-2

7

159

116

-18

12

-10

0

11

1

-7

30

©1992 Dataquest Incorporated July—Reproduction Prohibited

Worldwide MOS Memory Market Share, 1989-1991

Table 2-2

Top 10 Companies' Factory Revenue from Shipments of MOS DRAMs to the World

(MlUions of U.S. Dollars)

1991

Rank

1

2

3

4

5

6

7

8

9

10

1990

Rank

1

2

3

4

5

7

6

11

8

9

Toshiba

Samsung

NEC

Hitachi

Texas Instruments

Mitsubishi

Fujitsu

Micron Technology

Oki

Siemens

All Others

North American Companies

Japanese Companies

European Companies

Asia/Pacific Companies

Total Market

MA - Not available

NM - Not meaningful

Source; Dataquest (July 1992)

1990

Revenue

961

839

754

617

584

466

536

213

305

298

952

1,235

3,991

298

1,001

6,525

1991

Revenue

957

886

743

661

575

515

503

365

346

287

1144

1,384

4,011

287

1,300

6,982

15

Percent

Change

0

6

-1

7

-2

11

-6

71

13

-4

M

12

1

A

30

1991

Market:

Share

(%)

13.7

12.7

10.6

9.5

8.2

7.4

7.2

5.2

5.0

4.1

16.4

19.8

57.4

4.1

18.6

100.0

©1992 Dataquest Incorporated July—Reproduction Prohibited

16 Memories Worldwide

Table 2-3

Top 10 Companies' Factory Revenue from Shipments of MOS SRAMs to the World

(MiUions of U.S. Dollars)

1991

Rank

1

2

3

4

5

6

7

8

9

10

1990

Rank

1

4

2

3

5

6

8

7

9

10

NA - Not available

NM - Not meaningful

Source: Dataquest QxAy 1992)

Hitachi

Fujitsu

Toshiba

NEC

Sony

Mitsubishi

Motorola

Cypress Semiconduaor

Sharp

Samsung

All Others

Nofth American Companies

Japanese Companies

European Companies

Asia/Pacific Companies

Total Market

1990

Revenue

406

237

247

237

193

183

99

124

94

92

672

524

* 1,756

98

206

2,584

1991

Revenue

449

261

242

241

172

151

132

125

109

93

601

551

1,742

82

201

2,576

Percent

Change

11

10

-2

2

-11

-17

33

1

16

1

-11

5

-1

-16

-2

23.3

21.4

67.6

3-2

7.8

100.0

1991

Market

Share

(%)

17.4

10.1

9.4

9.4

6.7

5.9

5.1

4.9

4.2

3.6

©1992 Dataquest Incoipoiated July—Reproduction Prohibited

Worldwide MOS Memory Market Share, 1989-1991

Table 2-4

Top 10 Companies' Factory Revenue from Shipments of MOS EPROMs to the World

(Millions of U.S. Dollars)

1991

Rank

1

2

9

10

10

3

4

5

6

6

8

1990

Rank

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA - Not available

NM ~ Not meaningful

Source: Dataquest (July 1992)

Advanced Micro Devices

Intel

SGS-Thomson

Texas Instmments

Fujitsu

National Semiconduaor

NEC

Toshiba

Mitsubishi

Philips

Hitachi

All Others

North American Companies

Japanese Companies

European Companies

Asia/Pacific Companies

Total Market

1990

Revenue

NA

•.^ NA

NA

NA

1991

Revenue

225

NA

NA

205

158

136

86

81

NA

NA

NA

NA

NA

81

68

67

59

59

NA

NA

NA

NA

NA

133

774

367

217

0

NA

1,358

17

Percent

Change

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

1991

Market

Share

(%)

16.6

15.1

11.6

10.0

6.3

6.0

6.0

5.0

4.9

4.3

4.3

9.8

57.0

27.0

16.0

0.0

100.0

©1992 Dataquest Incorporated July—^Reproduction Prohibited

18

Memories Worldwide

Table 2-5

Top 10 Companies' Factory Revenue from Shipments of MOS Nonvolatile Memory to the World

(Millions of U.S. Dollars)

1991

Rank

1

2

3

4

5

6

7

8

9

10

1990

Rank

2

1

3

5

4

6

7

8

9

10

NA ' Not available

NM - Not meaningful

Source: Dataquest Qaiy 1992)

Intel

Sharp

NEC

Advanced Micro Devices

Toshiba

Hitachi

SGS-Thomson

Texas Instruments

Fujitsu

National Semiconductor

All Others

North American Companies

Japanese Companies

European Companies

Asia/Pacific Companies

Total Market

1990

Revenue

268

292

242

209

223

201

198

155

140

115

802

1,090

1,347

290

118

2,845

1991

Revenue

311

306

258

237

226

220

208

159

145

97

904

1,195

1,382

275

219

3,071

Percent

Change

16

9

5

3

4

-16

5

7

13

1

13

10

3

-5

86

1991

Market

Share

C%)

10.1

10.0

8.4

7.7

7.4

7.2

6.8

5.2

4.7

3.2

29.4

38.9

45.0

9.0

7.1

100.0

©1992 Dataquest Incorporated July—fieproduction Prohibited

Worldwide MOS Memory Market Share, 1989-1991

Table 2-6

Top 10 Companies' Factory Revenue from Shipments of Other MOS Memory to the World

(Millions of U.S. Dollars)

1991

Rank

1

2

3

4

5

6

6

8

9

10

1990

Rank

1

2

3

5

7

4

6

8

49

10

NA - Not available

NM - Not meaningful

Source: Dataquest (July 1992)

Integrated Device Technology

SGS-Thomson

Cypress Semiconduaor

Dallas Semiconduaor

Advanced Micro Devices

Siemens

MOSel

Harris

Sanyo

Sharp

Ail Others

North American Companies

Japanese Companies

European Companies

Asia/Pacific Companies

Total Market

19

1990

Revenue

61

31

21

14

11

14

13

5

0

1

3

128

1

45

0

174

1991

Revenue

83

27

24

21

17

11

11

6

5.

1

6

168

6

38

0

212

Percent

C h a i s e

36

-13

14

50

55

-21

-15

20

NM

0

100

31

500

-16

NM

22

1991

Market

Share

(%)

39.2

12.7

11.3

9.9

8.0

5.2

5.2

2.8

2.4

0.5

2.8

79.2

2.8

17.9

0.0

100.0

©1992 Dataquest Incorporated July—^Reproduction Prohibited

20

Memories Worldwide

Table 2-7

Top 10 Companies' Factory Revenue from Shipments of Bipolar Memory to the World

(MiUions of U.S. Dollars)

1991

Rank

1

2

3

4

5

6

7

8

8

8

1990

Rank

1

2

3

4

6

5

7

8

10

90

NA - Not availaUe

NM - Not meaningful

Source Dataquest (July 1992)

Fujitsu

Hitachi

Advanced Micro Devices

Philips

NEC

National Semiconductor

Raytheon

Texas Instniments

Motorola

Integrated Device Technology

All Others

North American Companies

Japanese Companies

European Companies

Asia/Pacific Companies

Total Market

1990

Revenue

144

95

65

45

20

25

18

10

3

0

6

126

259

46

0

431

1991

Revenue

113

99

52

36

19

13

8

3

3

3

7

89

231

36

0

356

Percent

Change

-22

4

-20

-20

-5

-^

-56

-70

0

NM

17

-29

-11

-22

NM

-17

1991

Market

Share

C%)

31.7

27.8

14.6

10.1

5.3

3.7

2.2

0.8

0.8

0.8

2.0

25.0

64.9

10.1

0.0

100.0

©1992 Dataquest Incorporated July—Reproduction Prohibited

North American Companies

Advanced Micro Devices

AT&T

Atmel

Catalyst

Worldwide MOS Memory Market Share, 1989-1991

Cypress Semiconductor

Dallas Semiconduaor

Gould AMI

Harris

Honeywell

Integrated Device Technology

Intel

Intl Microelectronic Prod.

mr

Microchip Technology

Micron Technology

MOSel

Motorola

NCR

National Semiconductor

Performance Semiconductor

SEEQ Technology

Texas Instmments

Vitelic

VLSI Technology

"WaferScale Integration

Xicor

Other North American Companies

0.3

0.0

2.0

4.7

0.2

0.0

0.7

4.6

0.3

4.1

0.1

1.7

0.3

0.6

6.8

0.6

0.1

0.6

1.0

0.2

36.5

2.7

0.3

0.7

0.3

3.0

0.3

0.2

21

Table 3-1

Each Company's Factory Revenue from Shipments of MOS Memory to North America

(Millions of U.S. Dollars)

Total Market

1989

5,772

Revenue

1990

4,325

1991

4,510

1989

100.0

Market Share (%)

1990

1991

100.0 100.0

119

252

17

0

55

286

16

2,126

125

13

28

13

119

8

21

27

2

191

7

68

14

31

537

30

22

26

53

46

0.3

0.0

1.0

5.0

0.3

3.3

0.1

1.2

2.1

0.1

0.4

0.5

0.0

2.1

4.4

36.8

2.2

0.2

0.5

0.2

0.2

0.5

9.3

0.5

0.4

0.5

0.9

0.8

1,742

114

25

285

37

0

13

48

28

3

14

341

25

193

3

49

14

17

0

94

206

6

3

44

9

148

13

10

1,578

115

13

31

13

128

12

10

12

2

88

205

8

0

32

201

13

176

3

72

14

24

295

28

5

25

43

10

3.3

0.3

0.2

0.4

0.0

2.1

4.6

0.1

0.1

0.3

7.6

0.6

4.3

0.1

1.1

38.6

2.5

0.1

1.0

0.2

0.3

0.6

6.3

0.8

0.0

0.3

1.1

0.6

Japanese Companies

Fujitsu

Hitachi

Matsushita

Mitsubishi

NEC

NMB Semiconductor

Oki

Ricoh

2,911

237

388

69

321

467

40

211

1

2,053

193

320

54

188

349

39

154

0

2,053

164

352

44

297

338

23

138

0

50.4

4.1

6.7

1.2

5.6

8.1

0.7

3.7

0.0

47.5

4.5

7.4

1.2

4.3

8.1

0.9

3.6

0.0

45.5

3.6

7.8

1.0

6.6

7.5

0.5

3.1

0.0

(Continued)

©1992 Dataquest Incorporated July—^Reproduction Prohibited

22 Memories Worldwide

Table 3-1 (Continued)

Each Company's Factory Revenue from Shipments of MOS Memory to North America

(Millions of U.S. Dollars)

Rohm

Sanyo

Seiko Epson

Sharp

Sony

Toshiba

Other J a p a n e s e Companies

1989

0

0

29

50

63

735

300

R e v e n u e

1 9 9 0

0

1

12

48

59

636

0

1 9 9 1

2

2

9

46

38

600

0

1 9 8 9

0.0

0.0

0.5

0.9

1.1

12.7

5.2

Market S h a r e (%)

1 9 9 0

0.0

0.0

0.3

1.1

1 9 9 1

0.0

0.0

0.2

1.0

1.4

14.7

0.0

0.8

13.3

0.0

European Companies

GEC Plessey

Matra MHS iVlEDL

Philips

Plessey

SGS-Thomson

Siemens

Asia/Pacific Companies

Goldstar

Hyundai

Macronix

Samsung

United Microelectronics

Winbond Electronics

NA - Not a^^ilaUe

NM •- Not meaningful

Source: Dataquest (July 1992)

148

0

2

4

21

1

67

53

587

14

110

21

387

55

NA

169

3

3

0

38

0

75

50

525

23

45

4

416

36

1

20

0

72

50

144

0

2

0

571

62

93

4

409

1

2

2.6

0.0

0.0

0.1

0.4

0.0

1.2

0.9

10.2

0.2

1.9

0.4

6.7

1.0

NA

3.9

0.1

0.1

0.0

0.9

0.0

1.7

1.2

12.1

0.5

1.0

0.1

9.6

0.8

0.0

12.7

1.4

2.1

0.1

9.1

0.0

0.0

3.2

0.0

0.0

0.0

0.4

0.0

1.6

1.1

©1992 Dataquest Incorporated July—Reproduction Prohibited

Worldwide MOS Memory Market Share, 1989-1991

23

Table 3-2

Each Company's Factory Revenue from Shipments of MOS DRAMs to North America

(MilUons of U.S. Dollars)

Total Market

1 9 8 9

3,600

R e v e n u e

1 9 9 0

2,429

1 9 9 1

2,601

1989

100.0

Market S h a r e (%)

1 9 9 0

100.0

1 9 9 1

100.0

North American Companies

Intel

Micron Technology

Motorola

Texas Instruments

Vitelic

947

60

252

151

456

28

602

71

144

130

231

26

696

48

275

118

219

36

26.3

1.7

7.0

4.2

12.7

0.8

24.8

2.9

5.9

5.4

9.5

1.1

26.8

1.8

10.6

4.5

8.4

1.4

J a p a n e s e Companies

Fujitsu

Hitachi

Matsushita

Mitsubishi

NEC

NMB Semiconductor

Oki

Sharp

Toshiba

Other J a p a n e s e Companies

European Companies

Siemens

Asia^acific Companies

Goldstar

Hyundai

Samsung

NA - Not available

NM - Not meaningful

Source: Oataquest Ouly 1992)

53

53

434

10

89

335

2,166

126

215

49

248

407

40

199

24

558

300

50

50

416

23

25

368

1,361

100

166

37

132

300

39

143

16

428

0

49

49

494

60

72

362

1,362

74

185

32

235

285

23

134

13

381

0

1.5

1.5

12.1

0.3

2.5

9.3

60.2

3.5

6.0

1.4

6.9

11.3

1.1

5.5

0.7

15.5

8.3

1.9

1.9

19.0

2.3

2.8

13.9

0.9

5.2

0.5

14.6

0.0

52.4

2.8

7.1

1.2

9.0

11.0

2.1

2.1

17.1

0.9

1.0

15.2

56.0

4.1

6.8

• 1.5

5.4 •

12.4

1.6

5.9

0.7

17.6

0.0

©1992 Oataquest Incorporated July—Reproduction Prohibited

24

Memories Worldwide

Table 3-3

Each Company's Factory Revenue from Shipments of MOS SRAMs to North America

(MilUons of U.S. Dollars)

Total Market

1989

1,053

Revenue

1990

903

1991

889

1989

100.0

Market Share (%)

1990

100.0

1991

100.0

North American Companies

Advanced Micro Devices

AT&T

Atmel

Catalyst

Cypress Semiconduaor

Harris

Honeywell

Integrated Device Technology

Intel

Micron Technology

MOSel

Motorola

NCR

National Semiconductor

Performance Semiconduaor

Texas Instruments

Vitelic

VLSI Technology

Other North American Companies

393

23

5

1

5

86

23

2

82

1

34

14

37

1

25

14

2

2

6

30

353

15

5

1

3

101

8

2

41

15

57

7

44

1

21

14

2

2

5

9

383

12

3

1

0

101

12

0

32

8

66

19

75

2

13

14

2

1

0

22

1.3

0.0

3.6

0.9

7.4

43.1

1.3

0.3

0.1

0.0

11.4

2.1

8.4

0.2

1.5

1.6

0.2

0.1

0.0

2.5

0.9

0.2

4.5

1.7

6.3

0.8

4.9

0.1

2.3

1.6

0.2

0.2

0.6

1.0

39.1

1.7

0.6

0.1

0.3

11.2

1.3

0.2

0.2

0.6

2.8

1.3

3.5

0.1

2.4

37.3

2.2

0.5

0.1

0.5

8.2

2.2

0.2

7.8

0.1

3.2

Japanese Companies

Fujitsu

Hitachi

Mitsubishi

NEC

Oki

Sanyo

Seiko Epson

Sharp

Sony

Toshiba

499

72

119

55

50

9

0

29

12

63

90

451

61

106

44

40

8

0

12

16

59

105

444

65

112

48

43

3

2

9

17

38

107

47.4

6.8

11.3

5.2

4.7

0.9

0.0

2.8

1.1

6.0

8.5

49.9

6.8

11.7

4.9

4.4

0.9

0.0

1.3

1.8

6.5

11.6

49.9

7.3

12.6

5.4

4.8

0.3

0.2

1.0

1.9

4.3

12.0

(Continued)

©1992 Dataquest Incraiporated July—Reproduction Prohibited

Table 3-3 (Conthiued.)

Each Company's Factory Revenue from Shipments of MOS SRAMs to North America

(Millions of U.S. Dollars)

European Companies

GEC Plessey

Matra MHS

MEDL

Philips

SGS-Thomson

1989

38

0

2

4

0

32

Revenue

1990

20

2

3

0

1

14

1991

15

0

2

0

1

12

Market Share (%)

1989 1990 1991

3.6

0.0

2.2

0.2

1.7

0.0

0.2

0.4

0.0

3.0

0.3

0.0

0.1

1.6

0.2

0.0

0.1

1.3

Asia/Pacific Cotnpanies

Goldstar

Hyundai

Samsung

United Microelectronics

Winbond Electronics

NA - Not available

NM - Not meaningful

Source: Dataquest Ouly 1992)

Worldwide MOS Memory Market Share, 1989-1991

123

2

21

45

55

NA

79

0

13

30

36

0

47

2

15

28

1

1

11.7

0.2

2.0

4.3

5.2

NA

8.7

0.0

1.4

3.3

4.0

0.0

25

5.3

0.2

1.7

3.1

0.1

0.1

©1992 Dataquest Incorporated July—^Reproduction Prohibited

26 Memories Worldwide

Table 3-4

Each Company's Factory Revenue from Shipments of EPROMs to North America

(Millions of U.S. Dollars)

Total Market

1 9 8 9

NA

R e v e n u e

1 9 9 0

NA

1991

438

1 9 8 9

NA

North American Companies

Advanced Micro Devices

Aimel

Catalyst

Cypress Semiconductor

Intel

Microchip Technology

National S e m i c o n d u a o r

Texas Instruments

WaferScale Integration

Other North American Companies

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

317

83

12

1

31

75

11

32

58

13

1

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

Japanese Companies

Fujitsu

Hitachi

Mitsubishi

NEC

Toshiba

Eurojjean Companies

Philips

SGS-Thomson

NA - Not available

NM - Not meaningful

Source: Dataquest Ouly 1992)

NA

NA

NA

NA

NA

NA

NA

NA

NA

"NA

NA

NA

NA

NA

NA

NA

NA

NA

67

20

14

10

3

20

54

18

36

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

1 9 9 0

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

15.3

4.6

3.2

2.3

0.7

4.6

12.3

4.1

8.2

1991

100.0

72.4

18.9

2.7

0.2

7.1

17.1

2.5

7.3

13.2

3.0

0.2

©1992 Dataquest Incorporated July—Reproduction Prohibited

North American Companies

Advanced Micro Devices

AT&T

Atmel

Catalyst

Cypress Semiconductor

Gould AMI

Harris

Intel

Int'l Microelectronic Prod.

rrr

Microchip Technology

Motorola

National Semiconduaor

Worldwide MOS Memory Market Share, 1989-1991

SEEQ Technology

Texas Instalments

VLSI Technology

WaferScale Integration

Xicor

Other North American Companies

709

100

8

27

8

15

21

1

191

17

0

55

3

43

31

79

16

26

53

15

528

91

8

30

10

11

10

1

119

8

0

32

2

51

24

62

0

25

43

1

543

90

0

43

9

31

10

0

150

6

3

14

0

36

25

64

0

13

48

1

Z7

Table 3-5

Each Company's Factory Revenue from Shipments of MOS Nonvolatile Memory to North America

(Millions of U.S. Dollars)

Total Market

1989

1,041

Revenue

1990

877

1991

882

1989

100.0

Market Share (%)

1990

100.0

1991

100.0

18.3

1.6

0.0

5.3

0.3

4.1

3.0

7.6

1.5

2.5

5.1

1.4

68.1

9.6

0.8

2.6

0.8

1.4

2.0

0.1

2.7

7.1

0.0

2.9

4.9

0.1

1.3

1.1

0.1

13.6

0.9

0.0

3.6

0.2

5.8

60.2

10.4

0.9

3.4

1.1

61.6

10.2

0.0

4.9

1.0

3.5

1.1

0.0

17.0

0.7

0.3

1.6

0.0

4.1

2.8

7.3

0.0

1.5

5.4

0.1

Japanese Companies

Fujitsu

Hitachi

Matsushita

Mitsubishi

NEC

Oki

Ricoh

Rohm

Sanyo

Sharp

Toshiba

245

39

54

20

18

10

0

0

3

1

13

87

9

3

0

0

1

15

103

240

32

48

17

12

246

25

55

12

14

10

1

0

2

0

15

112

23.5

3.7

5.2

1.9

1.7

1.0

0.3

0.1

0.0

0.0

1.2

8.4

27.4

3.6

0.3

0.0

0.0

0.1

1.7

11.7

5.5

1-9

1.4

1.0

27.9

2.8

6.2

1.4

1.6

1.1

0.1

0.0

0.2

0.0

1.7

12.7

(Continued)

©1992 Dataquest Incoipoiated July—^Reproduction Prohibited

2 8

Memories Worldwide

Table 3-5 (Continued)

Each Company's Factory Revenue £rom Shipments of MOS Nonvolatile Memory to North America

(Millions of U.S. Dollars)

European Companies

GEO Plessey

Philips

Plessey

SGS-Thomson

1989

57

0

21

1

35

Revenue

1990

79

1

37

0

41

1991

63

0

19

0

44

Market Share (%)

1989 1990 1991

5.5

0.0

2.0

0.1

3.4

9.0

0.1

4.2

0.0

4.7

7.1

0.0

2.2

0.0

5.0

Asia/Pacific Companies

Goldstar

Hyundai

Macronix

Samsung

Winbond Electronics

NA - Not available

NM - Not meaningful

Source: Dataquest (July 1992)

30

2

0

21

7

NA

30

0

7

4

18

1

30

0

6

4

19

1

2.9

0.2

0.0

2.0

0.7

NA

3.4

0.0

0.8

0.5

2.1

0.1

3.4

0.0

0.7

0.5

2.2

0.1

i

©1992 Dataquest Incorporated July-Reproduction Prohibited

Worldwide MOS Memory Marii:et Share, 1989-1991 29

Table 3-6

Each Company's Factory Revenue from Shipments of Other MOS Memory t o N o r t h America

OMJllions of U.S. Dollars)

Total Market

1989

78

Revenue

1990

116

1991

138

1989

100.0

Market Share (%)

1990

100.0

1991

100.0

North American Conipanies

Advanced Micro Devices

Cypress Semiconduaor

Dallas Semiconductor

Harris

Integrated Device Technology

MOSel

NCR

Other North American Companies

Japanese Companies

Sharp

77

2

18

8

3

37

2

6

1

95

9

16

12

3

47

6

2

0

120

12

16

13

5

62

6

1

5

1

1

98.7

2.6

23.1

10.3

3.8

47.4

2.6

7.7

1.3

81.9

7.8

13.8

10.3

2.6

40.5

5.2

1.7

0.0

87.0

8.7

11.6

9.4

3.6

44.9

4.3

0.7

3.6

0.7

0.7

European Companies

SGS-Thomson

Siemens

NA - Not available

NM - Not meaningful

Source: Dataquest (July 1992)

0

0

0

1

1

20

20

0

1

1

17

16

1

1.3

1.3

0.0

0.0

0.0

0.9

0.9

17.2

17.2

0.0

12.3

11.6

0.7

©1992 Dataquest Incorporated July—^Reproduction Prohibited

30

Table 3-7

Each Company's Factory Revenue from Shipments of Bipolar Memory to North America

(Millions of U.S. Dollars)

Total Market

1989

180

Revenue

1990

160

1991

131

1989

100.0

Market Share (%)

1990

100.0

1991

100.0

North American Companies

Advanced Micro Devices

Harris

Integrated Device Technology

Motorola

National Semiconduaor

Raytheon

Texas Instruments

Other North American Companies

Memories 'Worldwide

94

47

0

0

3

29

11

4

0

73

36

5

0

3

11

14

4

0

52

26

2

2

2

6

8

1

5

52.2

26.1

0.0

0.0

1.7

16.1

6.1

2.2

0.0

45.6

22.5

3.1

0.0

1.9

6.9

8.8

2.5

0.0

39.7

19.8

1.5

1.5

1.5

4.6

6.1

0.8

3.8

Japanese Comp>anies

Fujitsu

Hitachi

NEC

59

41

17

1

60

43

16

1

60

39

20

1

32.8

22.8

9.4

0.6

37.5

26.9

10.0

0.6

45.8

29.8

15.3

0.8

European Companies

Philips

NA - Not available

NM - Not meaningful

Source: Dataquest Ouly 1992)

27

27

27

27

19

19

15.0

15.0

16.9

16.9

14.5

14.5

©1992 Dataquest Incorporated July—Reproduction Prohibited

Worldwide MOS Memory Market Share, 1989-1991

31

Table 4^1

Each Company's Factory Revenue from Shipments of MOS Memory to Japan

(Millions of XJ.S. Dollars)

Total Market

1989

5,629

Revenue

1990

4,196

1991

4,228

1989

100.0

Market Share (»/o)

1990

100.0

1991

100.0

North American Companies

Advanced Micro Devices

Atmel

Catalyst

Cypress Semiconduaor

C>allas Semiconduaor

Gould AMI

Harris

Integrated Device Technology

Intel

Microchip Technology

Micron Technology

MOSel

Motorola

National Semiconductor

Performance Semiconductor

SEEQ Technology

Texas Instmments

Vitelic

WaferScale Integration

Xicor

Other North American Conipanies

163

3

0

8

4

1

98

' 4

1

3

421

41

3

10

7

1

1

8

10

45

8

2

3

152

2

0

5

6

3

2

9

103

3

0

406

47

7

12

30

3

9

8

1

1

470

58

7

13

8

0

1

1

3

37

19

16

25

91

4

0

3

162

5

3

13

1

7.5

0.7

0.1

0.2

0.1

0.0

0.0

0.1

0.2

0.8

0.1

0.0

0.0

1.7

0.1

0.0

0.1

2.9

0.1

0.0

0.1

0.1

0.1

3.6

0.0

0.0

0.1

0.1

0.3

0.7

0.1

0.0

0.2

2.5

0.1

0.0

0.2

0.0

0.0

0.2

9.7

1.1

0.1

0.2

11.1

1.4

0.2

0.1

0.0

0.1

3.8

0.9

0.4

0.4

0.6

2.2

0.1

0.1

0.3

0.0

0.3

0.2

0.0

0.0

0.0

0.1

Japanese Companies

Fujitsu

Hitachi

Matsushita

Mitsubishi

NEC

NMB Semiconductor

Oki

Ricoh

Rohm

Sanyo

Seiko Epson

Sharp

Sony

Toshiba

5,131

750

768

221

604

847

24

156

25

5

108

108

335

122

656

3,694

546

689

154

366

664

16

126

26

11

69

43

353

109

522

3,621

577

719

147

290

667

8

129

8

18

59

24

376

110

487

91.2

13.3

13.6

3.9

10.7

15.0

0.4

2.8

0.4

0.1

1.9

1.9

6.0

2.2

11.7

88.0

130

16.4

3.7

8.7

15.8

0.4

3.0

0.6

0.3

1.6

1.0

8.4

2.6

12.4

0.6

8.9

2.6

11.5

(ContinuecD

3.1

0.2

0.4

1.4

85.6

13.6

17.0

3.5

6.9

15.8

0.2

©1992 Dataquest Incorporated July—Reproduction Prohibited

32 Memories Worldwide

Table 4-1 (Continued)

Each Company's Factory Revenue from Shipments of MOS Memory to Japan

(Millions of U.S. Dollars)

Yamaha

Other Japanese Comjjanies

1989

0

402

Revenue

1990

0

0

1991

2

0

1989

0.0

Market Share (%)

1990

0.0

1991

0.0

7.1

0.0 0.0

European Companies

Matra MHS

Philips

SGS-Thomson

Asia/Pacific Companies

Goldstar

Hyundai

Macronix

Samsung

United Microelectronics

Winbond Electronics

NA - Not availaUe

NM - Not meaningful

Source Dataquest Only 1992)

17

1

13

3

60

3

3

8

46

0

NA

23

1

7

15

::73

14

2

1

55

0

1

27

0

5

22

110

20

5

6

73

5

1

0.3

0.0

0.2

0.1

1.1

0.1

0.1

0.1

0.8

0.0

NA

0.5

0.0

0.2

0.4

1.7

0.3

0.0

0.0

1.3

0.0

0.0

2.6

0.5

0.1

0.1

1.7

0.1

0.0

0.6

0.0

0.1

0.5

©1992 Dataquest Incorf>orated July—Reproduction Prolubited

North American Companies

Micron Technology

MOSel

Motorola

Texas Instruments

Vitelic

Worldwide MOS Memory Market Share, 1989-1991

Table 4-2

Each Company's Factory Revenue from Shipments of MOS DRAMs to Japan

(Millions of TJ.S. Dollars)

Total Market

1989

2,893

Revenue

1990

1,992

1991

1,948

1989

100.0

220

2

0

n

138

3

216

2

0

75

137

2

235

15

13

72

130

5

7.6

0.1

0.0

2.7

4.8

0.1

Japanese Companies

Fujitsu

Hitachi

Matsushita

Mitsubishi

NEC

NMB Semiconduaor

Oki

Sanyo

Sharp

Sony

Toshiba

Other Japanese Companies

Asia/Pacific Companies

Goldstar

Hyundai

Samsung

NA - Not available

MM * Not meaxungfiil

Source: Dataquest 0"'y 1992)

2,635

491

388

127

313

437

24

131

14

65

0

492

153

38

3

0

35

1,733

325

322

81

177

298

16

102

15

44

3

350

0

43

13

0

30

1,663

321

325

80

147

295

8

102

21

42

3

319

0

50

9

2

39

91.1

17.0

13.4

4.4

10.8

15.1

0.8

4.5

0.5

2.2

0.0

17.0

5.3

1.3

0.1

0.0

1.2

33

1990

100.0

10.8

0.1

0.0

3.8

6.9

0.1

87.0

16.3

16.2

4.1

8.9

15.0

0.8

5.1

0.8

2.2

0.2

17.6

0.0

2.2

0.7

0.0

1.5

1991

100.0

12.1

0.8

0.7

3.7

6.7

0.3

85.4

16.5

16.7

4.1

7.5

15.1

0.4

5.2

1.1

2.2

0.2

16.4

0.0

2.6

0.5

0.1

2.0

©1992 Dataquest Incorporated July—Reproduction Prohibited

34

Memories Worldwide

Table 4-3

Each Company's Factory Revenue from Shipments of MOS SRAMs to Japan

(Millions of U.S. Dollars)

Total Market

1989

1,545

Revenue

1990

1,109

1991

1,081

1989

100.0

Market Share (%)

1990

1991

100.0 100.0

North American Companies

Advanced Micro Devices

Catalyst

Cypress Semiconductor

Harris

Integrated Device Technology

Micron Technology

MOSel

Motorola

Performance Semiconduaor

Texas Instruments

Other North American Companies

19

1

0

0

49

8

1

6

6

7

0

1

60

8

0

7

5

9

0

3

27

0

0

1

40

2

0

7

0

1

1

8

19

0

2

0

3.2

0.5

0.0

0.1

1.2

0.1

0.0

0.0

0.5

0.1

0.4

0.4

5.4

0.7

0.0

0.6

0.5

0.8

0.0

0.3

2.4

0.0

0.0

0.1

3.7

0.2

0.0

0.6

0.0

0.1

0.1

0.7

1.8

0.0

0.2

0.0

Japanese Companies

Fujitsu

Hitachi

Matsushita

Mitsubishi

NEC

Oki

Rohm

Sanyo

Seiko Epson

Sharp

Sony

Toshiba

Other Japanese Companies

European Companies

Matra MHS

Philips

SGS-Thomson

1,018

140

237

21

109

148

6

6

44

43

72

99

93

0

4

1

2

1

1,484

165

250

25

176

158

7

5

72

108

60

115

94

249

4

1

0

3

83

100

89

0

9

8

26

24

995

162

260

19

69

146

2

0

1

1

3.9

7.4

6.1

16.1

0.5

0.3

4.7

7.0

96.1

10.7

16.2

1.6

11.4

10.2

0.3

0.1

0.0

0.2

13.3

0.5

0.5

4.0

3.9

6.5

91.8

12.6

21.4

1.9

9.8

8.9

8.4

0.0

0.4

0.1

0.2

0.1

0.2

0.0

0.1

0.1

(Continued)

13.5

0.8

0.7

2.4

2.2

7.7

9.3

8.2

0.0

92.0

15.0

24.1

1.8

6.4

©1992 Dataquest Incorporated July—Reproduction Piotiibited

\

Worldwide MOS Memory Market Share, 1989-1991

35

Table 4-3 (Continued)

Each Company's Factory Revenue from Shipments of MOS SRAMs to Japan

(MUlions of i;.S. Dollars)

Asia/Pacific Companies

Goldstar

Hyundai

Samsung

United Microelectronics

Winbond Electronics

NA - Not available

NM " Not meaningful

Source: Dataquest (July 1992)

1989

8

0

3

5

0

NA

Revenue

1990

27

1

2

23

0

1

1991

44

11

3

27

2

1

Market Share (%)

1989 1990 1991

0.5

0.0

2.4

0.1

4.1

1.0

0.2

0.2 0.3

0.3

0.0

NA

2.1

0.0

0.1

2.5

0.2

0.1

>

©1992 Dataquest Incorpoiated July—^Reproduction Prohibited

36 Memories Voridwide

Table 4-4

Each Company's Factory Revenue from Shipments of MOS EPROMs to Japan

(MilUons of U.S. Dollars)

Total Market

1 9 8 9

NA

R e v e n u e

1990

NA

1 9 9 1

414

1 9 8 9

NA

North American Companies

Advanced Micro Devices

Atmel

Intel

Microchip Technology

National Semiconductor

Texas Instruments

WaferScale Integration

Other North American Companies

NA

NA.

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

134

53

6

31

10

3

27

3

1

NA

NA

NA

NA

NA

NA

NA

NA

NA

J a p a n e s e Companies

Fujitsu

Hitachi

Mitsubishi

NEC

Oki

Sharp

Toshiba

NA

NA

NA

NA

NA

NA

NA

NA

NA

•NA

NA

NA

NA

NA

NA

NA

260

54

34

53

72

3

3

41

NA

NA

NA

NA

NA

NA

NA

NA

European Companies

Philips

SGS-ITiomson

NA " Not available

NM - Not meaningful

Source: Dataquest Quly 1992)

NA

NA

NA

NA

NA

NA

20

4

16

NA

NA

NA

1 9 9 0

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

1 9 9 1

100.0

32.4

12.8

1.4

7.5

2.4

0.7

6.5

0.7

0.2

62.8

13.0

8.2

12.8

17.4

0.7

0.7

9.9

4.8

1.0

3.9

©1992 Dataquest Incorporated July—Reproduction Prohibited

Worldwide MOS Memory Market Share, 1989-1991

37

Table 4-5

Each Company's Factory Revenue from Shipments of MOS Nonvolatile Memory to Japan

(MiUions of U.S. Dollars)

Total Market

1989

1,184

Revenue

1990

1,080

1991

1,181

1989

100.0

Market Share (%)

1990

100.0

1991

100.0

North American Companies

Advanced Micro Devices

Atmel

Catalyst

Gould AMI

Intel

Microchip Technology

Motorola

National Semiconductor

SESQ Technology

Texas Instruments

WaferScale Integration

Xicor

Other North American Companies

145

33

3

9

1

45

8

2

4

3

25

0

8

4

116

38

3

9

1

30

3

1

3

3

15

0

5

5

185

54

7

13

1

37

19

0

4

3

30

3

13

1

3.8

0.7

0.2

0.3

0.3

2.1

12.2

2.8

0.3

0.8

0.1

0.0

0.7

0.3

10.7

3.5

0.3

0.8

0.1

2.8

0.3

0.1

0.3

0.3

1.4

0.0

0.5

0.5

15.7

4.6

0.6

1.1

0.1

3.1

1.6

0.0

0.3

0.3

2.5

0.3

1.1

0.1

Japanese Companies

Fujitsu

Hitachi

Matsushita

Mitsubishi

NEC

Oki

Ricoh

Rohm

Sanyo

Sharp

Sony

Toshiba

Yamaha

1,012

94

130

69

115

252

18

25

0

22

210

7

70

0

943

81

130

52

80

218

18

26

5

10

237

7

79

0

958

94

134

48

74

226

18

8

10

7

251

7

79

2

85.5

7.9

11.0

5.8

9.7

21.3

1.5

2.1

0.0

1.9

17.7

0.6

5.9

0.0

87.3

7.5

12.0

4.8

7.4

20.2

1.7

2.4

0.5

0.9

21.9

0.6

7.3

0.0

81.1

8.0

11.3

4.1

6.3

19.1

1.5

0.7

0.8

0.6

21.3

0.6

6.7

0.2

European Companies

Philips

SGS-Thomson

13

13

0

18

5

13

22

4

18

1.1

1.1

0.0

1.7

0.5

1.2

1.9

0.3

1.5

(Continued)

©1992 I>ataquest Incorporated July—^Reproduction Prohibited

38 Memories Worldwide

Table 4-5 (Continued)

Each Compaay's Factory Revenue from Shipments of MOS Nonvolatile Memory to Japan.

(MilUons of U.S. Dollars)

Asia/Pacific Companies

Macronix

Samsung

United Microelectronics

MA - Not available

NM - Not meaningful

Source: E>ataquest Quiy 1992)

1989

14

8

6

0

Revenue

1990

3

1

2

0

1991

16

6

7

3

Market Share (%)

1989 1990 1991

1.2

0.3

1.4

0.7

0.1

0.5

0.5

0.0

0.2

0.0

0.6

0.3

i

©1992 Dataquest Incorporated July—Repioduction Prohibited

i

Worldwide MOS Memory Market Share, 1989-1991

39

Table 4-6

Each Company's Factory Revenue from Shipments of Other MOS Memory to Japan

(Millions o f U.S. Dollars)

Total Market

1989

7

Revenue

1990

15

1991

18

1989

100.0

Market Share (%)

1990

100.0

1991

100.0

North American Companies

Advanced Micro Devices

Cypress Semiconductor

Dallas Semiconductor

Harris

Integrated Device Technology

MOSel

1

2

3

0

7

0

1

14

1

1

1

2

3

6

10

2

1

0

1

2

4

100.0

0.0

14.3

14.3

28.6

42.9

0.0

93.3

6.7

6.7

6.7

13.3

20.0

40.0

55.6

11.1

5.6

0.0

5.6

11.1

22.2

Japanese Companies

Sanyo

European Companies

SGS-Thomson

NA. " Not available

NM " Not meanixigful

Source: Dataqxaest O^y 1992) ft

&

0

0

0

•tt

X

1

5

5

3

3

0.0

0.0

0.0

0.0

0.0

0.0

6.7

6.7

27.8

27.8

16.7

16.7

©1992 Dataquest Incorporated July—^Reproduction Prohibited

40 Memories Worldwide

Table 4-7

Each Company's Factory Revenue from Shipments of Bipolar Memory to Japan

(Millions of U.S. Dollars)

Total Market

1989

191

Revenue

1990

194

1991

165

Market Share (%)

1989 1990 1991

100.0

100.0

100.0

North American Companies

Advanced Micro Devices

Motorola

National Semiconduaor

Texas Instruments

20

9

1

6

4

15

5

0

6

4

7

3

0

3

1

10.5

4.7

0.5

3.1

2.1

7.7

2.6

0.0

3.1

2.1

42

1.8

0.0

1.8

0.6

Japanese Companies

Fujitsu

Hitachi

NEC

European Companies

Philips

NA - Not available

NM - Not meaningful

Source: Dataquest Ouly 1992)

169

87

68

14

2

2

178

94

71

13

1

1

157

71

73

13

1

1

88.5

45.5

35.6

7.3

1.0

1.0

91.8

48.5

36.6

6.7

0.5

0.5

95.2

43.0

44.2

7.9

0.6

0.6

I

©1992 Dataquest Incorporated July—Reproduction Prohibited

i

Worldwide MOS Memory Market Share, 1989-1991 41

Table 5-1

Each Company's Factory Revenue from Shipments of MOS Memory to Europe

(MiUions of U.S. Dollars)

Total Market

1989

2,417

Revenue

1990

2,050

1991

2,129

1989

100.0

Market Share (%)

1990

100.0

1991

100.0

North American Conpanies

Advanced Micro Devices

AT&T

Atmel

Catalyst

Cypress Semiconduaor

Dallas Semiconductor

Gould AMI

Harris

Integrated Device Technology

Intel

ITT

Microchip Technology

Micron Technology

MOSel

Motorola

National Semiconduaor

Performance Semiconduaor

SEEQ Technology

Texas Instruments

Vitelic

VLSI Technology

WaferScale Integration

Xicor

Other North American Companies

687

71

0

10

6

21

1

1

1

24

102

2

22

9

6

60

2

60

28

1

6

250

3

1

0

591

61

0

12

8

• 28

1

1

3

18

84

0

4

46

2

76

31

4

6

181

1

3

2

18

1

621

61

1

20

8

26

4

0

5

22

27

3

4

171

4

0

4

25

2

93

5

15

45

4

72

2.2

0.1

3.7

1.5

0.2

0.9

4.1

0.0

0.2

0.3

8.8

0.0

0.1

0.1

0.9

0.0

28.8

3.0

0.0

0.6

0.4

1.4

0.0

0.0

0.1

28.4

2.9

0.0

0.4

0.2

0.9

0.0

0.0

0.0

1.0

4.2

0.4

0.2

10.3

0.1

0.0

0.1

0.9

0.0

2.5

0.1

2.5

1.2

0.0

0.2

29.2

2.9

0.0

0.0

0.2

1.0

4.4

0.2

0.9

0.4

1.2

0.2

1.3

0.1

0.2

8.0

0.2

0.7

2.1

0.2

3.4

0.0

0.2

1.2

0.1

Japanese Companies

Fujitsu

Hitachi

Matsushita

Mitsubishi

NEC

NMB Semiconduaor

Oki

Ricoh

Rohm

Sanyo

Seiko Epson

1,040

110

158

61

97

214

42

45

1

0

1

0

808

87

140

46

63

164

10

30

0

0

1

0

803

67

159

8

86

164

5

57

0

3

8

3

43.0

4.6

6.5

2.5

4.0

8.9

1.7

1.9

0.0

0.0

0.0

0.0

39.4

4.2

6.8

2.2

3.1

8.0

0.5

1.5

0.0

0.0

0.0

0.0

37.7

3.1

7.5

0.4

4.0

7.7

0.2

2.7

0.0

0.1

0.4

0.1

(Continued)

©1992 Dataquest Incorporated July—Reproduction Prohibited

42

Memories Worldwide

Table 5-1 (Continued)

Each Company's Factory Revenue from Shipments of MOS Memory to Europe

(Millions of U^. Dollars)

Sharp

Sony

Toshiba

Other Jap)anese Companies

1 9 8 9

21

24

228

38

R e v e n u e

1 9 9 0

22

23

222

0

1991

23

21

199

0

1 9 8 9

Market S h a r e (%)

1 9 9 0

1 9 9 1

0.9

1.0

9.4

1.6

1.1

1.1

10.8

0.0

1.1

1.0

9.3

0.0

Eurof)ean Companies

Eurosil

GEC Plessey

Matra MHS

MEDL

Philips

Plessey

SGS-Thomson

Siemens

Asia/Pacific Companies

Goldstar

Hyundai

Macronix

Samsung

United Microelectronics

Winbond Electronics

NA. ' Not available

NM - Not meaningful

Source: Dataquest Qu\y 1992)

480

0

0

28

3

20

2

129

298

210

4

19

1

186

0

NA

447

0

5

32

0

29

0

143

238

204

8

11

1

184

0

0

417

1

0

32

0

26

0

134

224

288

33

25

0

226

3

1

19.9

0.0

0.0

1.2

0.1

0.8

0.1

5.3

12.3

8.7

0.2

0.8

0.0

7.7

0.0

NA

21.8

0.0

0.2

1.6

0.0

1.4

0.0

7.0

11.6

10.0

0.4

0.5

0.0

9.0

0.0

0.0

19.6

0.0

0.0

1.5

0.0

1.2

0.0

6.3

10.5

13.5

1.6

1.2

0.0

10.6

0.1

0.0

©1992 Dataquest Incorporated July-^teproduction Prohibited

Worldwide MOS Memory Market Share, 1989-1991

43

Table 5-2

Each Company's Factory Revenue from Shipments of MOS DRAMs to Europe

(Millions of U.S. Dollars)

Total Market

1989

1,537

Revenue

1990

1,155

1991

1,205

Market Share (%)

1989 1990 1991

100.0 100.0 100.0

North American Companies

Intel

Micron Technology

Motorola

Texas Instruments

Vjtelic

282

3

55

47

175

2

224

5

36

6:

• 122

0

222

11

35

45

127

4

18.3

0.2

3.6

3.1

11.4

0.1

19.4

0.4

3.1

5.3

10.6

0.0

18.4

0.9

2.9

3.7

10.5

0.3

Japanese Companies

Fujitsu

Hitachi

Matsushita

Mitsubishi

NEC

NMB Semiconductor

Oki

Sanyo

Sharp

Toshiba

Other Japanese Companies

European Companies

Siemens

Asia/Pacific Companies

Goldstar

Hyundai

Samsung

NA •• Not avajlaUe

NM - Not m e a n i n ^ ^

Source: Dataquest (July 1992)

774

72

111

61

85

152

42

36

0

5

172

38

298

298

183

4

9

170

523

55

89

46

36

110

10

24

0

4

149

0

238

238

170

7

6

157

517

43

99

8

57

106

5

55

7

2

135

0

224

224

242

32

13

197

50.4

4.7

7.2

4.0

5.5

9.9

2.7

2.3

0.0

0.3

11.2

2.5

19.4

19.4

11.9

0.3

0.6

11.1

45.3

4.8

7.7

4.0

3.1

9.5

0.9

2.1

0.0

0.3

12.9

0.0

20.6

20.6

14.7

0.6

0.5

13.6

42.9

3.6

8.2

0.7

4.7

8.8

0.4

4.6

0.6

0.2

11.2

0.0

18.6

18.6

20.1

2.7

1.1

16.3

©1992 Dataquest Incorporated July-Reproduction Prohibited

44 Memories Worldwide

Table 5-3

Each Company's Factory Revenue £rom Shipments of MOS SRAMs to Europe

(MiUions of U^. Dollars)

Total Market

1989

354

Revenue

1990

365

1991

377

1989

100.0

Market Share (%)

1990

100.0

1991

100.0

North American Companies

Advanced Micro Devices

AT&T

Catalyst

Cypress Semiconductor

Harris

Integrated Device Technology

Intel

Micron Technology

MOSel

Motorola

National Semiconductor

Performance Semiconductor

Vitelic

VLSI Technology

Other North American Companies

68

18

0

1

12

1

14

0

5

1

12

1

1

1

1

0

69

10

0

1

14

3

8

0

10

1

14

1

4

1

2

0

83

2

1

0

13

5

9

7

10

3

26

2

3

0

0

2

0.3

4.0

0.0

1.4

0.3

3.4

19.2

5.1

0.0

0.3

3.4

0.3

0.3

0.3

0.3

0.0

18.9

2.7

0.0

0.3

3.8

0.8

2.2

0.0

2.7

0.3

3.8

0.3

1.1

0.3

0.5

0.0

1.9

2.7

0.8

6.9

0.5

0.8

0.0

0.0

22.0

0.5

0.3

0.0

3.4

1.3

2.4

0.5

Japanese Companies

Fujitsu

Hitachi

Mitsubishi

NEC

Oki

Sanyo

Seiko Epson

Sharp

Sony

Toshiba

European Companies

GEC Plessey

Matra MHS

MEDL

Philips

SGS-Thomson

183

19

36

8

51

4

1

0

4

24

36

83

0

28

3

3

49

199

16

41

24

44

3

1

0

5

23

42

68

2

32

0

4

30

200

12

47

22

46

1

1

3

8

21

39

58

0

32

0

3

23

51.7

5.4

10.2

2.3

14.4

1.1

0.3

0.0

1.1

6.8

10.2

23.4

0.0

7.9

0.8

0.8

13.8

54.5

4.4

11.2

6.6

12.1

0.8

0.3

0.0

1.4

6.3

11.5

18.6

0.5

8.8

0.0

1.1

8.2

15.4

0.0

8.5

0.0

0.8

6.1

(Continued)

53.1

3.2

12.5

5.8

12.2

0.3

0.3

0.8

2.1

5.6

10.3

©1992 Dataquest Incorporated July—Reproduction Prohibited

^

Worldwide MOS Memory Market Share, 1989-1991 45

Table 5-3 (Continued)

Each Company's Factory Revenue from Shipments of MOS SRAMs to Europe

(MUUons o f U.S. Dollars)

Asia/Pacific Companies

Goldstar

Hyundai

Samsung

United Microelectronics

Winbond Electronics

NA - Not available

MM - Not meaningful

Source: Dataquest 0"ly 1992)

1989

20

0

10

10

0

NA

Revenue

1990

29

1

5

23

0

0

1991

36

1

9

22

3

1

Market Share (%)

1989 1990 1991

5.6

0.0

2.8

2.8

0.0

NA

7.9

0.3

1.4

6.3

0.0

0.0

9.5

0.3

2.4

5.8

0.8

0.3

I

©1992 Dataquest Incorporated July—Reproduction Piohibited

46

Memories Worldwide

Table 5-4

Each Company's Factory Revenue from Shipments of MOS EPROMs to Europe

(Millions of U.$. Dollars)

Total Market

1989

NA

Revenue

1990

NA

1991

303

Market Share (%)

1989 1990 1991

NA NA

100.0

North American Companies

Advanced Micro Devices

Atmel

Catalyst

Cypress Semiconductor

Intel

Microchip Technology

National Semiconductor

Texas Instruments

WaferScale Integration

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

186

53

6

1

6

55

7

21

33

4

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

61.4

17.5

2.0

0.3

2.0

18.2

2.3

6.9

10.9

1.3

Jajsanese Con^anies

Fujitsu

Hitachi

Mitsubishi

NEC

Toshiba

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

26

9

4

3

4

6

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

8.6

3.0

1.3

1.0

1.3

2.0

European Companies

Philips

SGS-Thomson

NA " Not available

NM " Not meaningful

Source: Dataquest Ouly 1992)

NA

NA

NA

NA

NA

NA

91

17

74

NA

NA

NA

NA

NA

NA

30.0

5.6

24.4

©1992 Dataquest Incorporated July—Reproduction Prohibited

Worldwide MOS Memory Market Share, 1989-1991

281

50

12

7

10

1

79

0

4

1

30

6

59

1

2

18

1

86

16

10

3

10

3

0

0

13

31

47

Table 5-5

Each Company's Factory Revenue from Shipments of MOS Nonvolatile Memory to Europe

(MiUions of U.S. Dollars)

Total Market

1989

511

Revenue

1990

503

1991

512

1989

100.0

Market Share (%)

1990

100.0

1991

100.0

North American Companies

Advanced Micro Devices

Atmel

Catalyst

Cypress Semiconduaor

Gould AMI

Intel

ITT

Microchip Technology

Motorola

National Semiconductor

SEEQ Technology

Texas Instruments

VLSI Technology

WafetScale Integration

Xicor

Other North American Conq>anies

Japanese Companies

Fujitsu

Hitachi

Mitsubishi

NEC

Oki

Ricoh

Rohm

Sharp

Toshiba

Eurofjean Comjsanies

Eurosil

GEC Plessey

Philips

Plessey

SGS-Thomson

322

52

10

99

9

6

5

7

1

1

27

6

75

0

2

22

0

83

19

11

4

11

5

1

0

12

20

99

0

0

17

2

80

131

0

3

25

0

103

289

57

20

8

6

0

75

5

15

1

25

4

44

0

4

25

0

86

12

13

7

12

1

0

3

13

25

127

1

0

23

0

103

63.0

10.2

2.0

1.0

1.4

0.2

19.4

1.8

1.2

0.2

5.3

1.2

14.7

0.0

0.4

4.3

0.0

16.2

3.7

2.2

0.8

2.2

1.0

0.2

0.0

2.3

3.9

19.4

0.0

0.0

3.3

0.4

15.7

55.9

9.9

2.4

1.4

2.0

0.2

15.7

0.0

0.8

0.2

6.0

1.2

11.7

0.2

0.4

3.6

0.2

17.1

3.2

2.0

0.6

2.0

0.6

0.0

0.0

2.6

6.2

26.0

0.0

0.6

5.0

0.0

20.5

24.8

0.2

0.0

4.5

0.0

20.1

16.8

2.3

2.5

1.4

2.3

0.2

0.0

0.6

2.5

4.9

56.4

11.1

3.9

1.6

1.2

0.0

14.6

1.0

2.9

0.2

4.9

0.8

8.6

0.0

0.8

4.9

0.0

(Continued)

©1992 Dataquest Incoipofated July—^Reproduction Prohibited

4S Memories Worldwide

Table 5-5 (Continued) "

Each Company's Factory Revenue from Shipments of MOS Nonvolatile Memory to Europe

(MilUons of U.S. Dollars)

Asia/Pacific Companies

Hyundai

Macronix

Samsung

MA - Not available

NM - Not meaningful

Source: Dataquest Ouly 1992)

Revenue Market Share (%)

1989 1990 1991 1989 1990 1991

7

5

10

1.4

1.0 2.0

0 0

3

0.0 0.0 0.6

1 1

0

0.2 0.2 0.0

6 4 7 1.2 0.8 1.4

i

©1992 Dataquest Incoipoiated July—Reproduction Prohibited

i

World^de MOS Memory Market Share, 1989-1991

Table 5-6

Each Company's Factory Revenue from Shipments of Other MOS Memory to Europe

(Millions o f U.S. £>ollars)

Total Market

1989

15

Revenue

1990

27

1991

35

1989

100.0

North American Companies

Advanced Micro Devices

Cypress Semiconductor

Dallas Semiconductor

Integrated Device Technology

MOSel

15

1

2

1

10

1

17

1

4

1

10

1

27

2

7

4

13

1

100.0

6.7

13.3

6.7

66.7

6.7

European Companies

SGS-Thomson

NA - Not available

NM - Not meaningful

Source: Dataquest 0"ly 199?)

0

0

10

10

8

8

0.0

0.0

1990

100.0

63.0

3.7

14.8

3.7

37.0

3.7

37.0

37.0

49

1991

100.0

77.1

5.7

20.0

11.4

37.1

2.9

22.9

22.9

©1992 Dataquest Incorporated July—^Eteproduction Prohibited

50 Memories Worldwide

Table 5-7

Each Company's Factory Revenue £rom Shipments of Bipolar Memory to Europe

(Millions of U.S. Dollars)

Total Market

1989

71

Revenue

1990

55

1991

43

1989

100.0

Market Share C/o)

1990

100.0

1991

100.0

North American Companies

Advanced Micro Devices

Integrated Device Technology

National Semiconduaor

Raytheon

41

27

0

13

1

31

22

0

5

4

25

21

1

3

0

57.7

38.0

0.0

18.3

1.4

56.4

40.0

0.0

9.1

7.3

58.1

48.8

2.3

7.0

0.0

Japanese Companies

Fujitsu

Hitachi

NEC

European Companies

Philips

NA - Not available

MM - Not meaningful

Source: Dataquest Quly 1992)

18

6

6

6

12

12

14

6

2

6

10

10

10

3

2

5

8

8

25.4

8.5

8.5

8.5

16.9

16.9

25.5

10.9

3.6

10.9

18.2

18.2

23.3

7.0

4.7

11.6

18.6

18.6

©1992 Dataquest Incorporated July—Reproduction Prohibited

Worldwide MOS Memory Market Share, 1989-1991

51

Table 6-1

Each Company's Factory Revenue from Shipments of MOS Memory t o Asia/Paclfiic-Rest of World

(Millions of U.S. Dollars)

Total Market

1989

1,587

Revenue

1990

1,557

1991

1,974

1989

100.0

Market Share (%)

1990

100.0

1991

100.0

North American Companies

Advanced Micro Devices

Atmel

Catalyst

Cypress Semiconduaor

Dallas Semiconductor

Gould AMI

Harris

Integrated Device Technology

Intel

ITT

Microchip Technology

Micron Technology

MOSel

Motorola

NCR

National Semiconduaor

Performance Semiconductor

SEEQ Technology

Texas Instruments

Vitelic

WaferScale Integration

Xicor

Other North American Companies

5

34

1

0

2

1

417

21

6

2

2

25

47

1

0

145

30

0

4

0

58

1

32

0

0

21

37

7

40

1

402

30

8

5

2

0

2

2

14

52

31

1

0

113

33

0

2

1

465

37

7

2

4

4

0

0

9

59

2

9

53

21

56

0

32

1

1

120

39

3

5

1

26.3

1.3

0.4

0.1

0.1

0.0

0.1

0.1

0.3

2.1

0.1

1.6

3.0

0.1

3.7

0.1

2.0

0.0

0.0

9.1

1.9

0.0

0.3

0.0

0.1

0.1

6.1

2.0

0.2

0.3

0.1

0.5

2.7

1.1

2.8

0.0

1.6

23.6

1.9

0.4

0.1

0.2

0.2

0.0

0.0

0.5

3.0

0.1

25.8

1.9

0.5

0.3

0.1

0.0

0.1

0.1

0.9

3.3

0.0

1.3

2.4

0.4

0.0

7.3

2.1

0.0

0.1

0.1

2.6

0.1

2.0

0.1

Japanese Companies

Fujitsu

Hitachi

Matsushita

Mitsubishi

NEC

NMB Semiconductor

Oki

Ricoh

Rohm

Sanyo

Seiko Epson

Sharp

596

91

82

11

95

66

21

29

1

0

9

0

28

540

87

75

11

128

56

31

40

0

2

15

0

31

664

101

100

18

89

73

24

56

0

5

13

1

31

1.3

1.8

0.1

0.0

0.6

0.0

1.8

37.6

5.7

5.2

0.7

6.0

4.2

34.7

5.6

4.8

0.7

8.2

3.6

2.0

2.6

0.0

0.1

1.0

0.0

2.0

0.3

0.7

0.1

1.6

CContinued)

33.6

5.1

5.1

0.9

4.5

3.7

1.2

2.8

0.0

©1992 Dataquest Incorporated July—Reproduction Prohibited

52

IMemorles Worldwide

Table 6-1 (Continued)

Each Company's Factory Revenue from Shipments of MOS Memory to Asia/Padflic-Rest of World

(Millions of U.S. Dollars)

Sony

Toshiba

Other Japanese Companies

1 9 8 9

6

62

95

R e v e n u e

1 9 9 0

13

51

0

1 9 9 1

14

139

0

1 9 8 9

0.4

3.9

6.0

Market S h a r e (%)

1 9 9 0

1 9 9 1

0.8

3.3

0.0

0.7

7.0

0.0

European Companies

Matra MHS

Philips

SGS-Thomson

Siemens

Asia/Pacific Companies

Goldstar

Hualon Microelectronics Corp.

Hyundai

Macronix

Samsung

Silicon Integrated Systems

United Microelectronics

Winbond Electronics

NA - Not available

NM - Not meaningful

Source: Dataquest Quly 1992)

503

61

NA

78

1

316

NA

47

NA

71

0

6

40

25

523

316

17

30

12

51

39

57

1

92

1

22

45

24

751

134

27

125

21

358

15

49

22

94

1

24

45

24

31.7

3.8

NA

4.9

0.1

19.9

NA

3.0

NA

4.5

0.0

0.4

2.5

1.6

33.6

3.3

2.5

3.7

0.1

20.3

1.1

1.9

0.8

5.9

0.1

1.4

2.9

1.5

38.0

6.8

1.4

6.3

1.1

18.1

0.8

2.5

1.1

4.8

0.1

1.2

2.3

1.2

©1992 Dataquest Incoipoiated Juljf—Reproduction Prohibited

Worldwide MOS Memory Market Share, 1989-1991

53

Table 6-2

Each Company's Factory Revenue from Shipments of MOS DRAMs to Asla/Padfic-Rest of World

(MlUlons of U.S. Dollars)

Total Market

1989

1,074

Revenue

1990

949

1991

1,228

1989

100.0

Market Share (p/6)

1990

100.0

1991

100.0

North American Companies

Intel

Micron Technology

MOSel

Motorola

Texas Instruments

Vitelic

Japanese Companies

Fujitsu

Hitachi

Matsushita

Mitsubishi

NEC

NMB Semiconduaor

Oki

Sanyo

Sharp

Toshiba

Other Jafjanese Companies

European Comjjanies

Siemens

Asia/Pacific Companies

Goldstar

Hyundai

Samsung

NA - Not available

NM = Not meaningful

Source: Dataquest (July 1992)

256

7

46

0

45

130

28

437

59

43

4

83

56

21

24

0

6

46

95

10

10

371

44

62

265

193

12

31

0

26

94

30

374

56

40

4

121

46

31

36

3

3

34

0

10

10

372

42

46

284

231

10

40

7

41

99

34

469

65

52

12

76

57

24

55

3

3

122

0

14

14

514

127

99

288

23.8

0.7

4.3

0.0

4.2

12.1

2.6

40.7

5.5

4.0

0.4

7.7

5.2

2.0

2.2

0.0

0.6

4.3

8.8

0.9

0.9

34.5

4.1

5.8

24.7

39.4

3.3

3.8

0.3

0.3

3.6

0.0

5.9

4.2

0.4

12.8

4.8

20.3

1.3

3.3

0.0

2.7

9.9

3.2

1.1

1.1

39.2

4.4

4.8

29.9

38.2

5.3

4.2

1.0

6.2

4.6

2.0

4.5

0.2

0.2

9.9

0.0

1.1

1.1

18.8

0.8

3.3

0.6

3.3

8.1

2.8

41.9

10.3

8.1

23.5

©1992 Dataquest Incorporated July—Reproduction Prohibited

54 Memories Worldwide

Table 6-3

Each Company's Factory Revenue from Shipments of MOS SRAMs to Asla/Padfic-Rest of World

(MilUons of U.S. Dollars)

Total Market

1989

219

Revenue

1990

207

1991

229

1989

100.0

1990

100.0

North American Companies

Advanced Micro Devices

Cypress Semiconductor

Harris

Integrated Device Technology

Micron Technology

MOSel

Motorola

Performance Semiconduaor

Vltelic

Other North American Companies

26

3

2

1

4

1

1

12

0

2

0

13

6

1

14

1

3

0

42

0

2

2

45

0

4

0

3

13

6

12

1

5

1

11.9

1.4

0.9

0.5

1.8

0.5

0.5

5.5

0.0

0.9

0.0

20.3

0.0

1.0

1.0

6.3

2.9

0.5

6.8

0.5

1.4

0.0

Japanese Companies

Fujitsu

Hitachi

Matsushita

Mitsubishi

NEC

Oki

Rohm

Sanyo

Seiko Epson

Sharp

Sony

Toshiba

European Companies

Matra MHS

PhUips

SGS-Thomson

80

20

24

0

8

4

3

0

9

0

0

6

6

4

0

0

4

88

20

22

1

6

5

3

1

10

0

1

12

7

6

1

1

4

103

22

30

1

12

6

13

4

2

1.4

0.0

4.1

0.0

0.0

2.7

2.7

36.5

9.1

11.0

0.0

3.7

1.8

1.8

0.0

0.0

1.8

1991

100.0

42.5

9.7

10.6

0.5

2.9

2.4

1.4

0.5

4.8

0.0

0.5

5.8

3.4

2.9

0.5

0.5

1.9

3.1

0.4

1.7

0.9

(Continued)

3.5

0.4

0.4

5.7

3.1

45.0

9.6

13.1

0.4

5.2

2.6

0.4

0.4

19.7

0.0

1.7

0.0

1.3

5.7

2.6

5.2

0.4

2.2

0.4

©1992 Dataquest Incorporated July—Reproduction Prohibited

Worldwide MOS Memory Market Share, 1989-1991

55

Table 6-3 (Continued)

Each Company's Factory Revenue from Shipments of MOS SRAMs to Asia/Padfic-Rest of World

(MilUons of U.S. Dollars)

Asia/Padfic Companies

Goldstar

Hualon Microelectronics Corp.

Hyundai

Samsung

Silicon Integrated Systems

United Microelectronics

Winbond Electronics

NA - Not available

NM - Not meaningful

Source: Dataquest Quly 1992)

1989

109

9

NA

15

40

NA

45

NA

R e v e n u e

1 9 9 0

71

4

10

10

16

2

28

1

1 9 9 1

74

3

10

21

16

2

16

6

1 9 8 9

49.8

4.1

NA

Market S h a r e (%)

1 9 9 0

1 9 9 1

34.3

1.9

4.8

32.3

1.3

4.4

6.8

18.3

NA

4.8

7.7

1.0

9.2

7.0

20.5

NA

13.5

0.5

0.9

7.0

2.6

©1992 Dataquest Incoipotated July—^Reproduction Prohibited

56

Table 6-4

Each Company's Factory Revenue from Shipments of MOS EPROMs to Asia/Pacific-Rest of World

(MilUons of U.S. Dollars)

Total Market

1989

NA

Revenue

1990

NA

1991

203

Market Share (%)

1989 1990 1991

NA NA

100.0

North American Companies

Advanced Micro Devices

Atmel

Intel

Microchip Technology

National Semiconduaor

Texas Instruments

WaferScale Integration

Memories Worldwide

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

137

36

6

44

5

25

18

3

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

67.5

17.7

3.0

21.7

2.5

12.3

8.9

1.5

Japanese Conpanies

Fujitsu

Hitachi

Mitsubishi

NEC

Toshiba

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

14

3

7

1

2

1

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

NA

6.9

1.5

3.4

0.5

1.0

0.5

European Companies

Philips

SGS-Thomson

NA. " Not a^^dlable

NM •" Not meaningful

Source: Dataquest Qxily 1992)

NA

NA

NA

NA

NA

NA

52

20

32

NA

NA

NA

NA

NA

NA

25.6

9.9

15.8

©1992 Dataquest Incorporated July^-Reproduction Prohibited

World^vide MOS Memory Market Share, 1989-1991

Table 6-5

Each Company's Factory Revenue from Shipments of MOS Nonrolatile Memory to

Asia/Padfic-Rest of World

(Millions of U.S. Dollars)

Total Market

1989

277

Revenue

1990

385

1991

496

1989

100.0

North American Companies

Advanced Micro Devices

Atmel

Catalyst

Gould AMI

Intel

ITT

Microchip Technology

MOSel

Motorola

National Semiconductor

SEEQ Technology

Texas Instruments

WaferScale Integration

Xicor

Other North American Companies

133

18

6

2

2

27

1

25

0

1

32

0

15

0

4

0

165

30

8

5

2

40

0

21

6

0

31

0

19

0

2

1

0.0

0.4

11.6

0.0

5.4

0.0

1.4

0.0

48.0

6.5

2.2

0.7

0.7

9.7

0.4

9.0

Japanese Companies

Fujitsu

Hitachi

Matsushita

Mitsubishi

NEC

Oki

Ricoh

Rohm

Sanyo

Sharp

Sony

Toshiba

79

12

15

7

4

6

2

0

10

1

0

0

22

78

11

13

6

1

5

1

0

1

2

27

1

10

9

8

3

32

1

21

3

5

0

178

36

7

2

0

49

2

92

14

18

10

0

5

1

27

1

10

0

4

2

28.5

4.3

5.4

2.5

1.4

2.2

0.7

0.4

0.0

0.0

7.9

0.0

3.6

European Companies

Philips

SGS-Thomson

42

6

36

62

21

41

63

20

43

15.2

2.2

13.0

57

1990

100.0

1991

100.0

42.9

7.8

2.1

1.3

0.5

10.4

0.0

5.5

1.6

0.0

8.1

0.0

4.9

0.0

0.5

0.3

20.3

2.9

3.4

1.6

0.3

1.3

0.3

0.0

0.3

0.5

7.0

0.3

2.6

16.1

5.5

10.6

12.7

4.0

8.7

(Continued)

6.5

0.2

4.2

0.6

1.0

0.0

9.9

0.4

1.8

1.6

0.6

35.9

7.3

1.4

0.4

0.0

0.0

0.8

0.4

5.4

0.2

2.0

18.5

2.8

3.6

1.0

0.2

2.0

0.0

©1992 Dataquest Incorporated July—^Reproduction Prohibited

58

Memories Worldwide

Table 6-5 (Continued)

Each Company's Factory Revenue from Shipments of MOS Nonvolatile Memory to

Asia/Pacific-Rest of World

(Millions of U.S. Dollars)

Asia/Pacific Companies

Goldstar

Hualon Microelectronics Corp.

Hyundai

Macronix

Samsung

Silicon Integrated Systems

United Microelectronics

Winbond Sectronics

NA - Not available

NM - Not meaningful

Source: Dataquest Qaty 1992)

1989

23

8

NA

1

1

11

NA

2

NA

Revenue

1990

80

5

29

1

1

16

15

2

11

1991

163

4

17

5

21

54

13

33

16

1989

8.3

2.9

NA

Market Share (%)

1990

1991

20.8

1.3

7.5

32.9

0.8

3.4

0.4

0.4

4.0

NA

0.7

NA

0.3

0.3

4.2

3.9

0.5

2.9

1.0

4.2

10.9

2.6

6.7

3.2

I

©1992 Dataquest Incorporated July—Reproduction Prohibited

I

>

Worldwide MOS Memory Market Share, 1989-1991

59

Table 6-6

Each Company's Factory Revenue from Shipments of Other MOS Memory to Asia/Pacific-Rest of World

(Millions of U.S. Dollars)

Total Market

19S9

17

Revenue

1990

16

1991

21

19S9

100.0

1990

100.0

1991

100.0

North American Companies

Advanced Micro Devices

Dallas Semiconductor

Integrated Device Technology

NCR

European Companies

Siemens

NA - Not available

NM - Not meaningful

Source: Dataquest (July 1992)

2

0

0

1

1

15

15

2

0

0

1

1

14

14

11

1

4

6

0

10

10

11.8

0.0

0.0

5.9

5.9

88.2

88.2

12.5

0.0

0.0

6.3

6.3

87.5

87.5

52.4

4.8

19.0

28.6

0.0

47.6

47.6

I

^

©1992 Dataquest Incorporated July—^Reproduction Prohibited

6 0

Table 6-7

Each Company's Factory Revenue from Shipments of Bipolar Memory to Asia/Pacific-Rest of World

(MiUions of U.S. Dollars)

Total Market

1989

18

Revenue

1990

22

1991

17

Market Share (%)

1989 1990 1991

100.0 100.0 100.0

North American Companies

Advanced Micro Devices

Motorola

National Semiconduaor

Texas Instruments

Memories Worldwide

5

2

0

1

2

7

2

0

3

2

5

2

1

1

1

27.8

11.1

0.0

5.6

11.1

31.8

9.1

0.0

13.6

9.1

29.4

11.8

5.9

5.9

5.9

Japanese Con^panies

Fujitsu

Hitachi

European Companies

Philips

SGS-Thomson

NA > Not available

NM - Not meaningM

Source: Dataquest O^Iy 1992)

7

1

6

6

6

0

8

7

1

7

1

6

8

8

0

4

0

4

38.9

5.6

33.3

33.3

33.3

0.0

31.8

4.5

27.3

36.4

31.8

4.5

23.5

0.0

23.5

47.1

47.1

0.0

©1992 Dataquest Incorporated July—Reproduction Prohibited

• Dataquest

^ K Dataquest Research and Sales Offices:

Dataquest Incorporated

1290 Ridder Park Drive

San Jose, California 95131-2398

United States

Phone: 01-408-437-8000

Facsimile: 01-408-437-0292

Dataquest Incorporated

Dataquest/Ledgeway

550 Gochituate Road

Framingham, Massachusetts 01701-9324

United States

Phone: 01-508-370-5555

^ ^ ^ ^ Facsimile: 01-508-370-6262

^ ^ ^ ^ B Dataquest Incorporated Invitational

^ ^ ^ ^ K Computer Conferences Division

^ ^ ^ ^ B 3151 Airway Avenue, G-2

^ ^ ^ B - Costa Mesa, California 92626

^ ^ ^ ^ K United States

^ ^ ^ K Phone: 01-714-957-0171

^ ^ ^ K Facsimile: 01-714-957-0903

^ ^ ^ ^ B Dataquest Australia

^ ^ ^ ^ K Suite 1, Century Plaza

^ ^ ^ K 80 Berry Street

^ ^ ^ 1 North Sydney, NSW 2060

^ ^ ^ ^ K Australia

^ ^ ^ ^ Phone: 61-2-959-4544

^ F Facsimile: 61-2-929-0635

^ K Dataquest Europe Limited

^ B - Roussel House, Broadwater Park

^ V Denham, Uxbridge

^f Middlesex UB9 5HP

^ B England

^ f e Phone: 44-895-835050

^M' Facsimile: 44-895-835260/1

^ B Dataquest Europe SA

^ K Tour Galli^ni 2

^ K ^ ^ 36, avenue du General-de-Gaulle

^ ^ ^ k 93175 Bagnolet Cedex

^ ^ ^ ^ B France

^ ^ • [ Phone: 33-1-48-97-3100

^ ^ ^ B Facsimile: 33-1-48-97-3400

^^^^m Dataquest GmbH

^ ^ ^ ^ H Kronstadter Strasse 9

^ ^ ^ B 8000 Munich 80

^ ^ ^ ^ ^ Germany

^ • ^ ^ Phone:49-89-930-9090

^ ^ ^ Facsimile: 49-89-930-3277

Dataquest Germaqy

In der Schneithohl 17

6242 Kronberg 2

Germany

Phone: 49-6173/61685

Facsimile: 49-6173/67901

Dataquest Hong Kong

Rm. 4A01

HKPC Building

78 Tat Chee Avenue

Kowloon, Hong Kong

Phone: 852-788-5432

Facsimile: 8525788-5433

Dataquest Japan Limited

Shinkawa Sanko Building

1-3-17 Shinkawa, Ghuo-ku

Tokyo, 104

Japan

Phone: 81-3-5566-0411

Facsimile: 81-3-5566-0425

Dataquest Korea

Daeheung Building 1105

648-23 Yeoksam-dong

Kangnam-gu

Seoul 135-080, Korea

Phone: 82-2-556-4166

Facsimile: 82-2-552-2661

Dataquest Singapore

4012 Ang Mo Kio Industrial Park 1

Ave. 10, #03-10 to #03-12

Singapore 2056

Phone: 65-4597181

Telex: 38257

Facsimile: 65-4563129

Dataquest Taiwan

Room 801/8th Floor

Ever Spring Building

147,Sec.2,ChienKuoN.Rd.

Taipei, Taiwan R.O.C. 104

Phone: 886-2-501-7960

886-2-501-5592

Facsimile: 886-2-505-4265

Dataquest Thailand

300/31 Raohdapisek Road

Bangkok 10310

Thailand

Phone: 66-2-275-1904/5

66-2-277-8850

Facsimile: 66-2-275-7005

1

0012973

1

1

1

I

DataQuest'

emories Worldwide

MMRY-SEG-UW-9201

December 28, 1992

IVIemories Worldwide

User Wants and Needs

\

DataQuest'

Memories Worldwide

MMRY-SEG-UW-9201

December 28, 1992

i i

Published by Dataquest Incorporated

The content of this report represents our interpretation and analysis of information generally available to tiie public or released by knowledgeable individuals in the subject industry, but is not guaranteed as to accuracy or completeness. It does not contain material provided to us in confidence by our dients.

Printed in the United States of America. All rights reserved. No part of this publication may be reproduced, stored in retrieval systems, or transmitted, in any form or by any means—mechanical, electronic, photocopjong, di^licating, microfilming, videotape, or otherwise—^without the prior permission of the publisher.

© 1992 Dataquest Incorporated

December 1992

0014238

i

>

Table of Contents

Page

1. Executive Summary 1-1

2. Methodology 2-1

3. Applications Types 3-1

Automotive, Consumer 3-2

Data Processing 3-2

Instrumentation and Test, and Industrial Control and Monitoring 3-6

Military/Aerospace 3-8

Office Eqtdpment 3-9

Telecommimications 3-9

^ Usage Trends 4-1

By Density 4-3

16Kb 4-3

64Kb 4-3

256Kb 4-4

1Mb 4-4

4Mb 4-5

List of Figures

Figure Page

2-1 Revenue of Responding Corporations 2-2

2-2 Number of Employees at This Location 2-2

3-1 Percent of Responses, by Application Type 3-12

3-2 North American Electronic Equipment Production, 1989-1996 3-12

3-3 SRAM As a Percent of All MOS Purchases, by Application 3-13

3-4 Density Preferences: Consimier Electronics Manufacturers 3-14

3-5 SRAM As a Percent of All MOS Purchases, Consumer/Automotive 3-14

3-6 Cost of Electronics in Average U.S. Vehicle, 1989-1996 3-15

3-7 Density Preferences: A u d i o / \ ^ u a l Manufacturers 3-15

3-8 Breakout of Data Processing Responses 3-16

3-9 Worldwide PC Shipments Forecast 3-16

3-10 Density Preferences: Data Processing Manufacturers 3-17

3-11 Worldwide Computer Systems Market Mix 3-17

3-12 Worldwide PC and Personal Workstation Market Mix 3-18

3-13 SRAM As a Percent of Ail MOS Purchases, Data Processing 3-19

3-14 Computer Manufacturers, by Application Type 3-20

3-15 Density Preferences: PC Caches 3-20

3-16 Deiisity Preferences: Workstation Caches 3-21

3-17 Density Preferences: Minicomputer Caches 3-21

3-18 Density Preferences: Mainframe Computer Caches 3-22

3-19 Density Preferences: Main Memory in Deskside Computers 3-22

3-20 I / O Device Manufacturers, by Application 3-23

3-21 Worldwide Disk Drive Production, 1990-1996 3-24

3-22 Density Preferences: Disk Cache Manufacturers 3-24

3-23 North America Page Printer Forecast, 1991-1996 3-25

3-24 Worldwide Display Terminal Production, 1990-1996 3-25

3-25 Density Preferences: CRT Terminal Manufacturers 3-26

3-26 U.S. Network Interface Card Forecast, 1989-1996 3-26

3-27 Density Preferences: LAN Board Manufacturers 3-27

3-28 Density Preferences: Fax/Modem Board Manufacturers 3-27

3-29 SRAM As a Percent of All MOS Purchases, Instrumentation/Industrial 3-28

3-30 Density Preferences: Iiistrumentation Test Manufacturers 3-28

3-31 Instrumentation Test Manufacturers, by Application 3-29

3-32 Density Preferences: Digital Storage Oscilloscopes/Logical Analyzers 3-29

3-33 Density Preferences: Battery-Operated Iiistruments 3-30

3-34 Density Preferences: Medical Instrumentation 3-30

3-35 Density Preferences: Remote Measurement Equipment 3-31

List of Figures (Continued)

Figure Page

3-36 North American Industrial Electronics Production 3-31

3-37 Density Preferences: Industrial Control and Monitoring Manufacturers 3-32

3-38 U.S. Defense Budget Procurements, 1991-1996 3-32

3-39 North American Military and Civil Aerospace Electronics Production 3-33

3-40 SRAM As a Percent of AU MOS Purchases, Military/Aerospace 3-33

3-41 Density Preferences: Military/Aerospace Manufacturers 3-34

3-42 Military Aerospace Manufacturers, by Application 3-34

3-43 Density Preferences: Radar Mantifacturers 3-35

3-44 Density Preferences: Navigational Equipment Manufacturers 3-35

3-45 Density Preferences: Satellite Manufacturers 3-36

3-46 Density Preferences: Office Equipment Manufacturers 3-36

3-47 U.S. Telecom System Shipments, 1987-1996 3-37

3-48 SRAM As a Percent of All MOS Purchases, Telecommimications 3-38

3-49 Density Preferences: Telecommimications Manufacturers 3-38

3-50 Telecommimications Equipment Manufacturers, by Application 3-39

3-51 Density Preferences: PBX Switch Manufacturers 3-39

3-52 Density Preferences: Central Office Digital Switching Equipment 3-40

3-53 Density Preferences: Voice/Data Terminals 3-40

4-1 Density Preferences: AU Manufacturers 4-6

4-2 Speed Preference, by Device Density 4-6

4-3 SRAM Usage Expectations for Next Year 4-7

4-4 Responses Anticipating No Change, by Device Density 4-7

4-5 Responses Anticipating Use Increase, by Device Density 4-8

4-6 Expected Unit Consumption Increase Next Year 4-8

4-7 Responses Anticipating Use Decrease, by Device Density 4-9

4-8 Expected Unit Consumption Decrease Next Year 4-9

4-9 Responses Anticipating Different Device Use, by Device Density 4-10

4-10 New Part Plaimed for Next Year 4-10

4-11 Anticipated Migration to New Part 4-11

4-12 Responses Expecting to Upgrade Speed, by Device Derisity 4-11

4-13 Responses Anticipating Application Phaseout, by Device Density 4-12

4-14 Volume of 16Kb Purchases 4-12

4-15 Package Preference for 16Kb SRAMs 4-13

4-16 Volume of 64Kb Purchases 4-13

4-17 Package Preference for 64Kb SRAMs 4-14

4-18 Volume of 256Kb Purchases 4-14

4-19 Package Preference for 256Kb SRAMs 4-15

i

List of Figures (Continued)

Figure Page

4-20 Volume of 1Mb Purchases 4-15

4-21 Package Preference for 1Mb SRAMs 4-16

i i

List of Tables

Table Page

3-1 Worldwide Computer Systems Forecast, Unit Shipments 3-11

Chapter 1

Executive Summary

This report is a snapshot of North American static RAM (SRAM) usage during the year 1992. Through the efforts of numerous interviewers, analysts, and with the help of SRAM manufacturers and users, Dataquest has compiled a list of applications of SRAMs that draws trends out of each application and helps improve understanding of the issues used to determine which SRAM is to be used.

Some of the broader trends to emerge from this survey are as follows:

• Most designs are expected to use the same device next year as used this year.

• Of those comparues planning to upgrade the device they use, the majority plan to upgrade to the next higher density above the density they currently purchase.

• The plastic dual-in-line package is still preferred by a majority of companies. However, it is not preferred in the highest volume applications.

• The 256JCb density is preferred by a majority of companies, followed by botti the 64Kb and 1Mb devices, which are on nearly equal footing.

MMRY-SEG-UW-9201 ©1992 Dataquest Incorporated December 28,1992

Chapter 2

Methodology

A three-pronged approach was used in the compilation of this report.

First, SRAM manufacturers were interviewed about the major North

American applications of SRAMs from their own viewpoint. Many contributed names of major users, device preferences, and estunated usage. Second, Dataquest interviewed by telephone more than

200 SRAM buyers, and asked about their end applications, speed and deiisity usage, package preferences, and projections of future usage.

Last, the resultant data were taken back to certain SRAM manufacturers for a "sanity" check.

A statistical rather than rigorous approach was followed in the user telephone interviews. With certain exceptions, each respondent was asked to answer only about the single application that used the most significant dollar amount of SRAMs, and then was asked only to answer about the most significant SRAM used in the design. Although this approach probably caused us to overlook several applications going on in the same facility at the same time, or to overlook different types of SRAMs that would be used together in a specific application, it allowed us to gamer a wider variety of users, because long questioimaires are patently unpopular. This approach gave us a sampling that we believe is statistically significant.

Exceptions were made when deabng with multidivisional companies that used corporate procurement offices, offices that procured ail

SRAMs for all projects from a single office. These companies were questioned about their five most significant SRAM uses, and all of the

SRAMs used in these applications.

Where appropriate, information from other groups within Dataquest is presented to show the growth or decline of the end markets for each application examined.

Figure 2-1 shows corporate revenue of the respondents. Figure 2-2 shows employee head counts. The average company surveyed had an employee head count of about 550 and average armual revenue of about $150 million.

MMRY-SEG-UW-9201 ©1992 Dataquest Incorporated December 28,1992

VI

Figure 2-1

Revenue of Responding Corporations

Memories Woridwide

Corporate Revenue ($)

Source: Dataquest (December 1992)

Figure 2-2

Number of Employees at This Location

Less than 100,000

^ n

100.000-499,999

El

500,000 - 999,999

n

1 Million - 4.999 Million

5 Million - 9.999 Million

^ m

10 Million-49.999 Million

M

50 Million - 99.999 Million

100 Million-499.999 Million

U

500 Million+

El

Decline Refusal

G200312a

1-9

^

m

10-20

u

21-50

u

51-100

@

101-500

m

501-1,000

1,000+

I

i

Employees at Site

Source: Dataquest (December 1992)

December 28,1992

©1992 Dataquest Incorporated

GZQCQ-)29

MMRY-SEG-UW-9201

i

Chapter 3

Applications Types

After discussions with several manufacturers of SRAMs, and based upon data resident within Dataquest, the survey was written to focus on a list of nine major applications groups: audio/visual, consumer electrorucs, data processing (including everything from palmtop computers through corporate mainframes), iiistrumentation and test, military and aerospace electronics, office equipment (not including personal computers and telecommimications equipment), hand-held devices that do not fit within any of the other categories mentioned in this list, industrial control and monitoring, and telecommunications equipment.

Each major category was broken into subcategories wherever appropriate, and in some cases, these subcategories were broken into further groups. As an example, data processing contains a subcategory of computers and PCs. Applications that fall into these categories include register storage, caches of all types, and main memory in certain designs. Figure 3-1 shows the percent of responses received in each top-level applications category.

The categories listed nearly correlate to the six standard semiconductor markets used in Dataquest's electronic equipment production as reported by Dataquest's Semiconductor Applications and Markets

(SAM) group in its MarketTrends: Electronic Equipment (publication ntimber SAWW-SVC-MT-9201). These markets are: data processing, communication, industrial, consumer, military/aerospace, and transportation. Forecasts for these markets are in Figure 3-2, and can be used with the data in the following sections to help to forecast trends for distinct devices.

In Dataquest's 1993 Semiconductor Procurement Insights User Wants and

Needs report (publication niimber SPWW-SVC-UW-9202), a different list of applicatioris was surveyed and asked to rank their SRAM purchases as a percent of their overall MOS purchases for 1992. Applications investigated in this survey were: personal computers, other data processing, premise commimications, public telecoirununications, instnmaentation and test, consumer and automotive, and military/ aerospace. Figure 3-3 shows an overview of all responses; these data wiU be broken out into a more readable form in each of the following sections.

MMRY-SEG-UW-9201 ©1992 Dataquest Incorporated December 28,1992

3-2 Memories Worldwide

Automotive, Consumer

The weakest response to the survey came from the automotive and consumer electronics sector. This comes as little surprise, because there is not much North American activity in the consumer electronics sector and because automotive electronics currently use SRAM only as a fraction of an existing (controller) semiconductor device. The North

American region does not compete weU in the consumer global market, and has let the predominant portion of it fall into the hands of

Japan and the Asia/Pacific covintries.

Of the few respondents who replied that they were involved in consumer electronics, device preferences centered mostly aroimd slow

256BCb SRAMs (see Figure 3-4). There was also limited interest in 4Mb devices, although unit volume purchases were almost nonexistent.

SRAMs do not compete well in North American consumer and automotive applications for overall MOS dollar expenditure levels

(see Figure 3-5). According to the Semiconductor Buyer Perceptions survey, 40 percent of combined automotive and corisimner respondents answered that they would purchase from zero to 9 percent of their

MOS budget in SRAM, and another 40 percent put their expenditure at 10 to 19 percent. The remainder placed their percentage at 90 percent and above, a figure that probably reflects the disparity in the markets served by different sorts of equipment (that is, engine controllers versus video games).

Further illustration of this can be found in Figure 3-6, which is a forecast of the cost of the electronics (including audio electronics) of the average U.S.-built vehicle from 1989 to 1996. In the narrow price band

Dataquest has forecast, there is litfle room to design less-integrated systems around costly, and possibly nonessential devices such as

SRAMs.

One interesting twist is that our survey imcovered a less-known market to SRAM manufacturers, one we will call "audio/visual."

Typical respondents manufacture specialized digital audio processing equipment and imaging processors for broadcast television. Density preferences in this group, which are shown in Figure 3-7, are spotty, with interviewees answering to a widespread usage of 16Kb, 256Kb, and 4Mb devices (the last being used in prototype quantities to date).

Widest usage is of ihe 32Kx8 at 100ns in plastic DIP, averaging 13,000 units per year per purchasing organization.

Data Processing

The data processing category has been broken into two levels of subcategories because of the widespread use of SRAMs in every aspect of data processing, including main memories, modem boards, LAN controllers, CPU caches, hard disks, terminals, and many other related devices. The largest volume (58 percent) of responses came from the computers and PCs section of the market, with I / O devices following at 38 percent (see Figure 3-8). Less than 4 percent of the responses

December 28.1992 ©1992 Dataquest Incorporated MMRY-SEG-UW-9201

Applications Types 3-3 came from manufacturers that claimed that their major application of

SRAMs was in any other category. This seems a bit peculiar in light of the expected growth of the laptop and notebook computer areas (see

Figure 3-9), but is probably more because of a lack of widespread use of SRAMs in these applications.

Figure 3-9 is an updated version of a figure used in a December 23,

1991 article entitled "On the Verge of 3 Volts" in the Semiconductors

Worldwide: Products, Markets, and Technologies Dataquest Perspective,

Vol. 1, No. 5. Certain pundits believe that power consumption can be reduced in battery-operated PCs t h r o u ^ the addition of cache memories, but the verdict has not yet been returned on this question.

It appears that few PC designers attempt to use cache memories as a power-saving device.

Before delving into the details, we wUl take a top-level look at the data processing market. All five of the most widespread SRAMs are represented by one respondent or another to be the most important density (from an expenditure perspective) in use at their facility.

Naturally, the 4Mb and 16Kb densities garnered the fewest responses, and the 256Kb density took the largest number (see Figure 3-10).

When we examine the market more closely, though, we see that the use of denser devices parallels the use of slower access times, and vice versa.

Table 3-1 scales to the overall size of the worldwide total available market for SRAMs. The numbers in this table are derived from data from Dataquest's Computer and Peripheral Systems group. Shifts in this market, and in the PC and personal workstation market, are shown in Figures 3-11 and 3-12.

Figure 3-13 shows SRAM as a percent of overall MOS purchases in a piece of data processing eqtiipment. It is interesting to note that, while all other data processing responses are strikingly divided, all of the PC manufacturers polled placed their SRAM purchases at 10 to 19 percent of their overall board purchases. This meshes well with the fact that a typical cache in a 486-based PC is implemented using $10 to $25 worth of SRAM, while the CPU is priced at about $100 to $200. The split in the "other data processing" category between 39 percent and

60 percent of system cost most probably results in the breadth of this category, which includes workstations and computers on one end, and keyboards and I / O cards on the other.

Figure 3-14 shows the number of respondents whose major end use of

SRAMs fell into one of a number of subcategories in the data processing category. The largest number of responses (39 percent) was by manufacturers of desktop PCs, whose major application for SRAMs was as a cache in those PCs. This was followed by an equal nvimber of responses (14 percent) by manufacturers using SRAM as cache in workstations and minicomputers, or as main memory in deskside computers. About half as many said that caches in laptop computers or mainframe computers were their major SRAM application, and the remainder used SRAM in writable control stores (a dying field) or other data processing applications.

MMRY-SEG-UW-9201 ©1992 Dataquest Incorporated December 28,1992

3-4 Memories Woridwide

For PC caches, a surprisingly large number of respondents still use

16Kb devices, yet the 256Kb and 1Mb densities dominate this market

(see Figtire 3-15). Usage of the 16Kb devices was reported to consist solely of 4Kx4-bit organizations in relatively low volumes, leading us to believe that the response represents orUy the use of 4Kx4 cache-tags, along with a possibility that the users of these devices are paying more per system for these narrowly sourced devices than ttxey are for the cache data SRAMs, a plausible scenario given the last year's price drops for 64Kb and 256Kb SRAMs. Although 4Mb densities were reported by a segment of the respondents, unit volumes were low enough to support only protot5rpe quantities.

Workstation caches reported a more balanced mix of 16Kb and 64Kb devices (see Figure 3-16), with the 256Kb totally missing from the sample (as the most important SRAM purchased for the application) and

1Mb and 4Mb devices consuming the lion's share of the responses. We believe that the use of higher-density parts is indicative of the trend of certain workstation manufacturers to try to pack the most waUop into their machines, while others (for example. Precision Architecture machines) are so hell-bent for speed that issues of SRAM density fade in comparison. On the other side of the argument, 1Mb SRAMs are used by some respondents in more modest speeds, but in relatively high volvimes averaging 63,000 units per year. Once again, the 4Mb density is used only in prototype quantities by the respondents questioned.

Minicomputer caches accotmted for a nearly equivalent portion of the survey as did workstation caches, but showed less divided results (see

Figtire 3-17). No 4Mb usage was revealed, while equivalent 1Mb and

16Kb responses were given to Dataquest. The largest two sectors are the equal niunber of respondents who said they used SRAMs of 64Kb and 256Kb densities, the two densities that made up the majority of parts purchased by a wide margin.

Mainframe computer manufacturers (see Figure 3-18) are a horse of a different color. One-third of our respondents use 256Kb SRAMs, onethird use 1Mb devices, and the last third use 4Mb devices. This proportion stands to reason, because leading-edge SRAM suppliers compete to be the first to supply the 4Mb SRAM to certain supercomputer manufacturers. Still, unit volume implies that, of those who would disclose their voliune usage to Dataquest, the 256Kb makes up about 90 percent of all units used in this application.

The last computer type for which significant information was gained was deskside computers that use SRAM for their main memory. It comes as no surprise that the least expensive high-speed SRAMs, from a price-per-bit viewpoint, were used in the majority of the applications. Figure 3-19 shows that 256Kb SRAMs were the choice of the overwhelming majority of respondents, while the 1Mb and 16Kb devices were the only others to be given as choices for this application. The vast majority of imits used in this application comprise slower 32Kx8s.

December 28,1992 ©1992 Dataquest Incorporated MMRr-SEG-UW-9201

Applications Types 3-5

Of the respondents in the data processing category who stated that their major SRAM application was in I / O devices, the number of responses does not closely correlate to the volume usage of SRAMs in the design categories (see Figure 3-20). One important application, disk drive caches, elicited relatively few responses, even though it accounts for an important part of the North American SRAM market. For this reason, the discussion will detail points not immediately obvious from the graphics.

Figure 3-21 shows Dataquest's worldwide disk drive production imit forecast, as reported by Dataquest's Computer and Peripheral Systems group. This market is the epitome of the global industry, where design and limited manufacture is performed in first-world nations, with the majority of production happerung in developing countries such as

Thailand and Singapore. As a result, the location of buys is variable, but is most often outside of North America, even though the majority of businesses are headquartered in the Uiuted States. Because this survey only covers North American purchases, volumes are deceptively low. The trends in Figure 3-21 show that rapid growth is occurring in

3.5-, 2.5-, and 1.8-inch disk drives. Caches in these drives serve two purposes: If used with a desktop computer, or other computer with tinlimited power availability, the SRAM disk drive cache is simply used to improve apparent latency. If a cached disk drive is used in a limited (battery)-power application, the disk is powered down when not in use, and the cache serves to allow the disk not to power up in about 30 percent of the attempted accesses by the computer.

Popular densities for disk drives are 1Mb and 256Kb densities (see

Figure 3-22), usually at speeds of 70 to 100ns, and always in 8-bit widths. Surface-mount devices were the exclusive choice of the respondents. This is a fast moving market, and one significant manufacturer claimed to be using more 64Kb devices than 256Kb devices six months before the study was performed, and not to be using any at the time of the study. In this light, and given the fact that at least one disk drive manufacturer now uses DRAMs instead of

SRAMs in this application, it would not be surprising to see a rapid abandonment of the 256Kb density in favor of the 1Mb (to occur after this report is published, it is hoped), and an eventual total abandonment of SRAMs in favor of wide-word DRAMs by all disk drive manufacturers.

Page printers that replicate an entire page at one time comprise both laser printers and LED printers. Dataquest's imit production forecast for page printers manufactured in North America is shown in

Figure 3-23, a forecast regularly provided to subscribers of Dataquest's

Document and Imaging Service. Typical SRAM usage in page printers is for cache memories and tends to foUow the device types used in

PCs and workstations, because the CPU used in page printers is often similar to those used in PC and workstation applications. It is not unusual for the processing engine used in a page printer to be more powerful than the resources in the PC or workstation that sends the document to the printer.

MMRY-SEG-UW-9201 ©1992 Dataquest Incorporated December 28,1992

3-6 Memories Worldwide

A surprisingly large number of respondents, slightly less than 25 percent of those who used SRAMs primarily in data processing I / O applications, said that CRT terminals were their most important SRAM application. Worldwide unit production of display terminals is regularly forecast by Dataquest's Computer and Peripheral Systems group, and the current version is shown in Figure 3-24. Dataquest expects the slump encountered in 1991 to be relatively long-lived in this market, and for production not to match 1990 levels until 1995. Both 64Kb and

1Mb devices were popular with many respondents (see Figure 3-25), with 16Kb and 256Kb densities used, but by not as many respondents.

Figure 3-26 shows the network interface card (NIC) unit forecast from

Dataquest's Telecoimnunications service. SRAMs are used as buffers in

LAN cards, and as a result do not need to be too large. T5^ically, only one or two SRAMs are used per card, if any are used at all. Alternatives include first-in/first-out memories (FIFOs) as well as certain lower-performance software techniques.

One-third of the respondents each used 256Kb, 16Kb, and 64Kb

SRAMs (see Figure 3-27). Unit volvunes ranged from tens of thousands of units to himdreds of thousands for aU those densities represented in the chart.

The final data processing I / O market for which this survey attained meaningful results was fax/modems. Figure 3-28 shows that manufacturers of these boards have a preference for 4Mb SRAMs, with onethird expressing a primary need for 256Kb SRAMs. The major quantities used were the 4Mb device. However, because of a large percentage of nonresponses to the question of organization, Dataquest is led to believe that the main organization used in this application is the

512Kx8 pseudo-SRAM.

A large percentage of the respondents (25 percent) ariswered that they had major data processing I / O SRAM applications that fit into none of the listed categories. Although we tried to find a way to group some of these responses into a new category, they were too far-flung to allow us to accomplish this task.

instrumentation and Test, and Industriai Controi and Monitoring

Figure 3-29 is a detail from Figure 3-3 in Dataquest's Semiconductor

Procurement Insights survey. For the instrumentation and test market and the industrial control and monitoring market, interviewees were asked what percentage of their overall MOS dollar purchases comprised SRAM. Generally, the instrumentation and test field broke into two categories: those with SRAM purchases accounting for less than

30 percent of their overall MOS purchases (about 70 percent) and those with SRAM ptirchases accounting for 50 percent or more of their

MOS purchases (the other 30 percent). Industrial control and monitoring eqmpment manufacturers offered a spread of responses, covering the entire range of zero to 100 percent, with a surprisingly large number responding that SRAM accounted for 90 to 100 percent of their

December 28,1992 ©1992 Dataquest Incorporated MMRy-SEG-UW-9201

Applications Types 3-7 overall MOS purchases. This could imply that SRAMs are used in industrial control and monitoring applications with non-MOS devices, with boards rather than with discrete MOS semiconductors.

Focusing first on the instrumentation and test applications, we see that the respondents were somewhat evenly divided in their use of 64K,

256Kb, and 1Mb devices, with a small portion using 16Kb SRAMs (see

Figure 3-30). Volume usage of the 16Kb is significantly lower than this figure indicates, and volumes for the other densities were generally lower than for other markets, with few respondents answering that their annual imit voliime was in the himdreds of thousands. Speed usage for these manufacturers is widespread, but nearly aU said they use either an x l or x8 organization. We fotmd little use of 4-bit-wide parts in this market. Similarly, strong preferences appeared for DIPs and SOJ/SOIC packages. Few respondents used anything else.

The profile of the instrumentation and test respondents is shown in

Figure 3-31. The strongest showing was in medical instrximentation, which is a broad field, but does not account for a major portion of overall imit sales. Second was the "other" category, in which the responses showed absolutely no overlap.

Respondents who said that their main SRAM application was digital storage oscilloscopes (DSO) or logic analyzers mainly used very fast

256Kb and 1Mb devices, with an element using a small quantity of

ECL I / O 16Kb synchronous SRAMs (see Figure 3-32).

Figure 3-33 shows density preferences of those whose major SRAM application was in battery-operated instruments. Virtues often sought in these applications are wide word width, high integration, small package size, low overall power consumption as measured by low operating current, and low-voltage operation. Half these respondents said that their major expenditure was on 256Kb parts, with 25 percent going to the 1Mb density, and the other quarter to the 64Kb density.

Medical instrumentation manufacturers were more evenly divided in their use of 64Kb, 256Kb, and 1Mb densities, as would be expected given the breadth of this field, and the current point in each of these

SRAM densities' life cycles (see Figure 3-34). Medical applications range from battery-operated pulse-rate recorders to sophisticated imaging systems found in ultrasound, NMR, CAT, and PET scanners.

The next three categories—^integrated circuit testers, system testers, and global positioiung system receivers—elicited responses of insufficient quality to provide useful information about their SRAM usage patterns. There is a growing trend in the fields of IC testers and system testers, however, toward the use of SRAMs with ECL I / O levels in order to support the escalating speed of clocks in today's systems.

Global positioning systems are likely to have their SRAM decisions driven by power consiimption (especially for battery-operated devices), package size, and cost more than by any other criterion.

Respondents whose main SRAM application was stated to be remote monitoring and measurement equipment were divided two-to-one ki

MMRY-SEG-UW-9201 ©1992 Dataquest Incorporated December 28,1992

3-8 Memories Woridwide their use of SRAMs in 64Kb and 256Kb densities (see Figure 3-35). The lack of responses favoring 1Mb SRAMs probably owes to a shortage of suppliers of 1Mb SRAMs in extended temperature ranges, as well as to the fact that much of this remote monitoring equipment has relatively long qualification periods and life cycles.

In the industrial control and moiutoring field, the North American industrial electronics equipment production market is expected to grow to a level about 40 percent larger than its current $35.4 billion level by 1996 (see Figure 3-36). This estimate is generated by Dataquest's Semiconductor Applications and Markets service. SRAM density preferences for respondents who placed themselves in this category were scattered across all available SRAM densities (see

Figure 3-37). Average imit volumes were highest for the 256Kb density, at about 20,000 vmits per year, with other densities selling an average of thousands or himdreds of units per year to any single respondent.

Those who responded that their most important SRAM purchases occurred in the 4Mb density used only a smattering of the product, and used oiily the fuUy static device, not a pseudo-static version. The vast majority of respondents picked the 8-bit width as their preferred device, with packages centering arovmd SOIC/SOJ and plastic DIR

Military/Aerospace

Figure 3-38 is the U.S. Department of Defense's projected procurements budget through 1996. This accoimts for the vast majority of military/aerospace spending in North America. The radical drop in

1993 spending will probably have resounding repercussions through all supporting industries, and is certain to continue to be felt by the electronics industry well after the $62 biUion level is again reached in

1995. Despite reduced North American defense electronics spending, growth in civil aerospace electronics spending wUl help grow the overall market by more than 10 percent (see Figure 3-39 from

Dataquest's Semiconductor Applications and Markets service).

Military and aerospace respondents in the Semiconductor Procurement

Insights survey predominantly put their SRAM purchases as a percentage of overall MOS in the lower two-thirds category, with 42 percent responding that their SRAM expenditure only made u p from zero to

9 percent of their overall MOS expenditure (see Figure 3-40).

Of those who do use SRAMs, the survey found that densities from

64Kb to 4Mb were being used as the major SRAMs, with the predominant number of responses favoring the 64Kb, 256Kb, and 1Mb densities (see Figure 3-41). Average annual volvime per respondent for

64Kb devices was 4,000 vmits, while the 256Kb was at 20,000 and the

1Mb at 13,000. Only a few hundred 4Mb devices were used by all respondents combined, none of these parts being pseudo-SRAMs.

Speeds used covered the available spectrum from 15ns to 150ns. Sixtyfive percent used 8-bit widths, with the balance evenly spHt between

4-bit and 1-bit organizations. A surprisingly large 65 percent used commercial plastic packaging.

December 28,1992 ©1992 Dataquest Incorporated MMRY-SEG-UW-9201

Applications Types 3-9

Applications within the military and aerospace category were split among radar, satellites and satellite support, navigational aids, and other applications that did not overlap into a single sort of application

(see Figure 3-42). A limited response was received from sonar manufacturers.

The overwhelming preference of radar manufacturers responding to the survey was for 1Mb SRAMs (see Figure 3-43). 256Kb devices ranked second. The penchant for using higher densities probably owes to the need to manage numerous data points simultaneously in imaging applications such as this.

Navigational equipment is just the opposite, with the lion's share of the responses favoring the 64Kb density (see Figure 3-44). This stands to reason, because the application reqioires relatively little storage to accomplish its basic task, and the only use for a larger SRAM would be to add discretionary differentiating features such as historical information. As mentioned previously, although a nimiber of respondents named the 4Mb device as the one upon which they spent the most, the unit volumes are tiny.

Satellite manufacturers were the final group of respondents in the military and aerospace category to comprise a sample of significant size. Given that qualification standards are hard to meet for such equipment, it is not surprising to see that the majority of the respondents claimed that 256Kb and 64Kb densities made up the majority of their dollar purchases. Those using 4Mb SRAMs consumed so few as to contribute negligibly to overall sales, despite their strong showing on the chart of responses (see Figure 3-45).

Office Equipment

Relatively few of the respondents gave information about the office equipment market, probably because this market, like consumer electronics, is not heavily supported by North American manufactiirers.

Figure 3-46 shows device preferences for those who responded that office equipment was their major SRAM application. All used slow

(greater than 100ns) devices in either plastic DIP or PLCC packages.

Volumes ranged from tens of thousands of units for 1Mb SRAMs through an average of 2,000 for the 256Kb device, to prototype quantities of the 4Mb device.

Teiecommunications

Dataquest's Telecommionications service provided the imit forecast for the U.S. telecommunications market shown in Figure 3-47. The axis on the left shows the number of PBX lines in thousands; all other curves are measured, again in thousands of imits, on the right-hand axis.

Although unit constimption for major capital equipment is expected to grow considerably, PBX lines are not expected to grow, but to remain around 5.5 million lines over the forecast period.

IVIMRY-SEG-UW-9201 ©1992 Dataquest Incorporated December 28,1992

3-10 Memories Woridwide

Figure 3-48 shows resporises from the Semiconductor Procurement

Insights survey regarding SRAM as a percent of all MOS purchases.

A word of explanation is warranted here regarding nomenclature.

Premise commijnications devices comprise PBXs and voice/data terminals, while public telecoirununications devices comprise T-1 multiplexers, other central office switching equipment, voice messaging systems, and automatic call distributors.

Figure 3-49 shows SRAM preferences of the telecommunications market in general, based upon responses from the Memory User Wants and

Needs survey. As should be expected, the largest number of responses came from users of 256Kb and 1Mb densities, with a strong showing for the 64BCb density and fewer responses for 16Kb devices. Most of the devices sold were of an 8-bit organization, in either plastic DIP or

SOJ/SOIC packages, in volumes averaging about 58,000 imits per year.

Speed preferences were spread across a wide spectrum. Responses also came in for the 4Mb density, but volumes purchased were low.

The breakout of telecommunications respondents in North America showed a poor turnout from manufacturers of automatic dialers and fax machines, because of the overall lack of North America-based manufacturers of such devices. Figure 3-50 shows a strong resporise from manufacturers of central office switching systems, followed by manufacturers of voice/data terminals and PBXs. These three markets will be more deeply explored in the following figures.

Figure 3-51 shows SRAM usage of respondents whose major SRAM application is in PBXs. These respondents by and large were users of the 256Kb density, with a good portion going to the 1Mb SRAM, and some purchasing the 4Mb part. As a general rule, SRAM requirements in PBXs are more oriented toward density rather than speed, and this is supported by Figure 3-51.

Central office switching equipment, shown in Figure 3-52, uses SRAMs to store cormection information, and in certain state machines and cache memories for the controlling microprocessor or CPU. Small but high-speed memories are the norm in this sort of application. As a result, the survey revealed a number of manufacturers who continued to use the 64Kb density, although the largest number of respondents claimed the 1Mb density as the one that contributed the most to their

SRAM dollar purchases. Because this is the main growth market shown in Figure 3-47, it is worth further investigation by SRAM manufacturers.

Figure 3-53 shows the cross section of responses received from manufacturers who said that the majority of their SRAM purchases were made for voice/data terminals. The low cost of 64Kb and 256Kb

SRAMs probably accoimts for the fact that two-thirds of the respondents named these densities as the ones upon which they spent the most. The voice/data terminal market is price-sensitive. This is probably also the reason that 16Kb SRAMs appear in this figure, but not in any of the preceding telecommimications charts. Certain 16Kb SRAMs sell for less than $1.

December 28,1992 ©1992 Dataquest Incorporated MI\/IRY-SEG-UW-9201

?

CO

m

Table 3-1

Worldwide Computer Systems Forecast, Unit Shipments

Supercomputgr

Mainframe

Mid range

Workstation

Superworks tation

Traditional Workstation

Entry-Level Workstation

Personal Workstation

PC Subtotal

Transportable

Laptop AC

Laptop DC

Notebook

Pen-Based

Hand-Held

Desktop

Deskside

Total

NM = Not meaningful

Source: Dataquest (December 1992)

1990

1,008

15,115

727,712

407,624

19,703

120,100

267,821

0

23,935,200

101,000

349,000

2,491,000

408,000

10,000

217,000

19,773,200

587,000

25,086,659

1991

1,062

14,142

754,917

528,915

15,925

230,618

282,371

0

24,987,000

78,000

124,000

2,764,000

1,136,000

41,000

238,000

19,626,000

981,000

26,286,036

1992

1,229

13,640

754,537

677,000

12,500

246,000

418,200

300

26,710,000

42,000

65,000

3,101,000

1,794,000

122,000

763,000

19,441,000

1,383,000

28,156,406

1993

1,424

13,167

754,418

1,130,000

14,950

317,150

547,900

250,000

29,836,000

24,000

37,000

3,392,000

2,816,000

800,000

2,042,000

19,078,000

1,648,000

31,735,009

1994

1,645

12,721

754,548

2,217,000

17,400

370,700

828,900

1,000,000

33,774,000

15,000

28,000

3,669,000

4,393,000

1,759,000

3,877,000

18,204,000

1,829,000

36,759,914

19

1,9

12,2

754,9

3,802,0

20,3

416,7

1,365,0

2,000,0

39,127,0

10,0

24,0

3,933,0

6,809,0

3,289,0

6,188,0

16,899,0

1,975,0

43,698,

58

3-12

Figure 3-1

Percent of Responses, by Application Type

Percent

40

35

3 0 -

2 5 -

2 0 -

1 5 -

1 0 -

5-j

0

,\Wl-r4?m . W"

II i

Source: Dataquest (December 1992)

Memories Woridwide

^ 2 5

= c--=

• E O ^

G2D031XI

Figure 3-2

North American Electronic Equipment Production, 1989-1996

Millions of U.S. Dollars

100,000

90,000

80,000

70,000

60,000-

50,000-

40,000-

30,000-

20,000

10.000

0

1989 1990

— • * - • • — * * '

1 1 1 1 1

1991 1992 1993 1994 1995 1996

Data Processing

Communicatbn

Industrial

Consumer

Military/Aerospace

Transportation

Source: Dataquest (December 1992)

G2003T31

December 28,1992 ©1992 Dataquest Incorporated MMRY-SEG-UW-9201

i

I

i

Applications Types

3-13

Figure 3-3

SRAM A s a Percent of All M O S Purchases, b y Application

Percent of Responses in Category

7}

100-

9 0 -

8 0 -

7 0 -

6 0 -

5 0 -

4 0 -

3 0 -

2 0 -

1 0 -

0 -

\

>

5

V

R

I

;J

0-9

1

1 ll

10-19 2 0 - 2 9

Percent of All MOS Purchases

1

70-79 80-69 90-100

02003132

Source: Oataquest (December 1992)

MMRy-SEG-UW-8201

©1992 Dataquest Incorporated December 28,1992

3-14

Figiire3-4

Density Preferences: Consumer Electronics Manufacturers

Memories Woridwide

i

Source: Dataquest (December 1992)

Figure 3-5

SRAM As a Percent of All MOS Purchases, Consumer/Automotive

Percent of Responses in Category

G2(»3133

i

0-9 10-19 20-29 30-39 40-49 50-59 60-69 70-79

80-89 90-100

Percent of All MOS Purcliases

Source: Dataquest (December 1992)

GZD0S134

MMRY-SEG-UW-9201

i

December 28,1992 ©1992 Dataquest Incorporated

Applications Types

Figure 3-6

Cost of Electronics in Average U.S. Vehicle, 1989-1996

Average Per-Vehicle Cost ($)

450

3-15

Source: Dataquest (December 1992)

Figure 3-7

Density Preferences: AudioA^isual Manuf achirers

X 4Mb

/ \ (14.2%)

16Kb \

(28.6%) 1

\ 256Kb

\ (57.2%)

Source: Dataquest (December 1992)

02003135

G200S13e

MMRY-SEG-UW-9201 ©1992 Dataquest Incorporated December 28,1992

3-16

Figure 3-8

Breakout of Data Processing Responses

others (1%)

Portables (3%) —

Memories Woridwide

QZ0O3137

Source: Dataquest (December 1992)

Figure 3-9

Worldwide PC Shipments Forecast

Thousands of Units

30-

0 All Others

2 5 -

5V Notebook,

Pen-Based, Hand-Held

^ 3V Notebook,

Pen-Based, Hand-Held

2 0 -

1 5 -

1 0 -

5 -

v/yy.WSA

'VVvvHillllllllljrrnil

y/y/^&WS^BA

1992

Source: Dataquest (December 1992)

1993 1994

1995 1996

GJ00313B

December 28,1992

©1992 Dataquest Incorporated

IVBVIRY-SEG-UW-9201

Applications Types

Figure 3-10

Density Preferences: Data Processing Manufacturers

3-17

G2{»3139

Source: Dataquest (December 1992)

Figure 3-11

Worldwide Computer Systems Market Mix

1992

Supercomputer (2.0%) —

Workstation (9.0%)-,

1996

Supercomputer (2.5%) —i

/ Midrange \ PC \

\ y ^ Mainframe \ m

\ (23.8%) \ M

/ M a i n f r a m e \

/ (15.9%) \

1 Midrange ^ ^

I (18.4%) X

PC \

(39.7%) 1

\ / Workstation

\ / (23.5%)

Total = $109 Billion

Source: Dataquest (December 1992)

Total = $142 Billion

GBD0314O

MMiW-SEG-UW-9201

©1992 Dataquest Incoipotated

December 28,1992

3-18 Memories Woridwide

Figure 3-12

Worldwide PC and Personal Workstation Market Mix

1992

Personal Workstation (0.1%)—n

Hand-Held (0.5%)

Pen-Based (0.6%) 1

Notebook PC (6.2%) " i i i i ^ - t n — ^ ^ > ^

1996

Hand-Held (3.0%) — |

/ N o t o b o o k l

/ PC \

/ (22.6%) \ | I

J

/ Others \ 1 \

u l

/ ' * " v ^ 4 . 8 % ) \ Desktop PC \ f P en-B ased^'-Nj^ 1

1 ^^,„--^ 1

L,,^-^'' Desktop PC M

(27.1%) 1

\ j/'^ \ Personal J

\ (70.0%) m

\ y^ VWorkstatbn m

\ y ^ Others \ (19.1%) m

X (18.3%) \ M x^^^^X^

Total = $45.9 Billion Total = $70.1 Billion

Q20C3141

Source: Dataquest (December 1992)

December 28,1992 ©1992 Dataquest Incorporated MMRY-SE6-UW-9201

Applications Types

Figure 3-13

S R A M A s a Percent of All M O S Purchases, Data Processing

Percent of Responses in Category

100 H mi

I

0 PCs m other Data Processing

3-19

0-9 10-19 20-29 30-39 40-49 50-59 60-69 70-79 80-89 90-100

Percent of All MOS Purchases

Source: Dataquest (December 1992)

G2DD3142

MMRY-SEG-UW-9201 ©1992 Dataquest Incorporated December 28,1992

3-20

Figure 3-14

Computer Manufacturers, by Application Type

Percent of Responses

40

Memories Woridwide

Desktop

Cache

1 r

Laptop Workstation Mini Mainframe Main Writable Others

Cache Cache Cache Cache Memory Control

Stores

Computer Application of SRAI\/I

Source: Dataquest (December 1992) (33003143

Figure 3-15

Density Preferences: PC Caches

4Mb (5.9%)

Source: Dataquest (December 1992)

December 28,1992

©1992 Dataquest Incorpotated

64Kb (5.9%)

Gzxau

MMFiY-SEG-UW-9201

Applications Types

Figure 3-16

Density Preferences: Workstation Caches

3-21

Source: Dataquest (December 1992)

Figure 3-17

Density Preferences: Minicomputer Caches

/ IWto

JC (16.7%)

16Kb \

(16.7%) ^ A

\ 256Kb

\ (33.3%)

64Kb 1

(33.3%) M

GS0Q314S

G2(iagi46

Source: Dataquest (December 1992)

MMRY-SEG-UW-9201 ©1992 Dataquest Incoiporated iDecember 28,1992

3-22

Figure 3-18

Density Preferences: Mainframe Computer Caches

/ 4Mb

/ (33.3%)

256Kb \

(33.3%) 1

\

^ 1Mb

\ (33.3%)

Note: Segments do not add to 100 percent because of rounding.

Source: Dataquest (December 1992)

Figure 3-19

Density Preferences: Main Memory in Deskside Computers

Memories Woridwide

02003147

Source: Dataquest (December 1992)

December 28,1992

©1992 Dataquest Incorporated

Q200314S

MI\/IRY-SEG-UW-9201

Applications Types

Figure 3-20

I/O Device Manufacturers, by Application

Percent of Responses

3 0 - ) l

2 5 -

2 0 -

3-23

Disk

Cache

Laser

Printer

Source: Dataquest (December 1992)

CRT

Terminal

LAN

Modem

Type of I/O Device

T r

GinxatAS

MMRY-SEG-UW-9201

©1992 Dataquest Incorporated December 28,1992

3-24

Figure 3-21

Worldwide Disk Drive Production, 1990-1996

Thousands of Units

50,000 •

4 5 , 0 0 0 -

4 0 , 0 0 0 -

3 5 , 0 0 0 -

30,000 -

2 5 , 0 0 0 -

2 0 , 0 0 0 -

1 5 , 0 0 0 -

1 0 , 0 0 0 -

5 , 0 0 0 -

0

1990 1991 1992

1993

1.8 Inch

2.5 Inch

3.5 Inch

5.25 Inch

1994 1995

8-10 Inch

14 Inch

Source: Dataquest (December 1992)

Figure 3-22

Density Preferences: Disk Cache Manufacturers

Memories Worldwide

1996

02003150

/ 1Mb

/ (33.3%)

256Kb 1

(66.7%) a

Source: Dataquest (December 1992)

December 28,1992

©1992 Dataquest Incorporated MMRY-SEG-UW-9201

Applications Types

Figure 3-23

North America Page Printer Forecast^, 1991-1996

Thousands of Units

3,500

3-25

1991 1992

Source: Dataquest (December 1992)

1993 1994

Figure 3-24

Worldwide Display Terminal Production, 1990-1996

Thousands of Units

6,400

1995

1996

essxa\S2

5,000

19SI0

1991

Source: Dataquest (December 1992)

1992

1993

1994

1995

1996

MMRY-SEG-UW-9201

©1992 Dataquest Incorporated

December 28,1992

3-26

Figure 3-25

Density Preferences: CRT Terminal Manufacturers

Memories Woridwide

Source: Dataquest (December 1992)

Figure 3-26

U.S. Network Interface Card Forecast, 1989-1996

Thousands of Units

8.000-

1989 1990

Source: Dataquest (December 1992)

1991 1992

1993 1994 1995

1996

Gaomiss

December 28,1992

©1992 Dataquest Incorporated

MMRY-SEG-UW-9201

Applications Types

Figure 3-27

Density Preferences: LAN Board Manufacturers

3-27

Source: Dataquest (December 1992)

Figxure 3-28

Density Preferences: Fax/Modem Board Manufacturers

i

V. 4Mb

V (67%)

256Kb \

(33%) 1

Gaocetse

62003157

Source: Dataquest (December 1992)

MMRY-SEG-UW-9201 ©1992 Dataquest Incorporated

December 28,1992

3-2S

Figure 3-29

SRAM As a Percent of All MOS Purchases, Instnunentation/Industrial

Percent of Responses in Category

40-

I

^ Instrumentation & Test

B Industrial

Memories Woridwide

0-9 10-19 20-29 30-39 40-49 50-59 60-69 70-79 80-89 90-100

Percent of All MOS Purchases

Source: Dataquest (December 1992) GZD0315S

Figure 3-30

Density Preferences: Instrumentation Test Manufacturers

16Kb / ^ ^

/ 1Mb

(7%)/ \

/ (22%)

/ 64Kb 1

( (30%) 1

V 256Kb

\ (41%)

Source: Dataquest (December 1992)

December 28,1992

©1992 Dataquest Incorporated

G20D3159

MMRY-SEG-UW-9201

Applications Types

Figure 3-31

Instrumentation Test Manufacturers^ by Application

Percent of Responses

30

3-29

DSO/Logic Battery Medical IC Tester System GPS Remote Others

Analyzer Instrument Tester Receiver Monitor

Type of Instruments Built

Source: Dataquest (December 1992)

Gsocaieo

Figure 3-32

Density Preferences: Digital Storage Oscilloscopes/Logical Analyzers

Source: Dataquest (December 1992)

MMRY-SEG-UW-9201 ©1992 Dataquest Incotporated

G^OCQIGt

December 28,1992

3-30

Figure 3-33

Density Preferences: Battery-Operated Instruments

Memories Woridwide

I

Source: Dataquest (December1992)

Figure 3-34

Density Preferences: Medical Instnmientation

02003162

i

Source; Dataquest (December1992)

December 28,1992

©1992 Dataquest Incorporated

G2003163

MMRY-SEG-UW-9201

Applications Types

Figure 3-35

Density Preferences: Remote Measurement Equipment

3-31

Source: Dataquest (December1992)

G£D031»

Figure 3-36

North American Industrial Electronics Production

1 9 9 2

1 9 9 6

Other Industrial

Systems —v

(11%) \

Other Industrial

Systems —>

(11%) \

Security/Energy y"^

Management / \

(11%) \ / \

Security/Energy y ^

Management / \

(10%) ^ \

Manufacturing 1

Systems

(34%)

Manufacturing \

Systems m

(34%) 1

1 Medical

\ Equipment

1 I Medical

• \ Equipment

\ (21%)

' I n s t r u m e n t a t i o n ^ ^ f \ (22%)

Instrumentation ^ ^ P

(23%) M

(23%) M

T o t a l = $ 3 5 . 4 Billion

Source: Dataquest (December 1992)

MMRY-SEG-UW-9201 ©1992 Dataquest Incorporated

T o t a l

= $ 4 6 . 5 Billion eewBiw

December 28,1992

3-32

Memories Woridwide

Figure 3-37

Density Preferences: Industrial Control and Monitoring Manufacturers

i

Source: Dataquest (December 1992)

Figure 3-38

U.S. Defense Budget Procurements, 1991-1996

Billions of Dollars

66-

5 6 -

5 4 -

5 2 -

5 0 -

4 8 -

6 4 -

6 2 -

6 0 -

5 8 -

1991 1992

1993 1994 1995

Budget Year

Source: U.S. Department of Defense

1 9 9 6

020031ee

i

December 28,1992

©1992 Dataquest Incorporated

MMRY-SEG-UW-9201

Applications Types

Figxire 3-39

North American Military and Civil Aerospace Electronics Production

1992

1996

3-33

Total = $54.9 Billion

Source: Oataquest (December 1992)

Total = $60.1 Billion

Figure 3-40

SRAM As a Percent of All MOS Purchases^ Militaty/Aerospace

Percent of Responses In Category

45-ri

Gaooaies

0-9 10-19 20-29 30-39 40-49 50-59 60-69 70-79 80-89 90-100

Percent of All MOS Purchases

Source: Oataquest (December 1992) GSODSies

MMRY-SEG-UW-9201

©1992 Oataquest Incoiporated December 28,1992

3-34

Figure 3-41

Density Preferences: Military/Aerospace Manufacturers

Memories Woridwide

i

GZ003170

Source: Dataquest (December 1992)

Figure 3-42

Military/Aerospace Manufacturers, by Application

i

Others

(26%)

Radar \

(26%) \

Sonar (4%) -—A

I Navigational M

Satellite

\ (22%) #

^ (22%)

Source: Dataquest (December 1992)

December 28,1992 ©1992 Dataquest Incoiporated

52003171

MMF=ri'-SEG-UW-9201

I

Applications Types

Figure 3-43

Density Preferences: Radar Manufacturers

3-35

Source: Dataquest (December 1992)

Figure 3-44

Density Preferences: Navigational Equipment Manufacturers

GZ0030M

Source: Dataquest (December 1992)

MMRY-SEG-UW-9201

©1992 Dataquest Incorporated

G^naoas

December 28,1992

3-36

Figure 3-45

Density Preferences: Satellite Manufacturers

/ 4Mb

/ (20%)

64(0) \

(40%) 1

V 256Kb

\ ('iO'A)

Source: Dataquest (December 1992)

Figure 3-46

Density Preferences: Office Equipment Manufacturers

/ 4Mb

/ (20%) t 1Mb /

256Kb 1

\ (20%) /

(60%) m

Memories Woridwide

Gzooooge

i

I

Source: Dataquest (December 1992)

MMRY-SEG-UW-9201

I

December 28,1992 ©1992 Dataquest Incorporated

Applications Types

Figure 3-47

U.S. Telecom System Shipments, 1987-1996

Thousands of PBX Lines

Thousands of Units

3-37

T 1 1 1 J 1 1 1 1 f-o

1987 1988 1989 1990 1991 1992 1993 1994 1995 1996

T-1 Muxes

Voice Messaging Lines

ACDs

PBX Lines

Source: Dataquest (December 1992)

MMRY-SEG-UW-9201

©1992 Dataquest Incorporated

December 28,1992

3-38

Memories Woridwide

Figure 3-48

SRAM As a Percent of All MOS Purchases, Telecommunications

Percent of Responses In Category

100-<|

^ Premise Communicalion n Public Telecom

10-19 20-29 30-39 40-49 50-59 60-69 70-79 80-89 90-100

Percent of All MOS Purchases

Source: Dataquest (December 1992)

Gzooao^

Figure 3-49

Density Preferences: Telecommunications Manufacturers

4Mb (7.1%)

16Kb (4.8%)

Source; Dataquest (December 1992)

December 28,1992

©1992 Dataquest Incorporated

G2i»3toa

MMRY-SEG-UW-9201

Applications Types

Figure 3-50

Telecommunications Equipment Manufacturers, by Application

Percent of Responses

3-39

Telecom Application of SRAM

Source: Dataquest (December 1992)

Figure 3-51

Density Preferences: PBX Switch Manufacturers

/C 4Mb

/ \ ( 1 4 . 3 % )

\ {28.6%) / (57.1%) f

Source: Dataquest (December 1992)

MMRY-SEG-UW-9201 ©1992 Dataquest Incorporated

G2003101

G2003102

December 28,1992

3-40

Figure 3-52

Density Preferences: Central Office Digital Switching Equipment

Memories Woridwide

i

Source: Dataquest (December 1992)

Figure 3-53

Density Preferences: Voice/Data Terminals

y<^ 4Mb

16[«) ^ v

/ 1Mb \

/ (11.1%) \

/ S4Kb \

^ (22.2%) 1

\ 256Kb ^ " V ^

\ (44.5%) ^

G20O3103

G20031CM

I

December 28,1992 ©1992 Dataquest Incorporated

MMRY-SEG-UW-9201

Chapter 4

Usage Trends

Figure 4-1 is a compilation of all responses by density preference. As in the earlier application-related splits, the vast majority of responses indicated that the 256Kb density was the single density that contributed most to their dollar spending on SRAMs. This is followed by almost even proportions going to the 64BCb and 1Mb densities, and likewise nearly equal portions going to the 16Kb and 4Mb densities.

Figure 4-2 shows speed preferences by device density. Slow devices

(slower than a 70ns access time) accoimted for the majority of responses for all densities except for the 4Mb SRAM, and the fastest speed grades were only mentioned in the mainstream densities of 64K,

256K, and 1Mb. Except in the case of the 256Kb SRAM, the 45ns to

70ns speed grade appears to be displacing the slow speed grade step by step as SRAM generatior\s progress. A similar trend can be seen in the ramping of popularity of ihe 20ns to 35ns speed grade from the

16Kb density through the 1Mb density. The reverse of this trend appears to hold with the 10ns to 15ns speed grade, but this is most probably because, imtU recently, the less dense a part was, the faster it coiild be expected to operate. Designers were specifying smaller-thanideal SRAMs in order to meet their speed needs.

The survey examined usage expectations for next year (see Figure 4-3).

More than 60 percent of respondents anticipated no change in their purchasing pattern for their most important SRAM in the coming year.

The next largest number of respondents expected to increase purchases of this device, while the smallest number expected to decrease usage.

Details of these responses are shown in Figures 4-4 through 4-13.

Oddly enough, the trend shown in Figure 4-4 is the inverse of what would normally be expected. More than 70 percent of the users of the product nearest to obsolescence, the 16Kb SRAM, believed that there would be no change in their purchasing pattern for the device over the next year, while the device that is just now reaching its peak usage, the 1Mb, showed the smallest percentage of "no change" responses, at about 40 percent. This may have something to do with the fact that the tally is presented by the number of responses instead of units ptirchased. If we weight the responses by the average vmit purchase data used to generate Figures 4-14, 4-16, 4-18, 4-20, and 4-22, the results show that the smallest unit volume players are the least likely to expect a change. The reason is time to market. Smaller players, or those playing into smaller markets, are often less responsive than the true tigers whose unit consumption is high. So those who

MMFiY-SEG-UW-9201 ©1992 Dataquest Incorporated December 28,1992

4-2 Memories Wortdwide purchase the highest volume are the piost likely to move quickly to respond to competitive change. For example, the average imit consumption of 16Kb users expecting no change is only 14 percent of the annual unit volume of those expecting change, and for Ihe 64Kb density, the ratio is down to 6.4 percent.

Of those expecting to see change, it is only natural that the respondents most expecting to see an increase in consumption are those using the most advanced densities: 1Mb and 4Mb SRAMs (see

Figure 4-5). Figure 4-6 shows the expected percent increase in uiut consumption of those who responded that an increase was expected.

Bets are guarded, with the largest block expecting a modest zero to

24 percent increase in unit consumption.

Figure 4-7 attempts to find a relationship between the respondents who anticipate use decrease and the density that is currently their most important SRAM. It appears that, for whatever reason, the same general level of response is attained (about 5 percent) regardless of the density used. The variance is small enough as to be dismissable as noise. Figure 4-8 shows that the respondents expecting a decrease were also guarded, predominantly expressing beliefs that imit consimiption decreases would be smaller rather than larger.

Nearly aU of those respondents who expected to use a different device next year expect to use a der^er device, rather than a different organization of the same density, or a lower-density device. The proportion of respondents planning to purchase new parts as their major SRAM purchase is shown by the density of the current device in Figure 4-9.

Once again, it is interesting that those who expect the largest change are those who are using the most current devices, just as it was in

Figure 4-4. It seems more rational to expect those using products that are more mature to be the first to anticipate density increases. Once again, we attribute this to the makeup of the base of respondents using the lower-density devices.

Naturally, the bulk of the respondents expecting to use a higherdensity part plan to use the parts that will not be in a decline phase: the 1Mb and 4Mb densities (see Figure 4-10). Nearly equivalent numbers of respondents plan to upgrade to 64Kb and 256Kb densities, despite the maturity of the 64Kb part. Figure 4-11 is perhaps a more revealing perspective of the same information. It shows the migration path of those plarming to make a change. By far, the largest portion of users plan simply to move to the next density, with a few respondents planning to stay within the same generation, and another few planning to skip a generation and move to the density-after-next (for example, from 16Kb to 256K).

Moving to another slice of the pie shown way back in Figure 4-3, respondents planned to upgrade the speed of the parts they purchase within the next year, but otherwise to stay with the same density and organization (see Figure 4-12). The largest changes are expected from users of 64Kb devices, most likely to be cache parts on the leading edge of speeds, which are used in RISC caches for demanding CPUs

December 28,1992 ©1992 Dataquest Incorporated IVIMRY-SEG-UW-9201

Usage Trends 4-3 such as the R3000 and Hewlett-Packard's Precision Architecture. Dataquest believes that the lower responses for the 256Kb and 1Mb densities can be attributed to the relatively wide availability of high-speed versions of these devices during the six-month period preceding this survey.

Figure 4-13 shows the number of respondents expecting to see their end applications phased out by the end of next year. As was the case with the anticipated decrease of Figure 4-7, all responses fell within the 6 percent area, with differences in response by density apparently caused by random sampling noise.

By Density

The following paragraphs split out volume and package preference trends by device density.

16Kb

Figure 4-14 shows 16Kb SRAM unit purchases by respondent.

Volume peaks in the 1,000 to 9,999 region, with only 5 percent of the respondents claiming to make purchases larger than

100,000 xmits annually. This will be seen to pale by comparison to the purchasing profiles for higher-density SRAMs. The package preference by respondent is shown in Figure 4-15. AU respondents use plastic packages, with more than 50 percent still using plastic

DIP The majority of other respondents chose PLCC and SOIC/SOJ packages, with responses evenly split between the two. The popularity of the PLCC probably can be attributed to the fact that the

SOIC and SOJ packages were not available until this device had entered the maturity phase of its life cycle.

Because Figure 4-15 is broken out by response, rather than by unit volimie, it does not accoimt for unit sales in showing package preference. Owing largely to the highest volume application reported during this survey, the volume by package type breaks out greatly in favor of the PLCC, which accoimted for 63 percent of aU devices reported, followed by plastic DIP at 20 percent and

SOIC/SOJ at 17 percent.

64Kb

Figure 4-16 is not aU that different from Figure 4-14, except that fewer respondents claimed to be using extremely small unit volumes of the device, with a higher peak now appearing in the modest 1,000 to 9,999 area. Volumes for the 64Kb device, like those for the 16Kb device, generally reflect that the biJk of those users for whom these parts represent their most significant SRAM purchases are not major market players.

Package preferences by respondent in the 64Kb market are also similar to those in the 16Kb market (see Figure 4-17). Plastic DIPs accoimt for more than 50 percent of the responses, and a strong showing exists for the ceramic DIP because of some military responses. The SOIC/SOJ package was a latecomer to these two

l\/IMRY-SEG-UW-9201 ©1992 Dataquest Incorporated December 28,1992

4-4 Memories Woridwide devices, which partly accoimts for the strength of the responses favoring the plastic DIP package. PLCCs had a short period of grace in the 64Kb market before being largely displaced in the

256Kb and denser markets by the SOIC and SO] packages. As a result. Figure 4-17 contains a small wedge of respondents who purchase a major volume of their SRAMs in the PLCC package. Some respondents also indicated that they were using bare SRAM dice, probably to be used in multichip modules, which are speed-driven and are well matched to the leading-edge speeds required by some

RISC CPUs, as we found in the paragraph discussing Figure 4-12.

When responses were weighted to vmits consumed, a different partitioning evolved, with 88 percent of the packages used being SOICs and SOJs, 10 percent plastic DIPs, and less than 1 percent either ceramic DIP or PLCC. All respondents who said they used more than 100,000 imits were using SOIC and SOJ packages.

256Kb

Figure 4-18 is indicative of the maturity stage of the 256Kb SRAM's life cycle. The bulk of the respondents said they were purchasing more than 1,000 units per year of their most important device, and about 15 percent purchased 100,000 or more imits per year.

Package usage differs from that of less-dense devices, in that the

SOIC and SOJ packages were used by more than one-third of the respondents (see Figure 4-19), whereas they were used by about one-fourth of the respondents whose major volume device was either 16Kb or 64Kb. Usage among respondents was divided almost exclusively between SOIC/SOJ and plastic DIP, with 56 percent going to the SOIC and SOJ packages, and 44 percent to the plastic

DIP. Only about 10 percent of the respondents used packages that were neither plastic DIP nor SOIC/SOJ, and none of these devices accounted for as much as 1 percent of the total units purchased by aU 256Kb respondents combined.

1Mb

As in the 256Kb market, the 1Mb market is established and not in a decline phase, so the statistics of Figure 4-20 are reasonable.

Seventy percent of the responses were from companies whose annual imit volume was more than 1,000 units, and more than

12 percent were purchasing 100,000 or more imits per year.

For a change, DIP packages accounted for less than half of the responses (see Figure 4-21), with surface-mount accounting for nearly 60 percent of the responses, and more than 70 percent of the units used. PLCCs make a surprisingly strong showing, in spite of the limited supplier base, accounting for 4 percent of the units reported, but are overwhelmed by the use of SOICs and SOJs, which accounted for a full 68 percent of all units. This density had the largest percentage of respondents who declined telling their package preference, accoimting for 19 percent of the imits tallied.

December 28,1992 ©1992 Dataquest Incorporated MMRY-SEG-UW-9201

I

Usage Trends

4Mb

Our survey results were not statistically significant for the 4Mb device, so package preferences and shipment volumes will not be shown graphically. Dataquest estimates that fewer than 10,000 units of 4Mb SRAM shipped in 1992, and that, while significant volumes of 4Mb P-SRAMs were used by two North American manufacturers of hand-held com.puters, the 4Mb P-SRAM is not generally popular in the system design community.

4-5

t

MMFtY-SEG-UW-9201 ©1992 Dataquest Incorporated December 28,1992

4-6

Figxire 4-1

Density Preferences: All Manufactturers

Memories Woridwide

«

GzooaiDS

Source: Dataquest (December 1992)

Figure 4-2

Speed Preference, by Device Density

Percent of Responses

50~fl

16Kb

0 4-8ns

Source: Dataquest (December 1992)

64Kb 256Kb 1Mb

Preferred Density

4Mb

10,12,15ns H 20,25,35ns • 45.55,70ns ^ Slow

GSMsioe

December 28,1992 ©1992 Dataquest Incorporated MMRy-SEG-UW-9201

4-7

Usage Trends

Figure 4-3

SRAM Usage Expectations for Next Year

Dont Know/Refusal (1.2%)

Decrease (4.9%)

Faster Part (6.9%)

Phase-Out (6.9%) eZ003107 Source: Dataquest (December 1992)

Figure 4-4

Responses Anticipating No Change, by Device Density

Percent of Responses

80-

^GKb

Source: Dataquest (December 1992)

64Kb

T T T m

256Kb 1 Mb 4Mb

All

Current Preferred Density

G200310B

l\/IMRY-SEG-UW-9201 ©1992 Dataquest Incorporated December 28,1992

4-8

Figure 4-5

Responses Anticipating Use Increase, by Device Density

Percent of Responses

25-

Memories Woridwide

10:-i

5 -

0 -

T

16Kb S4Kb 256Kb 1Mb 4Mb

AJI

Current Preferred Density

Source: Dataquest (December 1992)

GzoceiD9

Figure 4-6

Expected Unit Consumption Increase Next Year

Source: Dataquest (December 1992)

December 28,1992 ©1992 Dataquest Incorporated

Gzoosua

MMRY-SEG-UW-9201

Usage Trends

Figure 4-7

Responses Anticipating Use Decrease, by Device Density

Percent of Responses

7-

4-9

16Kb

T

64Kb 256Kb 1Mb 4Mb All

Current Preferred Density

Source: Dataquest (December 1992)

02003111

Figure 4-8

Ejqjected Unit Consumption Decrease Next Year

Source: Dataquest (December 1992)

MMRY-SEG-UW-9201 ©1992 Dataquest Incorporated

GZ0O3112

December 28,1992

4-10

Figure 4-9

Responses Anticipating Different Device Use, by Device Density

2 -

0 -

4 -

Percent of Responses

14

10

8 -

6 -

12

I

16Kb 64Kb

T

256Kb 1Mb 4Mb

Current Preferred Density

Source: Dataquest (December 1992)

All

Memories Worldwide

Figure 4-10

New Part Planned for Next Year

Source: Dataquest (December 1992)

December 28,1992 ©1992 Dataquest Incorporated

G20O3114

MMRY-SEG-UW-9201

Usage Trends

Figure 4-11

Anticipated Migration to New Part

Beyond Next Generation (5%) -^

Current Generation ( 5 % ) - A , „ * — ^ " ^

/ N o \ \

/ Answer \ \

/-,4ii%) \ \

A Next Generation M

\ (79%) #

Source: Dataquest (December 1992)

Figure 4-12

Responses Expecting to Upgrade Speed, by Device Density

Percent of Responses

16-

*4-

1 2 -

1 0 -

&^ e -

;4^

2 -

0

16Kb

"T

m

64Kb 256Kb 1IVIb 4Mb

Current Preferred Density

Source: Dataquest (December 1992)

4-11

G20CQ11S

G2003116

MMRY-SEG-UW-9201 ©1992 Dataquest Incorporated

December 28,1992

4-12

Memories Woridwide

Figure 4-13

Responses Anticipating Application Phaseout, by Device Density

Percent of Responses

8 -

7 -

6 -

5 -

4 -

3 -

2 -

1 -

0

16Kb

T

64Kb 256Kb 1Mb

4Mb

Current Preferred Density

Source: Dataquest (December 1992)

All

G20oa)i7

Figure 4-14

Volume of 16Kb Purchases

1 5 -

1 0 -

5 -

0-

Percent of Responses

30-(I

2 5 -

2 0 -

1-9 10-99

100-999 1,000- 10,000-

100.000+

9,999 99,999

Annual Usage (Units)

Source: Dataquest (December 1992)

GSOosTie

December 28,1992

©1992 Dataquest Incorporated

MMRY-SEG-UW-9201

Usage Trends

Figure 4-15

Package Preference for 16Kb SRAMs

Don't Know/Refusal (5%)

4-13

G20aQ119

Source: Dataquest (December 1992)

Figure 4-16

Volume of 64Kb Purchases

Percent of Responses

4 5 - 1 -

4 0 -

3 5 -

3 0 -

2 5 -

2 0 -

1 5 -

1 0 -

5-1

0

1-9 10-99

100-999 1,000-

9,999

Annual Usage (Units)

Source: Dataquest (December 1992)

^

10,000-

99,999

100,000+

Gzooaiso

MMRY-SEG-UW-9201

©1992 Dataquest Incorporated December 28,1992

4-14

Figure 4-17

Package Preference for 64Kb SRAMs

Die (2%) -1

PLCC (5%)~V_J_

/ceramic D i p \ I

/ (15%) \ \

\ SOIC/SOJ /

\ (25%) /

P-DIP 1

(53%) 1

Memories Woridwide

G20a3121

Source: Dataquest (December 1992)

Figure 4-18

Volume of 256Kb Purchases

Percent of Responses

4 0 -

3 5 -

30

25 HI

20

1 5 -

1 0 -

5 -

0

1-9

T T

10-99 100-999

1.000-

9,999

Annual Usage (Units)

Source: Dataquest (December 1992)

10,000-

99.999

100,000+

GS0031Z2

I

MMRY-SEG-UW-9201

December 28,1992 ©1992 Dataquest Incorporated

Usage Trends

Figure 4-19

Package Preference for 256Kb SRAMs

LCC(1%)-,

PLCC (4%)-,

Ceramic DIP (4%)-v^.^J—

w

1 (37%) 1

(54%) 1

4-15

G2a»123

Source: Dataquest (December 1992)

Figure 4-20

Volume of 1Mb Purchases

Percent of Responses

3 0 -

2 5 -

2 0 -

1 5 -

1 0 -

1-9

10-99 100-999 1,000-

9,999

Annual Usage (Units)

Source: Dataquest (December 1992)

10,000-

99,999

100,000+

G2C031Z4

MIVIRY-SEG-UW-9201

©1992 Dataquest incorporated

December 28,1992

4-16

Figure 4-21

Package Preference for 1MB SRAMs

Others (3%)-i

Dont Know/Refusal (7%) ^v,,——T"^

Die (3%) s y ^ ^ \

LCC (3%) ^ y \ \ \

Memories Woridwide

/ c e r a m i c D I P N N X \

I PLCC ^>^

\ (10%) ^ X ' ^

SOIC/SOJ 1

(44%) 1

V * ^ P-DIP

\ (23%)

Source: Dataquest (December 1992)

Gzcoatss

I

December 28,1992 ©1992 Dataquest Incorpoiated MMRY-SEG-UW-9201

Dataquest Incorporated

1290 Ridder Park Drive

Sail Jose, California 95131-2398

United States

Phone: 01408437-8000

Facsimile; 014084370292

Dataquest Incorporated

Dataq uestledge way

The Corporate Center

550 Cochituflte Road

FramiiJ^am, Massachusetts 01701-9324

United States

Phone: 01-508-370-5555

Facsimile: 01-508-370^262

Damquest Europe Limited

Roussel House Broadwater Park

Denham, Near Uxbridfie

Middlesex UB9 5HP

Eniliind

Phoae: 44-895-835050

Facsimile; 4a95-835260/l

Dataquest Japan Limited

Shinliawa Saiiko Building

1-3-17 Shinkawa Ghuo-ku

'Ibkyol04

Japan

Phone; 81-3-5566-0411

Facsimile; 81-3-5566-0425

Offices in

Costa Mesa, Munich,

Paris, and Seoul

Representative Agencies in

Bangkok, Hong Kong,

Kronberg, North Sydney,

Sngapofe, and Taipei

^1992 Dataquest Incorporated

0014238

Library* Corporate

SIS-1140357

Bt 1

197-1234

Internal Distribution

n i

I

Worldwide MOS Memory Forecast

August 31,1992

DataQuest

Market Statistics

Library Copy

DO NOT REMOVE!

Memories Worldwide

MMRY-5EG-MS-9203

}

INFORMATION RESOURCE CENTER

IJAT:'-\(:(UEST INCORPORATED

!2''w R:dder Park D n v e

S-M.los^?, C A 95131-2398

i i M ^ 4 ^ J 6 0 0

Worldwide MOS MIHidry Forecast

August 31,1992

I

Dataoyest'

Source:

Dataquest

Market Statistics

File behind die Market Statistics tab inside the binder labeled Memories Worldtoide

Published by Dataquest Incorporated

The content of this report represents our interpretation and analysis of information generally available to the public or released by knowledgeable individuals in the subject industry, but is not guaranteed as to accuracy or completeness. It does not contain material provided to us in confidence by our clients.

Printed in the United States of America. All fights reserved. No p>att of this publication may be reproduced, stored in retrieval systems, or transmitted, in any form or by any means—mechanical, electronic, photocopying, duplicating, microiilining, videotape, or otherwise—without the prior permission of the publisher.

© 1992 Dataquest Incorporated

August 1992

i

Worldwide MOS Memory Forecast

Table of Contents

Introduction 1

Segmentation 1

Definitions 1

Product Definitions 1

Regional Definitions 2

Line Item Definitions 2

Forecast Methodology 2

MOS Memory Forecast Assumptions 2

Regional Issues 3

Worldwide Economic Growth Expectations 3

DRAM Forecast Assumptions 3

Bit Growth 4

Product Life Cycles and Trends 4

Product Differentiation 4

Impact of Flash 4

Major Applications 4

SRAM Forecast Assumptions 5

Nonvolatile Memory Forecast Assumptions 5

EPROM 5

EEPROM 5

ROM 5

Flash :.. 5

Exchange Rates 6

Table Page

1-1 Factory Revenue from Shipments of MOS Memory to the World, 1989-1996

(Millions of U.S. Dollars) 7

1-2 Shipments of MOS Memory to the World, 1989-1996 (Millions of Units) 8

1-3 Average Selling Price for Shipments of MOS Memory to the World, 1989-1996

(U.S. Dollars) 9

1-4 Shipments of MOS Memory to the World, 1989-1996 (Trillions of Bits) 10

1-5 Price per Bit for Shipments of MOS Memory to the World, 1989-1996 (Micro Dollars) j j

2-1 Factory Revenue from Shipments of DRAMs to the World, 1989-1996 (Millions of U.S.

Dollars) 12

2-2 Shipments of DRAMs to the World, 1989-1996 (Millions of Units) 13

2-3 Average Selling Price for Shipments of DRAMS to the World, 1989-1996 (U.S. Dollars) 14

2-4 Shipments of DRAMs to the World, 1989-1996 Clrillions of Bits) 15

Note: All tables show estimated data.

1

Table Page

2-5 Price per Bit for Shipments of DRAMs to the World, 1989-1996 (Micro Dollars) 16

3-1 Factory Revenue from Shipments of SRAMs to the World, 1989-1996 (Millions of U.S.

Dollars) 17

3-2 Shipments of SRAMs to the World, 1989-1996 (Millions of Units) 19

3-3 Average Selling Price for Shipments of SRAMs to the World, 1989-1996 (U.S. Dollars) 21

3-4 Shipments of SRAMs to the World, 1989-1996 (Trillions of Bits) 23

3-5 Price per Bit for Shipments of SRAMs to the World, 1989-1996 (Micro Dollars) 25

4-1 Factory Revenue from Shipments of EPROMs to the World, 1989-1996 (Millions of U.S.

Dollars) 27

4-2 Shipments of EPROMs to the World, 1989-1996 (Millions of Units) 28

4-3 Average Selling Price for Shipments of EPROMs to the World, 1989-1996 (U.S. Dollars) 29

4-4 Shipments of EPROMs to the World, 1989-1996 (Trillions of Bits) 30

4-5 Price per Bit for Shipments of EPROMs to the World, 1989-1996 (Micro Dollais) 31

5-1 Faaory Revenue from Shipments of ROMs to the World, 1989-1996 (Millions of U.S.

Dollars) 32

5-2 Shipments of ROMs to the World, 1989-1996 (Millions of Units) 33

5-3 Average Selling Price for Shipments of ROMs to the World, 1989-1996 (U.S. Dollars) 34

5-4 Shipments of ROMs to the World, 1989-1996 (Trillions of Bits) 35

5-5 Price per Bit for Shipments of ROMs to the World, 1989-1996 (Micro Dollars) 36

6-1 Factory Revenue from Shipments of EEPROMs to the World, 1989-1996 (Millions of U.S.

Dollars) 37

6-2 Shipments of EEPROMs to the World, 1989-1996 (Millions of Units) 38

6-3 Average Selling Price for Shipments of EEPROMs to the World, 1989-1996 (U.S. Dollars) 39

6-4 Shipments of EEPROMs to the World, 1989-1996 (Trillions of Bits) 40

6-5 Price per Bit for Shipments of EEPROMs to the World, 1989-1996 (Micro Dollars) 4l

7-1 Faaory Revenue from Shipments of Flash Memory to the World, 1989-1996 (Millions of

U.S. DoUars) 42

7-2 Shipments of Flash Memory to the World, 1989-1996 (Millions of Units) 43

7-3 Average Selling Price for Shipments of Flash Memory to the World, 1989-1996 (U.S.

Dollars) 44

7-4 Shipments of Flash Memory to the World, 1989-1996 (Trillions of Bits) 45

7-5 Price per Bit for Shipments of Flash Memory to the World, 1989-1996 (Micro Dollars) 46

Note: All tables show estimated data.

wmry'

Worldwide MOS Memory Forecast

Introduction

This document contains detailed information on Dataquest's view of the MOS memory market. Included in this document is:

• 1992-1996 MOS memory forecast

Analyses of the MOS memory market provide insight into high-technology markets and reinforce estimates of consumption, production, and company revenue.

More detailed data on this market may be requested through our client inquiry service.

Dataquest's qualitative analysis of these data can be found within the Dataquest PerspecHt/es located within the binder of the same name.

Segmentation

This section defines the market segments that are specific to this document. For a complete description of all market segments tracked by

Dataquest, please refer to the Dataquest High-

Technology Guide: Segmentation and Glossary.

Dataquest defines the MOS memory market as

DRAM, SRAM, EPROM, ROM, EEPROM, and flash memory. In this quarterly memory shipment volume, Dataquest segments the MOS memory market by product type and density according to the following scheme:

• DRAM (densities from 64K through 256Mb)

• Fast SRAM (densities from 16K through

16Mb)

• Slow SRAM (densities from l6K through

16Mb)

• EPROM (densities fi-om 16K through 16Mb)

• ROM (densities ft-om 32K through 256Mb)

• EEPROM (densities from 256b through 1Mb)

• Flash memory (densities from 256K through

64Mb)

Definitions

This section lists the definitions that are used by Dataquest to present the data in this document. Complete definitions for all Dataquest terms can be found in the Dataquest

High-Technology Guide: Segmentation and

Glossary.

Product Definitions

DRAM: Includes E>ynamic RAM, Multiport-

DRAM (M-DRAM), and Video-DRAM (V-DRAM).

DRAMs have memory cells consisting of a single transistor, and require regular externally cycled memory cell refreshes. These are volatile memories and addressing is multiplexed.

SRAM: Includes Static RAM, Multiport-SRAM

(M-SRAM), Battery Backed-Up SRAM (BB-

SRAM), and Pseudo-SRAM (P-SRAM). SRAMs have memory cells consisting of a minimum of four transistors (P-SRAMs have memory cells consisting of a single transistor and are similar to DRAMs). SRAMs do not require externally cycled memory cell refreshes. These are volatile memories and addressing is not multiplexed (except in the case of P-SRAM).

EPROM: Erasable Programmable Read-Only

Memory. This product classification indudes

Ultraviolet EPROM (UV EPROM) and One-Time

Programmable Read-Only Memory (OTPROM).

EPROMs have memory cells consisting of a single transistor, and do not require any memory cell refreshes. These devices are considered nonvolatile memories.

Mask ROM: Mask-Programmable Read-Only

Memory. Mask ROM is a form of memory that is programmed by the manufacturer to a user specification using a mask step. Mask ROM is programmed in hardware rather than software.

These devices are considered nonvolatile memories.

EEPROM: Electronically Erasable Programmable Read-Only Memory. Included are Serial

EEPROM (S-EEPROM), Parallel EEPROM (P-

EEPROM), and Electronically Alterable Read-

Only Memory (EAROM). EEPROMs have memory cells consisting of a minimum of two transistors, and do not require memory cell refreshes. This product classification also indudes Nonvolatile RAM (NV-RAM), also known as Shadow RAM. These semicondurtor products are a combination of SRAM and

EEPROM technologies in each memory cell.

The EEPROM functions as a shadow backup for the SRAM when pow^er is lost. These devices are considered nonvolatile memories.

Flash Memory: Includes nonvolatile products designated as Flash EPROM/EEPROM that incorporate either 5V or 12V programming supplies and one-transistor (IT) or twotransistor (2T) memory cells with electrical programming and fast bulk/chip erase. These devices are considered nonvolatile memories.

Memories Worldwide

single unit of that product contains. This number is reported in microdollars; there are 1 million microdollars per U.S. dollar. For an overall product category (for example, DRAM), this metric is calculated by dividing the category's total factory revenue by its total number of bits.

Regional Definitions

North America: Includes United States and

Canada

United States: Includes 48 contiguous states,

Washington, D.C., Alaska, Hawaii, and Puerto

Rico

Europe: Western Europe

Japan: Japan

Asia-Pacific/Rest of World: All other coimtries

Forecast Methodology

Dataquest publishes five-year unit shipments and factory revenue forecasts for the MOS memory market. In doing so, Dataquest utilizes a variety of forecasting techniques (both qualitative and quantitative) that vary by technology area. An overview of Dataquest forecasting techniques can be found in the

Dataquest Research Methodology Guide.

Line Item Definitions

Factory revenue: Calculated by multiplying a product's overall unit shipment total by the product's ASP.

Unit shipments: All unit shipments, both merchant and captive, for memory suppliers selling to the merchant market; excludes totally captive suppliers, w^here devices are manufactured solely for the company's own use.

Average selling price (ASP): The average billing price per unit that is paid for a product when it leaves the factory; takes into accoxmt discounts given to the distribution channel and multiple-purchase discounts. Prices are averaged over all companies, package types, lot sizes, and the entire speed mix, and they represent sales to both military and commercial accounts.

Number of bits: Calculated by multiplying a product's unit shipment total by the number of bits that a single unit of that product contains.

Price per bit (PPB): Calculated by dividing a product's ASP by the number of bits that a

MOS Memory Forecast

Methodology

The following is Dataquest's MOS memory forecast methodology:

• Survey the leading memory vendors throughout the year for company expectations, as well as for their views of the markets that they participate in.

• Examine statistics provided by a number of industry organizations (such as WSTS and

M m ) for up-to-date monthly trends.

• Perform time-series analysis as well as supply judgmental industry knowledge to produa and applications trends.

MOS Memory Forecast

Assiunptions

The following are assumptions for market cycle issues:

• Price elasticity is the basic driving mechanism for all MOS memories. Prices are now

• about 30 percent of those the industry offered at the end of the last cyclical upturn (summer 1989). These reductions

TJ^dll drive the next cyclical upturn, which we believe is now under way.

©1992 Dataquest Incorporated August—Kepioduction Prohibited

i

Worldwide MOS Memory Forecast

The market growth will go through another

"typical" growth cyde, with significant expansion beginning in 1992, accelerating in 1993, peaking in 1994, and contracting in 1995.

We assume that this cycle will exhibit about the same evolutionary path as the strong growth cycles that crested in 1988-1989,

1983-1984, 1979-1980, and 1973-1974.

- During those cycles, which ran from 16 to 20 quarters in length, the quarterly revenue run rate grew fourfold to tenfold from trough to the following peak. During the subsequent contractions, which ran three to four quarters, the quarterly revenue run rate dropped 25 to 50 percent, before stabilizing and establishing a new base. Prices per bit dropped 50 to

80 percent during these contractions.

- We expea this cycle to be more moderate, both in its expansionary and contraction phase, because of the slower overall growth rate of the market, the increased attention being paid to profitability in all corporate strategies, and the restrictions put on pricing by the intervention of various government agencies in Europe and the United States.

• Prices have risen just once (in 1988) during the supply-constrained shortage. We do not anticipate such a severe imbalance that would again raise prices in any but a temporary way; that is, there may be product imbalances, such as package types, or organization mix, but we do not expect any aggregate, across-the-board shortages.

Regional Issues

The weakness of the Japanese market in the first half of 1992 will serve as a significant constraint on worldwide memory market growth for 1992. We expect demand in Japan to be turned around by year-end, and all four regions of the word -will advance in concert in the early part of 1993. The secular trend calls for Asia-Pacific/Row taking an increasing share of the MOS memory market, at the expense of both Japan and the United States. We expect Europe to manage to retain its present share of consumption by a tariff structure that encourages domestic production of both MOS memories and the systems using them.

We expect that, for this forecast period,

MOS memory will gain in its share of the overall semiconduaor industry revenue, as a part of a natural cyclical pattern, but that in the contraction phase it will retreat from the cyclical high-water mark of 1994.

The capacity-demand balance that existed in the market, and the net strength of demand, has historically determined the madcet dynamics of revenue and profitability. Demand for bits has always grown, though from time to time not enough to compensate for declining per-bit prices brought on by supply excesses. This has caused a market contraction.

Worldwide Economic Growth

Expectations

Overall, the world economies continue to face an uncertain future, and there is some concern that we will drift back into a recession. The

Dataquest view of future economic activity anticipates the growth rates for 1992 through

1996 shown at the bottom of this page.

DRAM Forecast Assumptions

The following sections detail our DRAM forecast assumptions.

Estimated Real GDP Growth Rates, 1991-1996 CPercentage)

United States

Europe

J^an

Asia-Padfic/ROW

1991

-0.7

2.1

4.4

7.5

1992

2.1

1.2

2.0

6.8

1993

Z5

2.9

3.3

7.2

1994

2.3

3.5

3.5

7.4

1995

2.6

3.7

4.0

7.7

©1992 Dataquest Incorporated August—Reproduction Prohibited

Bit Growth

In the short term, we assume that the DRAM market is stumbling through the beginning of a qrclical upturn that will accelerate the bitgrowth rate, absorb available capacity, slow the PPB rate of decline, and improve profits through the end of 1994. At that time, as supply again passes demand, the market will weaken and revenue will contract.

Over the long term, we expect to see a continued decline in the rate of bit growth rate, to average about 60 to 65 percent per year from 1992 through 1996, and continue to slow thereafter.

Product Life Cycles and Trends

We assume that the price crossover from the

4Mb to the 16Mb DRAM will occur in late

1994 or early 1995, thereby giving the 4Mb product a slightly longer liiTetime than earlier generations. We further expect this lengthening trend to continue at the 16Mb and 64Mb densities.

As bit growth slows and processing becomes more expensive, we expect the floor price, under which the product cannot be sold profitably, to rise from generation to generation. This trend will be a contributing faaor in the gradual lengthening of DRAM product lifetimes. (This trend may be slowed through advanced-technology cost-sharing joint ventures, such as have been increasingly fi^uent in the memory/DRAM business.)

Although 1Mb DRAMs showed a new resurgence of life early in 1992, we do not expect their life-cyde curves to be significantly deferent from those of their predecessors; that is, the three-generations-at-a-time hypothesis may be real, but will in fact be very similar from what has gone before.

Product Differentiation

Though the DRAM market is becoming differentiated with the growth of wide DRAMs,

LP, 3.3V DRAMs, and new architectures, we believe that the forecast period here will continue to be dominated by mostly standard, mainstream parts. Even in the outer years, more than 80 percent of the units will

Memories Worldvvlde

continue to be 5V, and more than 70 percent are expected to be xl or x4.

Impact o f Flash

Over the long term, flash memories will have minor impact on DRAMs, and only in those applications where software is downloaded into DRAM and read repeatedly. In the longer term, flash has the potential for significant cost-per-bit advantages because of superior scaling and reduced cell complexity.

At the major market interfaces with DRAMs and flash memories, we expea SRAMs and

DRAMs to continue to coexist in the forecast period, though several high-data-rate DRAM architectures may absorb both standard DRAM and SRAM while creating new, bit-hungry applications. Flash's greatest impact is expected to be at the later 16Mb and 64Mb densities, and will mostly be an NVM replacement and new-market development product

Major Applications

Software is emerging as the silent driver of

DRAM demand. It is no longer so easy to count hardware/boxes and multiply to get

DRAM demand. Software moves independently, often finding its way into the installed base long after the hardware has been sold.

At present, fully 70 percent of DRAMs go into small computer systems, from hand-helds to workstations. Another 15 percent go to other

EDP and office equipment, such as laser printers, fax machines, and copiers. The remainder go elsewhere. But despite this categorical concentration, the DRAM end-use market is actually quite broad, as distributed processing power finds its way into all marmer of electronic equipment and computers of all sizes are made useful in a broad range of endeavors and activities. PCs, or PC-like small sjretems, are found in the home, at school, in industrial environments, in white-collar office and small business, and at virtually every retail outlet.

This pervasiveness is both good and bad. It is a good buffer and will prevent any rapid deterioration of demand, as was seen in earlier cycles. But because demand is diffused, explosive growth is also precluded from ever repeating the 1983 to 1984 first PC wave.

©1992 Dataquest Incorporated August—Reproduction Prohibited

dose to market prices.

Worldwide MOS Memory Forecast

At the same time, the aftennarket for DRAMs has also grown, which appears to be buffering the industry from the strong upturns and downturns experienced in the past. The installation of the STA in 1986, we believe, has helped moderate the a^ressive price cutting in the down cycle, and kept production costs

Graphics applications are becoming major forces driving the market However, ±ese applications will see their greatest growth period after the 16Mb comes into volume production in the niid-1990s.

SRAM Forecast Assumptions

The following are our SRAM forecast assumptions:

• Historical trends will tend to repeat themselves over the next five years. These drive the following:

- Market composition by density of device

- Price per bit, and relative PPB for various densities

— Migration toward faster devices

— Migration toward wider parts

• PC caches will continue to consume the lion's share of fast SRAMs, even as battery-operated PCs grow in stature.

• Static RAM ASPs will track dynamic RAM

ASPs, however, slower SRAMs will be sold at a bargain as the demand for faster parts tends to obsolete speeds slower than 100ns.

• Pseudo-static RAMs will grow in acceptance from their limited stance today.

• There will continue to be a speed gap in all SRAM densities where sales will be low.

• Applications for slow SRAMs will continue to be far-flung.

• There will be a cydical softening of the market in 1995.

• The economy will strongly impact the bitgrowth rate of slow SRAMs and the ASPs of fast SRAMs.

• The SRAM market will grow more quickly than will the DRAM market, but will continue to stay the significantly smaller of the two.

• The SRAM market is not seriously threatened by new technologies such as flash, cached DRAM, Rambus, and microprocessors with on-board cache. These technologies will coexist with SRAM, and may even create the opportunity for new SRAM applications.

Nonvolatile Memory Forecast

Assumptions

EPROM

This market segment will remain relatively flat for the forecast period and may in faa drop off more rapidly than the present forecast indicates. Flash is starting to replace EPROM/OTP devices in many applications including data processing, telecom, industrial, and automotive.

This trend will accelerate, especially for higher densities. The future of EPROMs for densities above 16Mb is rather bleak.

EEFROM

The low densities is where the action is and is expected to continue to be. Consumer applications are driving this market segment, which is expected to experience significant unit growth. As with all silicon for consumer applications, the ASPs will be low. The highdensity EEPROMs (parallel devices with densities above 1Mb) are dead. No activity is expected here because flash memories effectively perform the same ftmction at a fraction of the cost.

ROM

The ROM market segment is the least volatile and is expected to sustain reasonable growth rates. The applications are still driven by consumer electronics such as games. No replacement technology appears in the horizon at this time, and ROMs still represent the lowest-cost

(albeit least-flexible) solution.

Flash

Flash is the brightest spot within nonvolatile memories and will replace high-density

©1992 Dataquest Incoipoiated August—Reproduction Prohibited

Memories Worldwide

EEPROMs and EPROMs. They should be used in some portable applications in lieu of

DRAMs/P-SRAMs. ASPs are dropping rapidly and should cross over DRAM price per bit within the forecast's horizon. revenue to U.S. dollar amounts. The following table outlines these rates for 1989 through

1991.

Exchange Rates

As mentioned previously, Dataquest utilizes an average annual exchange rate in converting

Japan CYenAJ.S.$)

France (FrancAJ.S.$)

Germany (Deutsche Mark/U.S.$)

United Kingdom (U.S.$/Pound

Sterling)

1989 1990 1991

138 144 136

6.39 5.44 5.64

1.88 1.62 1.66

1.50 1.79 1.77

i

©1992 Dataquest Incoipoiated August-Reproduction Proliibited

i

Worldwide MOS Memorf Forecast

At the same time, the aftermarket for DRAMs has also grown, which appears to be buffering the industry from the strong upturns and downturns experienced in the past. The installation of the STA in 1986, we believe, has helped moderate the aggressive price cutting in the down cycle, and kept production costs dose to market prices.

Graphics applications are becoming major forces driving the market. However, these applications will see their greatest growth period after the 16Mb comes into volume production in the mid-1990s.

SRAM Forecast Assumptions

Hie following are our SRAM forecast assumptions:

• Historical trends will tend to repeat themselves over the next five years. These drive the following:

- Market composition by density of device

- Price per bit, and relative PPB for various densities

- Migration toward faster devices

- Migration toward wider parts

• PC caches will continue to consume the lion's share of fast SRAMs, even as battery-operated PCs grow in stature.

• Static RAM ASPs will track dynamic RAM

ASPs, however, slower SRAMs will be sold at a bargain as the demand for faster parts tends to obsolete speeds slow^er than 100ns.

• Pseudo-static RAMs will grow in acceptance from their limited stance today.

• TTiere will continue to be a speed gap in aU SRAM densities where sales wiU be low.

• Applications for slow SRAMs will continue to be far-flimg.

• There wiU be a cyclical softening of the market in 1995.

• The economy will strongly impact the bitgrowth rate of slow SRAMs and the ASPs of fast SRAMs.

• The SRAM market will grow more quickly than will the DRAM market, but will continue to stay the significantly smaller of the two.

• The SRAM market is not seriously threatened by new technologies such as flash, cached DRAM, Rambus, and microprocessors with on-board cache. These technologies will coexist with SRAM, and may even create the opportunity for new SRAM applications.

Nonvolatile Memory Forecast

Assumptions

EPROM

This market segment will remain relatively flat for the forecast period and may in faa drop off more rapidly than the present forecast indicates. Flash is starting to replace EPROM/OTP devices in many applications including data processing, telecom, industrial, and automotive.

This trend will accelerate, especially for higher densities. The future of EPROMs for densities above 16Mb is rather bleak.

EEPROM

The low densities is where the action is and is expected to continue to be. Consumer applications are driving this market segment, which is expected to experience significant unit growth. As with all silicon for consumer applications, the ASPs will be low. The highdensity EEPROMs (parallel devices with densities above 1Mb) are dead. No activity is expected here because flash memories effectively perform the same function at a fiaction of the cost.

ROM

The ROM market segment is the least volatile and is expected to sustain reasonable growth rates. The applications are still driven by consumer electronics such as games. No replacement technology appears in the horizon at this time, and ROMs still represent the lowest-cost

(albeit least-flexible) solution.

Flash

Flash is the brightest spot within nonvolatile memories and will replace high-density

®1992 Dataquest Incorporated August—Keproduction Prohibited

Memories Worldwide

EEPROMs and EPROMs. They should be used in some portable applications in lieu of

DRAMs/P-SRAMs. ASPs are dropping rapidly and should cross over DRAM price per bit within the forecast's horizon. revenue to U.S. dollar amounts. The following table outlines these rates for 1989 through

1991.

Exchange Rates

As mentioned previously, Dataquest utilizes an average annual exchange rate in converting

Japan CVenAJ.S.$)

France (FrancAJ.S.$)

Gennany (Deutsche MarkAJ.S.$)

United Kingdom (U.S.$/Pound

Sterling)

1989 1990 1991

138 144 136

6.39 5.44 5.64

1.88 1.62 1.66

1.50 1.79 1.77

©1992 Dataquest Incorporated August-Reproduction Piobibited

i

Worldwide MOS Memory Forecast

Table 1-1

Factory- Revenue from Shipments o f MOS Memory to the World, 1989-1996

CMilUons of U^. Dollars)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

DRAM 8,323.4 6,456.6 6,849.4 7,802.6 9,936.9 12,220.8 10,415.1 11,769.2 11.4

EEPROM 319.6 292.1 326.2 373.8 446.2 467.5 430.4 420.8 5.2

EPROM 1,809.1 1,445.8 1,362.4 1,275.5 1,381.8 1,318.9 1,270.4 1,183.4 -2.8

Flash 11.1 35.3 119.6 273.8 557.5 1,199.9 1,716.8 1,989-7 75.5

ROM 1,069.2 1,131.7 1,197.6 1,277.2 1,343.4 1,415.5 1,571.1 1,692.4 7.2

SRAM 3,329.1 2,433.6 2,569-3 2,811.1 3,722.4 4,435.8 5,538.2 6,8l6.1 21.5

Total/Average l4,86l.6 11,775.2 12,424.6 13,813.9 17,388-1 21,058.4 20,941.9 23,871.4 14.0

Percent Change (%) 22.0 -20.8 5.5 11.2 25.9 21.1 -0.6 14.0

Soufce: Dataquest (August 1992)

®1992 Dataquest Incorporated August—^Reproduction Prohibiled

Memories 'VRn-ldwide

Table 1-2

Shipments of MOS Memory to the World, 1989-1996

(MilUons of Units)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

DRAM 1,254.3 1,335.7 1,285.9 1,377.8 1,425.0 1,445.0 1,156.0 1,156.0 -2.1

EEPROM 118.3 127.1 212.9 281.6 349.9 422.9 470.7 523.6 19.7

EPROM 402.1 424.0 476.0 480.3 476.0 463.7 429.0 386.7 -4.1

Flash 0.6 2.7 11.8 33.0 77.3 163.0 268.7 349.8 97.0

ROM 299.4 315.4 383.3 367.0 370.0 318.6 276.7 255.4 -7.8

SRAM 630.5 620.5 703.6 759.1 766.2 787.1 797.3 914.6 5.4

Total/Average 2,705.1 2,825.3 3,073.6 3,298.8 3,464.4 3,600.2 3,398.4 3,586.1 3.1

Percent Change (%) 7.7 4.4 8.8 7.3 5.0 3.9 -5.6 5.5

Source: Dataquest (August 1992)

©1992 Dataquest Incoipoiated August—ReprcxluctiQn Prohibited

Table 1-3

Average SeUing Price for Shipments of MOS Memory to the World, 1989-1996

CUS. Dollars)

DRAM

1989 1990 1991 1992 1993 1994 1995

CAGR (%)

1996 1991-1996

6.64 4.82 5.33 5.66 6.97 8.46 9.01 10.18

13.8

2.70 2.30 1.53 1.33 1-28 1.11 0.91 0.80 -12.1

EEPROM

EPROM

4.50 3.41 2.86 2.66 2.90 2.84 2.96 3.06

1.3

Flash

17.21 13.23 10.13 8.30 7.21 7.36 6.39 5.69

-10.9

ROM

3.57 3.59 3.12 3.48 3-63 4.44 5.68 6.63 16.2

SRAM

Worldwide MOS Memory Forecast

5.28 3.92 3.65 3.70 4.86 5.64 6.95 7.45 15.3

Total/Average

Percent Change (%)

Source Dataquest (August 1992)

5.49

4.17 4.04 4.19 5.02 5.85 6.16

6.66

13.3 -24.1 -3.0 3.6 19-9 l6.5 5.4

8.0

10.5

©1992 Dataquest Incorporated August—Keproduction Prohibited

10 Memories Worldwide

Table 1-4

Shipments of MOS Memory to the World, 1989-1996

(Trillions of Bits)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

DRAM 641.6 964.1 1,535.0 2,707.7 4,510.2 7,267.9 8,480.0 12,042.1 51.0

EEPROM 1.0 1.3 2.1 2.8 3.8 4.5 4.5 4.8 17.7

EPROM 132.6 182.4 221.0 305.2 438.2 578.9 678.2 730.9 27.0

Flash 0.3 1.7 10.4 37.3 117.9 433.8 1,047.3 1,706.4 177.2

ROM 425.5 681.5 954.7 1,204.1 1,760.5 2,353.9 3,661.3 5,329.6 41.0

SRAM 72.5 90.5 127.0 237.7 425.9 661.6 1,055.6 1,611.2 66.2

TotaVAverage 1,273.6 1,921.5 2,850.3 4,494.8 7,256.6 11,300.8 14,926.9 21,425.1 49.7

Percent Change (%) 45.1 50.9 48.3 57.7 61.4 55.7 32.1 43.5

Sovuce: E>ataquest CAugust 1992)

©1992 Dataquest Incoipoiated August—Reproduction Prohibited

Wotldwide MOS Memory Forecast 11

Table 1-5

Price per Bit for Shipments of MOS Memory to the World, 1989-1996

OVUcro Dollars)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

DRAM 13.0 6.7 4.5 2.9 2.2 1.7 1.2 1.0 -26.2

EEPROM 311.6 224.0 153.8 135.3 116.0 102.9 96.2 87.9 -10.6

EPROM 13.6 7.9 6.2 4.2 3.2 2.3 1.9 1.6 -23.5

Flash 39.9 21.0 11.5 7.3 4.7 2.8 1.6 1.2 -36.7

ROM 2.5 1.7 1.3 1.1 0.8 0.6 0.4 0.3 -24.0

SRAM 45.9 26.9 20.2 11.8 8.7 6.7 5.2 4.2 -26.9

Total/Average 11.7 6.1 4.4 3.1 2.4 1.9 i . | 1.1 -24.2

Percent Change (%) -15.9 -47.5 -28.9 -29.5 -22.0 -22.2 -24.7 -20.6

Source: E>ataquest CAugust 1992)

®1992 Oaiaquest Incorporated August—Reproduction Prohibited

12 Memories Worldwide

Table 2-1

Factory Revenue £rom Shipments of DRAMs t o the World, 1989-1996

(Millions of U.S. Dollars)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

64K 112.3 38.1 20.0 0 0 0 0 0

256K 2,445.5 1,323.4 620.7 345-8 175.0 123.3 73.8 39.9 -42.2

1Mb 5,601.6 4,231.0 3,776.0 2,415.0 1,395.0 812.5 379.5 295.8 -39.9

4Mb 164.0 844.1 2,400.0 4,719.8 6,511.9 6,105.0 3,685.0 2,576.0 1.4

16Mb 0 0 32.7 322.0 1,855.0 5,180.0 6,270.0 8,482.5 204.0

64Mb 0 0 0 0 0 0 6.8 375.0

Total/Average 8,323.4 6,436.6 6,849.4 7,802.6 9,936.9 12,220.8 10,415.1 11,7692 11.4

Percent Change (%) 23.8 -22.7 6.4 13.9 27.4 23.0 -14.8 13.0

Source: Dataquest (August 1992)

©1992 Dataquest Incorporated August—Reproduction Prohibited

^

I

Worldwide MOS Memory Forecast 13

Table 2-2

Shipments of DRAMs to the World, 1989-1996

(MiUlons of Units)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

64K 65.3 25.7 13.5 0 0 0 0 0

2 5 6 K 780.1 620.5 299.1 190.0 125.0 85.0 41.0 21.0 - i l . 2

1Mb 407.5 665.0 835.4 750.0 450.0 250.0 115.0 87.0 -36.4

4Mb 1.3 24.4 137.7 435.0 815.0 925.0 670.0 460.0 27.3

16Mb 0 0 0.1 2.8 35.0 185.0 330.0 585.0 435.8

64Mb 6 i) 6 0 0 0 0 3-0

Total/Average 1,254.3 1,335.7 1,285.9 1,377.8 1,425.0 1,445.0 1,156.0 1,156.0 -2.1

Percent Change (%) -3.2 6.5 -3-7 7.1 3.4 1.4 -20.0 0

Sotuce: Dataquest (August 1992)

©1992 Dataquest Incoipoiated August—Seproduction Prohibited

14 Memories 'Worldwide

Table 2-3

Average Sellii^ Price for Shipments o f DRAMs to the World, 19S9-1996

CU.S. Dollars)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

64K 1.72 1.48 1.48 . - -

256K 3.13 2.13 2.08 1.82 1.40 1.45 1.80 1.90 -1.7

1Mb 13.74 6.36 4.52 3.22 3.10 3.25 3.30 3.40 -5.5

4Mb 125.01 34.59 17.43 10.85 7.99 6.60 5.50 5.60 -20.3

16Mb - - 246.60 115.00 53.00 28.00 19.00 14.50 -43.3

64Mb - - - - . - 225.00 125.00

Total/Average 6.64 4.82 5.33 5.66 6.97 8.46 9.01 10.18 13.8

Percent Change (%) 27.9 -27.4 10.5 6.3 23.1 21.3 6.5 13.0

Source Dataquest (August 1992)

©1992 Dataquest Incorporated August—Reproduction Prohibited

Wotldwlde MOS Memory Forecast 15

Table 2-4

Shipments of DRAMs t o the World, 1989-1996

(Trillions of Bits)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

64K 4.3 1.7 0.9 0 0 0 0 0

256K 204.5 162.7 78.4 49.8 32.8 22.3 10.7 5.5 -41.2

1Mb 427.3 697.3 876.0 786.4 471.9 262.1 120.6 91.2 -36.4

4Mb 5.5 102.4 577.4 1,824.5 3,418.4 3,879.7 2,810.2 1,929-4 27.3

16Mb 0 0 2.2 47.0 587.2 3,103.8 5,536.5 9,814.7 435.8

64Mb 0 0 0 0 0 0 2.0 2013

Total/Average 641.6 964.1 1,535.0 2,707.7 4,510.2 7,267.9 8,480.0 12,042.1 51.0

Percent Change (%) 33.1 50.2 59.2 76.4 (££ 61.1 l6.7 42.0

Source: Oataquest (August 1992)

©1992 Dataquest Incoiporated August—Bepioduction Prohibited

16 Memories Worldwide

Table 2-5

Price per Bit for Shipments of DRAMs to the World, 1989-1996

(Micro Dollars)

CAGR C%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

64K 26.2 22.6 22.5 - . . . .

256K 12.0 8.1 7.9 6.9 5.3 5.5 6.9 7.2 -1.7

1Mb 13.1 • 6.1 4.3 3.1 3.0 3.1 3.1 3.2 -5.5

4Mb 29.8 8.2 4.2 2.6 1.9 1.6 1.3 1.3 -20.3

16Mb - - 14.7 6.9 3.2 1.7 1.1 0.9 -43.3

64Mb .

.

.

_ .

. 3.4 1.9

Total/Average 13.0 6.7 4.5 2.9 2.2 1.7 1.2 1.0 -26.2

Percent Change (%) -7.0 -48.5 -33.2 -35.4 -23.5 -23.7 -27.0 -20.4

Source: Dataquest (Augtist 1992)

©1992 Dataquest Incorporated August—Reproduction Prohibited.

Worldwide MOS Memoir Forecast

17

Table 3-1

Factory Revenue Crom Shipments of SRAMs to the World, 1989-1996

(MllUons o f U.S. Dollars)

16K 10-19ns

CAGR (%)

1989 1990 1991 1992 1993

1994 1995 1996 1991-1996

0 5.1 2.0

1.0 0.7 0.5 0.3

-42.4

16K 20-44ns 33.4 16.6 8.8 5.3 3.5 2.2

-41.9

16K 45-70ns

13.0 6.7 4.0 3.0 2.3 1.6 -34.2

16K >70ns

16K <70ns

64K 0-9ns

64K 10-19ns

64K 20-44ns

64K 45-70ns

64K <70ns

143.3 77.2

341.8 116.9 63.1 17.2 13.3 7.5 5.5

5.4

0

0

-38.9

15.5 30.2 25.9 8.9 3.1

42.4 23.6 20.3 15.5 8.7 5.0 -34.7

0 235.2 88.6 79.1 54.0 28.0 I6.O

-41.6 0

0 62.6 45.9 29.8 22.0 12.0 7.2

-35.1

480.0 380.2

516.5 367.8 316.6 230.5 49.6 47.0 24.2 15.1

-45.6

64K >70ns

64K >70ns PSRAM

10.0 3.2 6.3 5.6 2.8 1.4 0.1

256K 0-9ns 98.2 130.3 173.7 114.5 80.4

46.4 ' 74.4 92.7 165.6 93-9 84.4

2 5 6 K 10-19ns

2 5 6 K 20-i4ns

2 5 6 K 45-70ns

0 273.1 338.0 291.5 423.9 279.7 205.0

57.5 98.2 104.3 151.8 104.3 83.0

12.7

-5.6

7.6

2 5 6 K <70ns

2 5 6 K >70ns

256K >70ns PSRAM

1Mb 0-9ns

349.6 341.2

6

1,138.9 703.2 718.2 570.3 557.4 255.4 119.2 125.6 - ^ . 4

173.6 97.5 162.0 109.5 138.7 103.2 6O.I 20.4 -33-9

0 18.0 75.0 139.0

(CcKKinuedD

®1992 Dataquest Incorporated August—Reproduction Profaibtted

18 Memories Worldwide

Table 3-1 (Continued)

Factory Revenue Crom Shipments of SRAMs to the World, 1989-1996

(Millions o f U.S. Dollars)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

1Mb 10-19ns 0 0 110.2 78.5 87.3 149.2 321.4 405.4 29.8

• 1Mb 20-44ns 0 0 45.5 l6l.2 285.2 338.4 924.9 961.8 84.1

1Mb 45-70ns 0 0 58.9 154.0 l6l.4 187.9 373.3 373.0 44.6

1Mb <70ns 0 68.8 0 0 0 6 0 0

1Mb >70ns 105.1 191.0 244.2 548.5 782.3 732.5 514.3 737.4 24.7

1Mb >70ns PSRAM 70.2 86.6 6l.2 51.3 71.5 79.6 84.2 83.5 6.4

4Mb 0-9ns 0 0 0 .

0 9 0 53.1 148.8

4Mb 10-19ns 0 0 0 1.0 27.5 71.0 253-5 491.0

4Mb 20-44ns 0 0 0 0.8 31.7 69.9 276.2 504.3

4Mb 45-70ns 0 0 0 9.5 246.8 201.6 588.1 709.8

4Mb <70ns 0 0 0 19.0 399.0 976.2 1,025.2 1,185.2

4Mb >70ns PSRAM 0 0 14.4 46.7 75.9 143.7 136.2 190.7 67.8

16Mb >70ns 0 0 0 0 0 0 4.0 132.3

16Mb >70ns PSRAM 0 0 0 0 0 12.0 43.6 99.1

Total/Average 3,329.1 2,433.6 2,569.3 2,811.1 3,722.4 4,435.8 5,538.2 6,816.1 21.5

Percent Change C%) 43.7 -26.9 5.6 9.4 32.4 19.2 24.9 23.1

Sotirce: Dataquest (August 1992>

©1992 Dataquest Incoipoiated August—Reproduction Prohibited

Worldwide MOS Memory Forecast

19

Table 3-2

Shipments of SRAMs to the World, 1989-1996

(Millions of Units)

16K 10-19ns

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

1.5 0.7 0.4 0.3 0.2 0.2 -36.4

16K 20-44ns

16K 45-70ns

16K <70ns

16K >70ns

64K 0-9ns

64K 10-19ns

0 13.9 7.4 4.4 3.3 2.5 1.8

0 10.4 5.4 3.2 2.4 1.8 1.3

-33.9

-34.2

40.9 31.7

150.9 91.0 90.2 21.5 13-3 7.5 3.6 2.1 -52.7

0.8 2.0 2.6 1.9 1.1

0 10.2 7.4 7.4 6.2 3.5 2.2

-26.1

0 64.8 46.6 39.5 30.0 14.7 8.6 -33.2

64K 20-44ns

64K 45-70ns 0 29.6 27.0 18.1 12.9 6.7 4.0 -33.0

64K <70ns

64K >70ns

64K >70ns PSRAM

59.5 69.2

190.4 199.6 185.9 159-0 38.2 29.4 14.2 8.4 -46.2

2 5 6 K 0-9DS

3.5 1.7 3.7 3.8 2.2 1.1 0.1

0.9 2.9 11.6 11.4 13.4

2 5 6 K 10-19ns

2 5 6 K 20-44ns

256K 45-70ns

2 5 6 K <70ns

256K >70ns

256K >70ns PSRAM

1Mb 0-9ns

0 1.9 6.0 11.6 25.5 22.9 24.1

0 28.1 82.4 95.6 141.3 93.2 68.3

66.1

19.4

0 10.8 29.8 34.8 53.3 36.0 28.1

12.1 21.4

21.1

145.7 168.4 179.5 203.7 202.7 94.6 42.6 42.6

•m&

22.2 23.3 40.5 54.8 69.4 59.0 35.4 11.3 -22.5

0.4 3.6 8.7

CCondnusD

©1992 Dataquest Incorporated August—Reproduction Prohibited

20 Memories Worldwide

Table 3-2 (Continued)

Shipments o f SRAMs to the World, 1989-1996

(Millions of Units)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

1Mb 10-19ns 0 0 0.6 1.2 4.0 10.4 35.7 57.9 153.7

1Mb 20-44ns 0 0 1.0 8.9 24.8 39.6 132.1 156.4 176.4

1Mb 45-70ns 0 0 2.3 14.0 20.8 29.6 66.7 (££ 95.8

1Mb <70ns 0 1.0 0 0 0 0 0 0

1Mb >70ns 1.4 6.2 18.2 59-6 120.3 124.1 102.9 149.0 52.2

1Mb >70ns PSRAM 3.9 7.0 9.4 13.7 26.0 29.5 30.1 28.3 24.6

4Mb 0-9ns 0 0 0 0 0 0 0.5 2.0

4Mb 10-19ns 0 0 0 0.0 0.2 1.1 7.2 21.8

4Mb 2a44ns 0 0 0 0.0 0.4 2.3 12.6 30.8

4Mb 45-70ns 0 0 0 0.1 3.7 7.2 28.0 44.6

4Mb >70ns 0 0 0 0.2 9.5 37.5 60.3 87.8

4Mb >70ns PSRAM 0 0 1.1 4.4 10.8 24.0 24.3 34.7 100.9

16Mb >70ns 0 0 0 0 0 0 0.0 1.3

16Mb >70ns PSRAM 0 0 0 0 0 0.5 2.5 7.1

Total/Average 630.5 620.5 703.6 759.1 766.2 787.1 797.3 914.6 5.4

Percent Change (%) 23.0 -1.6 13.4 7.9 .9 2.7 1.3 14.7

Source: Dataquest (August 1992)

©1992 Dataquest Incorporated August—Reproduction Prohibited

'\K%>rldwide MOS Memory Forecast 21

Table 3-3

A v e r s e Selling Price for Shipments o f SRAMs to the World, 1989-1996

CU.S. Dollars)

16K 10-19ns i9Sy 1990 1991 1992 1993 1994

3.30 3.00 2.50 2.25 2.10 2.00

CAGR (%)

1995 1996 1991-1996

-9.5 l^K 20-44ns

2.40 2.25 2.00 1.60 1.40 1.25 -12.2

16K 45-70ns

1.25 1.25 1.25 1.25 1.25 1.25 0.0

16K <70ns

16K >70ns

3.50 2.44

29.0

64K 0-9ns

2.27 1.28 0.70 0.80 1.00 1.00 1.50 2.50

19.00 15.00 10.00 4.75 2.80

-11.6

64K 10-19ns

64K 20-44ns

418 3.20 2.75 2.50 2.50 2.25

3.63 1.90 2.00 1.80 1.90 1.85

64K 45-70ns 2.11 1.70 1.65 1.70 1.80 1.80

-12.6

-3.2

64K <70ns

6 4 K >70ns

64K >70ns PSRAM

8.07 5.49

2.71 1.84 1.70 1.45 1.30 1.60 1.70 1.) 1.1

2.85 1.88 1.72 1.45 1.30 1.20 1.30

110.00 45.00 15.00 10.00 6.00 256K 0-9ns

256K 10-19ns

256K 20-44ns

24.31 12.50 8.00 6.50 4.10 3.50

9.71 4.10 3.05 3.00 3.00 3.00

-32.1

-20.9

256K 45-70ns

5.33 3.30 3.00 2.85 2.90 2.95 -11.1

256K <70ns

2 5 6 K <70ns

2 5 6 K >70ns PSRAM

1Mb 0-9ns

28.97 15.94

7.82 4.18 4.00 2.80 2.75 2.70 2.80 2.95

7.84 4.19 4.00 2.00 2.00 1.75 1.70 1.80

-*?

-14^

45.00 21.00 16.00

CCcmtiaiieO

©1992 Dataquest Incorporated August—^Reproduction Prohibited

22 Memories Worldwide

Table 3-3 (Continued)

A v e n ^ Selling Price for Shipments of SRAMs to the World, 1989-1996

(U.S. Dollars)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

1Mb 10-19ns - - 200.00 65.00 22.00 14.35 9.00 7.00 -48.9

1Mb 20-44ns - - 46.94 18.05 11.50 8.55 7.00 6.15 -33.4

1Mb 45-70ns - - 25.47 11.00 7.75 6.35 5.60 5.60 -26.1

1Mb <70ns - 68.78 - - - -. _ -

1Mb >70ns 74.30 30.76 13.39 9.20 6.50 5.90 5.00 4.95 -18.0

1Mb >70ns PSRAM 17.85 12.29 6.50 3.75 2.75 2.70 2.80 2.95 -14.6

4Mb 0-9ns - - - - - - 110.00 75.00

4Mb 10-19ns - - - 500.00 125.00 67.00 35.00 22.50

4Mb 20-44ns - - - 250.00 72.00 30.00 22.00 l6.40

4Mb 45-70ns - - - 100.00 66.00 28.00 21.00 15.90

4Mb >70ns - - - 95.00 42.00 26.00 17.00 13.50

4Mb >70ns PSRAM - - 13.54 10.50 7.00 6.00 5.60 5.50 -16.5

16Mb >70ns - - - » _ - 180.00 99.50

16Mb >70ns PSRAM .

.

. - . . 26.IO 17.15 14.00

Total/Average 5.28 3.92 3.65 3.70 4.86 5.64 6.95 7.45 15.3

Percent Change (%) 16.8 -25.7 -6.9 1.4 31.2 I6.O 23.3 7.3

Source: Dataquest (August 1992)

©1992 Dataquest Incorporated August—Reproduction Prohibited

Worldwide MOS Memoiy Forecast

23

Table 3-4

Shipments of SRAMs to the World, 1989-1996

(Trfflions of Bits)

16K 10-19ns

1989 1990 1991 1^92 1993 1994 1995

CAGR (%)

1996 1991-1996

0 0

-36.4

16K 20-44ns

16K 45-70ns

16K <70ns

0.7 0.5

0.2 0.1 0.1 0.1

0.2 0.1 0.1

0

0

0

0

0

0

-33.9

-34.2

1 6 K >70ns

64K 0-9ns

64K 10-19ns

64K 20-44ns

64K 45-70ns

2.5 1.5 1.5 0.4 0.2 0.1 0.1

0.1 0.1 0.2

0.1

0.7 0.5 0.5 0.4

0.2

4 2 3.1 2.6 2.0 1.0

0

0.1

0.1

0.6

-52.7

-26.1

-33.2

-33.0

64K <70ns

3.9 4.5

1.9 1.8 1.2 0.8 0.4

0

0.3

0

64K >70ns

64K. >70ns PSRAM

12.5 13.1 12.2 10.4 2.5 1.9

0.2 0.1 0.2 0.3 0.1 0.1

0.9

0

0.5

0

-46.2

2 5 6 K 0-9ns

0.2 0.8 3.0

3.0

0 0 0.5 1.6 3.0 6.7 6.0

3.5

6.3

2 5 6 K 10-19ns

256K 20-44ns

66.1

19.4

2 5 6 K 45-70ns

0 0 7.4 21.6 25.1 37.0 24.4

0 2.8 7.8 9.1 14.0

9A

3.2 5.6

0 0

0

17.9

7.4

0

21.1

2 5 6 K <70ns

2 5 6 K >70ns

256K >70ns PSRAM

1Mb 0-9as

38.2 44.1 47.1 53.4 53.1 24.8 11.2

5.8 6.1 10.6 144 18.2 15.5

0.4

9.3

3.7

11.2

3.0

9.1

-25.0

-22.5

(Continued)

©1992 Dataquest Incorporated August—^Reproduction Prohibited

24 Memories Woridwide

Table 3-4 (Continued)

Shipments of SRAMs t o the World, 1989-1996

(TrilUons of Bits)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

1Mb 10-19ns 0 0 0.6 1.3 4.2 10.9 37.4 60.7 153.7

1Mb 20-44ns 0 0 1.0 9.4 26.0 41.5 138.5 164.0 176.4

1Mb 45-70ns 0 0 2.4 14.7 21.8 31.0 69.9 69.8 95.8

1Mb <70ns 0 1.0 0 0 0 0 0 0

1Mb >70ns 1.5 6.5 19.1 62.5 126.2 130.2 107.9 156.2 52.2

1Mb >70ns PSRAM 4.1 7.4 9.9 14.4 27.3 30.9 31.5 29.7 24.6

4Mb 0-9ns 0 0 0 0 0 0 2.0 8.3

4Mb 10-19ns 0 0 0 0 0.9 4.4 30.4 91.5

4Mb 20-44ns 0 0 0 0 1.8 9.8 52.7 129.0

4Mb 45-70ns 0 0 0 0.4 15.7 30.2 117.5 187.2

4Mb >70ns 0 0 0 0.8 39.9 157.5 252.9 368.2

4Mb >70ns PSRAM 0 0 4.4 18.7 45.5 100.5 102.0 145.4 100.9

16Mb >70ns 0 0 0 0 0 0 0.4 22.3

16Mb >70ns PSRAM 0 0 0 0 0 7.7 42.7 118.7

Total/Average 72.5 90.5 127.0 237.7 425.9 661.6 1,055.6 1,611.2 66.2

Percent Change (%) 67.3 24.9 40.3 87.1 79.2 55.3 59-6 52.6

Soufce: Dataquest (August 1992>

©1992 Dataquest Incoipoiated August—Reproduction Piohibited

Woridwlde MOS Memory Forecast

25

Table 3-5

Price p e r Bit for Shipments of SBAMs to the World, 1989-1996

(Micro Dollars)

16K 10-19ns

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

201.4 183.1 152.6 137.3 128.2 122.1

-9.5

146.5 137.3 122.1 97.7 85.4 76.3 -12.2 16K 2(>44ns

16K 45-70ns

16K <70ns

76.3 76.3 76.3 76.3 76.3 76.3

213.8 148.9

138.3 78.4 42.7 48.8 6l.O 6l.O 91.6 152.6 29.0

I 6 K >70ns

64K 0-9ns

64K 10-19ns

289.9 228.9 152.6 72.5 42.7

63.7 48.8 42.0 38.1 38.1 34.3 -11.6

-12.6

64K 2a44ns

64K 45-70ns

55.3 29.0 30.5 27.5 29.0 28.2

32.3 25.9 25.2 25.9 27.5 27.5

-3.2

64K <70ns

123.1 83.8

41.4 28.1 26.0 22.1 19.8 24.4 25.9 27.5

1.1

64K >70ns

64K >70ns PSRAM

43.5 28.6 26.2 22.1 19.8 18.3 19.8

256K 0-9ns

419.6 171.7 57.2 38.1 22.9

92.7 47.7 30.5 24.8 15.6 13.4

-32.1

2 5 6 K 10-19ns

2 5 6 K 2a44ns

37.1 15.6 11.6 11.4 11.4 11.4

20.3 12.6 11.4 10.9 11.1 11.3

-20.9

-11.1

2 5 6 K 45-70ns

256K <70ns

110.5 60.8

2 5 6 K >70ns

256K >70ns PSRAM

1Mb 0-9ns

29.8 15.9 15.3 10.7 10.5 10.3 10.7 11.3

29.9 16.0 15.3 7.6 7.6 6.7 6.5 6.9

-5.9

-14.8

42.9 20.0 15.3

(Continued)

©1992 Dataquest Incorporated August—Keproduction Frobibited

26 Memories Worldwide

Table 3-5 (Continued)

Price per Bit for Shipments o f SRAMs to the World, 1989-1996

OMicro Dollars)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

1Mb 10-19IK - - 190.7 62.0 21.0 13.7 8.6 6.7 -48.9

1Mb 20-44ns - - 44.8 17.2 11.0 8.2 6.7 5.9 -33.4

1Mb 45-70ns - - 24.3 10.5 7.4 6.1 5.3 5.3 -26.1

1Mb <70ns - 65.6 .

.

.

.

1Mb >70ns 70.9 29.3 12.8 8.8 6.2 5.6 4.8 4.7 -18.0

1Mb >70ns PSRAM 17.0 11.7 6.2 3.6 2.6 2.6 2.7 2.8 -14.6

4Mb 0-9ns .

.

.

.

.

. 26.2 17.9

4Mb 10-19ns - - - 119.2 29.8 16.0 8.3 5.4

4Mb 20-44ns - - - 59.6 17.2 7.2 5.2 3.9

4Mb 45-70ns - - - 23.8 15.7 6.7 5.0 3.8

4Mb >70ns - - - 22.6 10.0 6.2 4.1 3.2

4Mb >70ns PSRAM - - 3.2 2.5 1.7 1.4 1.3 1.3 -16.5

16Mb >70ns .

.

.

.

- 10.7 5.9

16Mb >70ns PSRAM _ _ .

.

. 1.6 1.0 0.8

Total/Average 45.9 26.9 20.2 11.8 8.7 6.7 5.2 4.2 -26.9

Percent Change (%) -14.1 -41.5 -24.7 -41.5 -26.1 -23.3 -21.7 -19.4

Source: Dataquest (August 1992)

©1992 Dataquest Incorporated August—Reproduction Prohibited

Woridwlde MOS Memoiy Forecast Z7

Table 4-1

Factory Revenue from Shipments of EPROMs to the World, 1989-1996

(MiUions of U ^ . Dollars)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

1 6 K

22.9 12.6 11.7 10.8 9.5 8.2 7.0 5.3 -14.6

32K 40.3 18.2 25.3 21.4 17.1 14.3 10.8 8J: -20.2

64K 170.5 88.3 99.9 85.0 64.0 55.8 43.5 36.0 -18.5

128K 174.3 103.2 74.6 57.6 45.0 36.0 30.0 25.5 -19.3

256K 519.6 360.9 450.0 341.3 277.5 212.5 176.6 136.0 -21.3

512K 442.0 278.8 204.8 160.6 128.0 99.0 77.6 67-2 -20.0

1Mb 410.1 479.5 299.3 299.0 268.8 226.9 177.6 155.3 -12.3

2Mb 25.3 69.7 99.9 143.8 220.5 231.8 181.5 156.8 9A

4Mb 4.1 34.8 96.8 153.8 332.5 348.0 421.2 375.0 31.1

8Mb a & 0 2.4 19.0 75.0 101.3 142.5

16Mb 0: 6 6 0 0 11.4 43.5 75.6

Total/Average 1,809.1 1,445.8 1,362.4 1,275.5 1,381.8 1,318.9 1,270.4 1,183.4 -2.8

Percent Qiange (%) -5.4 -20.1 -5.8 -6.4 8.3 -4.5 -3.7 -6.9

Source: Dataqtiest (August 1992)

©1992 Dataquest Incorporated August—{teproduction Prohibited

28 Meaiorles Worldwide

Table 4-2

Shipments of EPROMs t o the World, 1989-1996

(MiUions of Units)

CAcai (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

16K 8.0 5.9 5.3 4.9 4.4 3.9 3.5 2.8 -11.8

32K 13.7 8.1 10.8 9.3 7.6 6.5 5.0 3-9 -18.5

64K 55.0 39.2 57.3 50.0 40.0 36.0 29.0 24.0 -l6.0

128K 54.0 45.5 41.6 36.0 29.0 24.0 20.0 17.0 -l6.4

256K 152.7 157.7 204.8 175.0 150.0 125.0 107.0 85.0 -l6.1

512K 84.1 98.3 71.9 73.0 64.0 55.0 47.0 42.0 -10.2

1Mb 33.7 63.8 66.3 92.0 96.0 89.0 74.0 69.0 0.8

2Mb 0.7 4.6 11.9 25.0 49.0 6l.O 55.0 49.0 32.6

4Mb 0.1 1.0 6.1 15-0 35.0 58.0 78.0 75.0 65.4

8Mb 0 0 0 0.1 1.0 5.0 9.0 15.0

16Mb . 0 0 0 0 0 0.3 1.5 4.0

Total/Average 402.1 424.0 476.0 480.3 476.0 463.7 429.0 386.7 -4.1

Percent Change (%) 11.9 5.5 12.3 0.9 -0.9 -2.6 -7.5 -9.9

Source: Dataquest (August 1992)

®1992 Dataquest Incorporated August—Reproduction Prohibited

>

Worldwide MOS Memory Forecast

29

Table 4-3

A v e r s e Selling Price for Shipments of EPROMs to the World, 1989-1996

CU.S. VoUars)

1 6 K

CAGR (%)

1989 1990 1991 1992 1993 1994 X99S 1996 1991-1996

2.15 2.24 2.20 2.15 2.10 2.00 1.90

-3.2

32K

64K

128K

2.95 2.24 2.33 2.30 2.25 2.20 2.15 2.10

3.10 2.25 1.74 1.70 1.60 1.55 1.50 1.50

-2.1

-3.0

256K

512K

3.23 2.27 1.79 1.60 1.55 1.50 1.50 1.50 -3.5

-6.2

3.40 2.29 2.20 1.95 1.85 1.70 1.65 1.60

5.25 2.84 2.85 2.20 2.00 1.80 1.65 1.60

-10.9

12.15 7.52 4.51 3.25 2.80 2.55 2.40 2.25 -13.0 1Mb

2Mb 33.80 15.25 8.37 5.75 4.50 3.80 3.30 3.20 -17.5

59.32 35.16 15.99 10.25 9.50 6.00 5.40 5.00 -20.7 4Mb

8Mb

16Mb

24.00 19.00 15.00 11.25 9.50

38,00 29.00 18.90

Total/Average

Percent Change (%)

Source: Dataquest (Augiut 1992}

4.50 . 3.41 2.86 2.66 2.90 2.84 2.96 3.06

-15.4 -24.2 -16.1 -7.2 9.3 -2.0 4.1 3.3

1.3

©1992 Dataquest Incorporated August—^Reproduction Prohibited

30 Memories Worldwide

Table 4-4

Shipments of EPROMs to the World, 1989-1996

(Trilltons of Bits)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

16K 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0 -11.8

32K 0.4 0.3 0.4 0.3 0.2 0.2 0.2 0.1 -18.5

64K 3.6 2.6 3.8 3.3 2.6 2.4 1.9 1.6 -16.0

128K 7.1 6.0 5.5 4.7 3.8 3-1 2.6 2.2 -l6.4

256K 40.0 41.3 53-7 45.9 39.3 32.8 28.0 22.3 -l6.1

512K 44.1 51.5 37.7 38.3 33.6 28.8 24.6 22.0 -10.2

1Mb 35.4 66.9 69.6 96.5 100.7 93.3 77.6 72.4 .8

2Mb 1.6 9.6 25.0 52.4 102.8 127.9 115.3 102.8 32.6

4Mb 0.3 4.1 25.4 62.9 146.8 243.3 327.2 314.6 65.4

8Mb 0 0 0 0.8 8.4 41.9 75.5 125.8

16Mb 0 0 0 0 0 5.0 25.2 67.1

Total/Average 132.6 182.4 221.0 305.2 438.2 578.9 678.2 730.9 27.0

Percent Change (%) 46.3 37.5 21.2 38.1 43.6 32.1 17.2 7.8

Source Dataquest (August 1992)

©1992 Dataquest Incoipoiated August—Seproduction Piohibited

2 5 6 K

512K

1Mb

2Mb

4Mb

8Mb

16Mb

Worldwide MOS Memoiy Forecast

31

Table 4-5

Price per Bit for Shipments o f EPROMs to the World, 1989-1996

(Micro Dollars)

16K

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

174.4 131.0 136.4 134.3 131.2 128.2 122.1 116.0

-3.2

32K 90.1 68.5 71.2 70.2 68.7 67.1 65.6 64.1

47.3 34.4 26.6 25.9 24.4 23.7 22.9 22.9

-2.1

-3.0

64K

128K

24.6 17.3 13.7 12.2 11.8 11.4 11.4 11.4 -3.5

13.0 8.7 8.4 7.4 7.1 6.5 6.3 6.1

-6.2

10.0 5.4 5.4 4.2 3.8 3.4 3.1

3.1

-10.9

11.6 7.2 4.3 3.1 2.7 2.4 2.3 2.1

-13.0

16.1 7.3 4.0 2.7 2.1 1.8 1.6

1.5

-17.5

14.1 8.4 3.8 2.4 2.3 1.4 1.3 1.2 -20.7

2.9 2.3 1.8 1.3 1.1

2.3 1.7 1.1

Total/Average

Percent Change (%)

Source: Dauquest (August 199:9

13.6 7.9 6.2 4.2 3.2 2.3 1.9 1.6

-35.3 -41.9 -22.2 -32.2 -24.6 -27.7 -17.8 -13.6

-23.5

©1992 Dataquest IiKorporated August—Reproduction Prohibited

32

Memories Worldwide

Table 5-1

Factory Revenue Cram Shipments of ROMs to the World, 1989-1996

(MiUions o f U^. Dollars)

16K

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

0.5

32K

64K

1.5 0.8 0.9

0.2

19-8 11.8 9.1 5.1

128K

11.9

8.4 13.1 4.6 2.2 1.2

256K

512K

1Mb

2Mb

4Mb

8Mb

16Mb

62.3 53.1 66.7 33.0 23.5 11.7 1.9 0.2 -67.6

109.5 62.7 53.0 38.8 20.9 6.7 1.8

0.2

-68.1

402.7 285.2 334.4 277.9 191.1 132.6 53.2 27.0

-39.5

168.0 200.0 179.3 134.8 116.3 85.5 63.8 36.5

-27.3

267.1 385.9 306.4 338.6 294.3 195.0 124.8 71.3

-25.3

16.3 92.3 213.7 280.2 370.2 361.5 385.2 313.2

7.9

9.6 31.4 21.1 55.5 150.5 239.3 257.9 318.5

72.2

32Mb

64Mb

0 108.5 161.0 245.1 367.2 383.0

128Mb

13.5 130.5 233.7 430.5

6.5 81.7 104.4

256Mb

0 7.5

Total/Average

Percent Change (%)

Source: Dataquest CAu^jUst 1992)

1,069.2 1,131.7 1,197.6 1,277.2 1,343.4 1,415.5 1,571.1 1,692.4

12.2 5.8 5.8 6.6 5.2 5.4 11.0

7.7

7.2

®1992 Dataquest Incorporated August—Reproduction Prohibited

'Worldwide MOS Memory Forecast 33

Table 5-2

Shipments o f ROMs to the "World, 1989-1996

CMillions of Units)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

16K 0:2 O 0 0 0 0 0 0

32K 0.9 0.5 0.4 0.1 0 0 0 0

64K 12.0 7.4 5.5 3.3 1.0 0 0 0

128K 7.0 5.1 8.7 3.2 1.8 1.1 0 0

2 5 6 K 34.6 31.6 37.7 22.0 17.4 9.0 1.5 0.2 -64.9

512K 43.8 26.7 26.6 19.9 11.0 3-6 1.0 0.1 -67.3

1Mb 120.2 103.7 130.9 123.5 91.0 66.3 28.0 15.0 -35.2

2Mb 39.1 52.6 57.9 49.0 46.5 38.0 29.0 17.4 -21.4

4Mb 39.9 74.2 76.5 91.5 107.0 78.0 52.0 31.0 -l6.5

8Mb 1.4 11.9 36.6 47.1 72.3 76.1 85.6 72.0 14.5

16Mb 0.3 1.7 2.5 4.3 14.7 29.0 38.2 49.0 80.9

32Mb 0 0 0 3.1 7.0 12.9 27.2 38.3

64Mb 0 0 0 0 0.3 4.5 12.3 28.7

128Mb 0 0 0 0 0 0.1 1.9 3.6

256Mb 0 0 0 0 0 0 0 0.1

Total/Average 299-4 315.4 383.3 367.0 370.0 318.6 276.7 255.4 -7.8

Percent Change (%) 22.8 5.4 21.5 -4.3 0.8 -13.9 -13.2 -7.7

Source: Dauquest (August 1992)

©1992 Dataquest Incoiporated August—Reproduction Prohibited

256K

512K

1Mb

2Mb

4Mb

8Mb

16Mb

32Mb

64V[b

34

Memories Worldwide

Table 5-3

Average Sellic^ Price for Shipments of ROMs to the World, 1989-1996

(U.S. Dollars)

16K

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

2.50

32K 1.60 1.60 2.29 2.20

64K

128K

128Mb

256Mb

1.65 1.60 1.65 1.55

1.70 1.65 1.51 1.45 1.20 1.12

1.80 1.68 1.77 1.50 1.35 1.30 1.25 1.20

2.50 2.35 1.99 1.95 1.90 1.85 1.80 1.75

3.35 2.75 2.55 2.25 2.10 2.00 1.90 1.80

4.30 3.80 3.10 2.75 2.50 2.25 2.20 2.10

6.70 5.20 4.00 3.70 2.75 2.50 2.40 2.30

-7.5

-10.5

-7.5

-2.6

•6.8

-5.7

-4.8

11.90 7.75 5.84 5.95 5.12 4.75 4.50 4.35

32.00 19.00 8.32 12.90 10.24 8.25 6.75 650

35.00 23.00 19.00 13.50 10.00

45.00 29.00 19.00 15.00

65.00 4300 29.00

75.00

TotaVAvefage

Percent Change (%)

Source: Dataquest (August 1992)

3.57 3.59 3.12 3.48 3.63 4.44 5.68 6.63

-8.6 0.5 -12.9 11.4 4.3 22.4 27.8 16.7

16.2

©1992 Dataquest Incoipoiated August—Reproduction Prohibited

4Mb

8Mb

16Mb

32Mb

2 5 6 K

512K

1Mb

2Mb

64Mb

128Mb

256Mb

Worldwide MOS Memory Forecast

35

Table 5-4

Shipments o f ROMs to the World, 1989-1996

CTriUions o f Bits)

16K

CAGR (%)

1989 1990 1991 1992 1993 1994 1995

1996 1991-1996

0 0

32K

64K

128K

0.8 0.5 0.4

0.2

0.1

0.2

0.9 0.7 1.1

0.4

0.1

9.1 8.3 9.9 5.8

23.0 14.0 13.9 10.4

4.6

5.8

2.4

0.4

0.1

0.1

1.9 0.5

126.0 108.8 137.3 129.5 95.4 69.5 29.4 15.7

•64.9

-67.3

-35.2

82.0 110.4 121.5 102.8 97.5 79.7 60.8 36.5

-21.4

167.2 311.3 321.0 383.8 448.8 327.2 218.1 130.0 -16.5

11.5 99.9 307.2 395.1 606.5 638.4 718.1 604.0

5.0 27.7 42.4 72.1 246.6 486.5 640.9 822.1

14.5

80.9

104.0 234.9 432.9 912.7 1,285.1

0 20.1 302.0 825.4 1,926.0

13.4 255.0 483.2

26.8

Total/Average

Percent Change (%)

Soufce: Dataquest (August 1992)

425.5 681.5 954.7 1,204.1 1,760.5 2,353.9 3,661.3 5,329.6

63.2 60.2 40.1 46.2 33.7 55.5 45.6

41.0

©1992 Dataquest Incorpoiated August—Septoduction Prohibited

36

Memories Worldwide

Table 5-5

Price per Bit for Shipments of ROMs to the World, 1989-1996

CMicro Dollars)

16K

CAGR (Vo)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

152.6

32K

48.8 48.8 70.0 67.1

64K

25.2 24.4 25.2 23.7

128K

13.0 12.6 11.5 11.1 9.2 8.5

256K

6.9 6A 6.8 5.7 5.1 5.0 4.8 4.6 -7.5

512K

1Mb

2Mb

4Mb

8Mb

16Mb

32Mb

4.8 4.5 3.8 3.7 3.6 3.5 3.4 3.3

3.2 2.6 2.4 2.1 2.0 1.9 1.8 1.7

-2.6

2.1 1.8 1.5 1-3 1.2 1.1 1.0 1.0

-6.8

-7.5

1.6 1.2 1.0 • 0.9 0.7 0.6 0.6 0.5

-10.5

1.4 0.9 0.7 0.7 0.6 0.6 0.5 0.5

-5.7

-4.8

1.9 1.1 0.5 0.8 0.6 0.5 0.4 0.4

1.0 0.7 0.6 0.4 0.3

0.7 0.4 0.3 0.2 64Mb

128Mb

256Mb

0.5 0.3 0.2

0.3

Total/Average

Percent Change

Source: Dataquest (August 1992)

2.5 1.7 1.3 1.1 0.8 0.6 0.4 0.3

-24.0

-31.2 -33.9 -24.5 -15.5 -28.1 -21.2 -28.6 -26.0

©1992 Dataquest IiKOtpoiated August—Reproduction Piohibited

Woildwide MOS Memory Forecast

37

Table 6-1

Factory Revenue from Shipments of EEPROMs to the World, 1989-1996

(Millions of U.S. Dollars)

256b

CAGR (%)

19S9 1990 1991 1992 1993 1994 1995 1996 1991-1996

27.4 16.9 17.8 15.0 13.2 11.2 8.1 5.9 -19.9

512b

5.6 8.3 10.0 10.2 11.0 11.5 15.6

45.1 48.7 62.1 63.7 67.5 64.4 47.6 33.4

IK

2K 39.2 38.3 51.9" 60.2 62.0 65.0 65.5 68.9

-11.7

5.8

4K 44.1 54.1 28.3 48.8 63.0 72.5 78.8 89.3 25.8

8K

I 6 K

64K

256K

512K

1Mb

0.4 1.9 3.0 6.6 9.3 9.5

31.3 27.1 31.9 30.6 36.3 48.2 64.8 78.0

68.8 59.3 74.1 83.1 92.3 89.7 90.8 101.5

60.5 45.0 48.1 46.5 50.0 59.0 28.7 17.0

90.7

19.6

6.5

-18.8

0.9

3.1 2.8 5.3 15.8 49.0 40.8 26.0 6.0 2.6

Total/Average

Percent Change (%)

Soufce: Dataquest (August 1992)

319.6 292.1 326.2 373.8 446.2 467.5 430.4 420.8

16.9 -8.6 11.7 14.6 19.4 4.8 -7.9 -2.2

5.2

®1992 Dataquest Incorporated August—Reproduction PFOfaibited

3S Memories Worldwide

Table 6-2

Shipments of EEPROMs to the World, 1989-1996

(MiUlons of Units)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

256b 34.2 22.5 24.8 23.0 22.0 20.0 18.0 15.0 -9.6

512b 0 0 6.8 11.0 14.3 17.0 20.0 23.0 27.6

IK 41.0 48.7 100.1 130.0 150.0 165.0 140.0 115.0 2.8

2K 15.7 17.4 38.9 57.3 77.5 100.0 131.0 153.0 31.5

4K 11.0 19.7 13.4 25.0 42.0 63.0 87.5 119.0 54.7

8K 0 0 0.2 0.9 1.5 3.8 6.2 9.5 129.3

I 6 K

7.1 7.3 11.3 13.6 17.7 25.5 37.0 52.0 35.6

64K 8.4 10.3 14.8 17.5 20.5 23.0 27.5 35.0 18.8

2 5 6 K 0.8 1.2 2.5 3.1 3.7 5.0 3.1 2.0 -4.1

512K 0 0 0 0 0 0 0 0

1Mb 0 0 0.1 0.2 0.7 0.6 0.4 0.1 11.9

Total/Average 118.3 127.1 212.9 281.6 349.9 422.9 470.7 523.6 19.7

Percent Change (%) 17.7 7.4 67.5 32.3 24.3 20.8 11.3 11.2

Soufce: Dataquest (August 1992)

©1992 Dataquest Incorporated August—Reproduction Prohibited

Worldwide MOS Memory Forecast

39

Table 6-3

A v e r s e Selling Price for Shipments of EEPROMs to the World, 1989-1996

CU.S. Dollars)

256b

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

0.80

0.75 0.72 0.65 0.60 0.56 0.45 0.39

-11.4

512b 0.82 0.75 0.70 0.60 0.55 0.50

-9A

-14.1

IK

2K

1.10 1.00 0.62 0.49 0.45 0.39 0.34 0.29

2.50 2.20 1.33 1.05 0.80 0.65 0.50 0.45 -19.5

4K

8K

4.00 2.75 2.11 1.95 1.50 1.15 0.90 0.75

2.51 2.25 2.00 1.75 1.50 1.00

-18.7

-16.8

I6K

64K

4.40 3.73 2.82 2.25 2.05 1.89 1.75 1.50

8.19 5.75 5.01 4.75 4.50 3.90 3.30 2.90

-11.8

-10.4

256K

512K

75.29 38.19 19.46 15.00 13.50 11.80 9-25 8.50

44.88

-15.3

1Mb 165.00 100.00 92.56 79.00 70.00 68.00 65.00 60.00

-8.3

Total/Average

Percent Change (%)

Source: Dataquest (August 1992)

2.70

2.30 1.53 1.33 1.28 1.11 0.91 0.80

-0.7 -14.9 -33.4 -13.4 -3.9 -13.3 -17.3 -12.1

-12.1

©1992 Dataquest Incorporated August—Reproduction Piobibiied

64K

256K

512K

1Mb

40

Memories Woridwlde

Table 6-4

Shipments of EEPROMs to the World, 1989-1996

(TrlUions o f Bits)

256b

CAGR (%)

1989 1990 1991 1992 1993

1994 1995 1996 1991-1996

0

-9.6

512b 27.6

IK

2K

4K

8K

1 6 K

0.1 0.1 0.2 0.2 0.1 0.1

2.8

0.1 0.1 0.2 0.2

0.3 0.3

31.5

0.1 0.1 0.1 0.2 0.3 0.4 0.5

54.7

0.1 0.1

129.3

0.1 0.1 0.2 0.2 0.3 0.4 0.6 0.9

35.6

0.6 0.7 1.0 1.1 1.3 1.5 1.8 2.3

18.8

-4.1

0.1

0.2 0.7 0.6 0.4 0.1

11.9

Total/Average

Percent Change (%)

Source Dataquest (August 1992)

1.0 1.3 2.1 2.8 3.8 4.5 4.5 4.8

364 27.2 62.6 30.3 39-2 18.0 -1.4 6.9

17.7

©1992 Dataquest IiKoipoiated August—Reproduction Prohibited

512b

Worldwide MOS Memory Forecast

0.82 0.75 0.70 0.60 0.55 0.50

39

Table 6-3

A v e r s e Selling Price for Shipments o f EEPROMs to the World, 1989-1996

CU.S. Dollars)

256b

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

0.80 0.75 0.72 0.65 0.60 0.56 0.45 0.39

-11.4

-9A

-14.1

IK

2K

1.10 1.00 0.62 0.49 0.45 0.39 0.34 0.29

2.50 2.20 1.33 1.05 0.80 0.65 0.50 0.45 -19.5

4K

8K

4.00 2.75 2.11 1.95 1.50 1.15 0.90 0.75

2.51 2.25 2.00 1.75 1.50 1.00

-18.7

-16.8

4.40 3.73 2.82 2.25 2.05 1.89 1.75 1.50 -11.8

1 6 K

64K

2 5 6 K

8.19 5.75 5.01 4.75 4.50 3.90 3.30 2.90

75.29 38.19 19.46 15.00 13.50 11.80 9-25 8.50

-10.4

-15.3

512K

1Mb

44.88

165.00 100.00 92.56 79.00 70.00 68.00 65.00 60.00

•S3

Total/Average

Percent Change (%)

Souice: Dataquest (August 1992)

2.70 2.30 1.53 1.33 1.28 1.11 0.91 0.{

-0.7 -14.9 -33.4 -13.4 -3.9 -13.3 -17.3 -12.1

-12.1

©1992 Dataquest Incorporated August-^teptoduction Prohibited

40

Memories Worldwide

Table 6-4

Shipments of EEFROMs to the World, 1989-1996

CtriUions of Bits)

256b

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

-9.6

27.6 512b

IK

2K

4K

8K

16K

64K

256K

512K

0.1

Q.l

0.2 0.2

0.1 0.1

2.8

0.1 0.1 0.2 0.2

0.3 0.3

31.5

0.1 0.1 0.1 0.2 0.3 0.4 0.5

54.7

0.1 0.1

129.3

0.1 0.1 0.2 0.2 0.3 0.4 0.6 0.9

35.6

0.6 0.7

1.0 1.1

1.3 1.5 1.8 2.3

18.8

0.2 0.3 0.6 0.8 1.0 1.3

0.8 0.5

-4.1

1Mb

0.1 0.2

0.7

0.6 0.4 0.1

11.9

TotaVAverage

Percent Change (%)

Source: Dataquest (August 1992)

1.0

1.3 2.1

2.8

3.8 4.5 4.5 4.8

364 27.2 62.6 30.3 39.2 18.0 -1.4 6.9

17.7

©1992 Data<piest Incorpoiated August—Reproduction Prohibited

Woridwide MOS Memory Forecast 41

Table 6-5

Price per Bit for Shipments of EEFROMs to the World, 1989-1996

(Micro Dollars)

CAGR (%)

" 1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

256b 3,125.0 2,929.7 2,796.3 2,539.1 2,343.8 2,187.5 1,757.8 1,523.4 -11.4

512b - - 1,598.3 1,464.8 1,367.2 1,171.9 1,074.2 976.6 -9.4

IK 1,074.2 976.6 605.5 478.5 439.5 380.9 332.0 283.2 -14.1

2K 1,220.7 1,074.2 650.6 512.7 390.6 317.4 244.1 219.7 -19.5

4K 976.6 671.4 514.8 476.1 366.2 280.8 219.7 183.1 -18.7

8K - - 306.4 274.7 244.1 213.6 183.1 122.1 -l6.8

16K 268.3 227.7 172.0 137.3 125.1 115.4 106.8 91.6 -11.8

64K 125.0 87.8 76.5 72.5 68.7 59-5 50.4 44.3 -10.4

256K 287.2 145.7 74.2 57.2 51.5 45.0 35.3 32.4 -15.3

512K - - 85.6 . . - - .

1Mb 157.4 95.4 88.3 75.3 66.8 64.8 62.0 57.2 -8.3

Total/Average 311.6 224.0 153.8 135.3 116.0 102.9 96.2 87.9 -10.6

Percent Change (%) -14.3 -28.1 -31.3 -12.1 -14.2 -11.3 -6.6 -8.6

Source: Dataquest (August 1992)

©1992 Dataquest Incxsrporated August—Beproduction ProbibiKd

42 Memories Worldwide

Table 7-1

Factory Revenue fix>m Shipments of Flash Memory to the World, 1989-1996

(MiUions of U.S. Dollars)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

2 5 6 K 5.2 10.3 17.0 39.0 67.7 90.0 74.4 43.4 20.6

512K 1.1 1.7 17.5 36.6 73.5 104.0 108.0 73.6 33.2

1Mb 4.9 22.5 51.5 99.0 156.8 222.8 292.6 ' 322.0 44.3

2Mb 0 0.8 33.6 63.9 97.5 171.6 243.2 334.8 58.4

4Mb 0 0 0 16.1 39.8 100.8 170.0 235.8

8Mb 0 0 0 19.2 105.8 381.6 585.0 619.2

16Mb 0 0 0 0 16.5 129.2 230.6 304.5

32Mb 0 0 0 0 0 0 0 3.6

64Mb .

0 0 0 0 . 0 0 13.0 52.8

Total/Average 11.1 35.3 119-6 273.8 557.5 1,199.9 1,716.8 1,989-7 75.5

Percent Change (%) 256.4 218.4 238.9 128.9 103.6 115.2 43-1 15-9

Source: Dauquest (August 1992)

©1992 Dataquest Incorporated August—Reproduction Prohibited

I

4Mb

8Mb

16Mb

32Mb

64Mb

512K

Worldwide MOS Memorjr Forecast

Table 7-2

Shipments of Flash Memory to the World, 1989-1996

(Millions of Units)

256K

CAGR C%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

0.5 1.4 3.0 7.8 16.5 25.7 24.0 15.5 38.7

0.1 2.3 6.0 15.0 26.0 30.0 23.0

43

58.3

1Mb

2Mb

0.1 1.2 4.9 13.2 28.0 49.5 77.0 92.0

1.6 4.5 10.0 22.0 38.0 62.0

79.7

108.8

0.9 3.0 12.0 25.0 41.0

0.6 4.5 24.0 60.0 86.0

0.3 3.8 14.5 29.0

0.1

0.2 1.2

Total/Average

Percent Change C%)

Source: DaUquest (August 1992)

0.6 2.7 11.8 33.0 77.3 163.0 268.7 349.8

338.1 314.3 342.4 179.6 134.2 110.9 64.8 30.2

97.0

©1992 Dataquest Incoipotated August—^Reproduction Prohibited

44 Memories Woridwjde

Table 7-3

A v e n g e Selling Price for Shipments of Flash Memory to the World, 1989-1996

(U.S. Dollars)

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

256K 10.86 7.58 5.62 5.00 4.10 3.50 3.10 2.80 -13.0

512K 23.89 13.10 7.59 6.10 4.90 4.00 3.60 3.20 -15.9

1Mb 39.13 19.54 10.49 7.50 5.60 4.50 3.80 3.50 -19.7

2Mb - 31.63 21.50 14.20 9.75 7.80 6.40 5.40 -24.1

4Mb - - - 17.90 13.25 8.40 6.80 5.75

8Mb - - - 32.00 23.50 15.90 9.75 7.20

.

16Mb

32Mb .

.

.

.

_

. 55.00 34.00 15.90 10.50

.

.

. 36.00

64Mb .

.

.

.

_ 65.00 44.00

Total/Avetage 17.21 13.23 10.13 8.30 7.21 7.36 6.39 5.69 -10.9

Percent Change (%> -18.6 -23.1 -23.4 -18.1 -13.1 2.1 -13.2 -11.0

Source: Dataquest (August 1992)

©1992 Dataquest Incorporated August—Kepioduction Piohibiled

I

'Woridwlde MOS Memory Forecast 45

Table 7-4

Shipments of Flash Memory to the World, 1989-1996

(TrilUons of Bits)

C A ^ l (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

256K 0.1 0.4 0.8 2.0 4.3 6.7 6.3 4.1 38.7

512K 0 0.1 1.2 3.1 7.9 13.6 15.7 12.1 58.3

1Mb 0.1 1.2 5.1 13.8 29.4 51.9 80.7 96.5 79.7

2Mb 0 0.1 3.3 9.4 21.0 46.1 79.7 130.0 108.8

4Mb 0 0 0 3.8 12.6 50.3 104.9 172.0

8Mb 0 0 0 5.0 37.7 201.3 503.3 721.4

16Mb 0 0 0 0 5.0 63.8 243.3 486.5

32Mb 0 0 0 0 0 0 0 3.4

64Mb 0 0 0 0 0 0 13.4 80.5

Total/Average 0.3 1.7 10.4 37.3 117.9 433.8 1,047.3 1,706.4 177.2

Percent Change (%) 640.7 505.0 519.5 257.6 216.2 268.0 141.4 62.9

Source Dataquest (August 1992)

®1992 Dataquest iDCOipoiated August—^Reproduction Prohibited

46

Memories Worldwide

Table 7-5

Price per Bit for Shipments of Flash Memory to the World, 1989-1996

(Micro Dollars)

256K

CAGR (%)

1989 1990 1991 1992 1993 1994 1995 1996 1991-1996

41.4 28.9 21.5 19.1 15.6 13.4 11.8 10.7 -13.0

512K

1Mb

45.6 25.0 14.5 11.6 9.3 7.6 6.9 6.1

37.3 18.6 10.0 7.2 5.3 4.3 3.6 3.3

-15.9

-19.7

2Mb

15.1 10.3 6.8 4.6 3.7 3.1 2.6

-24.1

4Mb

8Mb

16Mb

4.3 3.2 2.0 1.6 1.4

3.8 2.8 1.9 1.2 0.9

3.3 2.0 0.9 0.6

32Mb

1.1

64Mb

1.0 0.7

Total/Average

Percent Change (%)

Source: Dataquest (August 1992)

39.9 21.0 11.5 7.3 4.7 2.8 1.6 1.2 -36.7

-51.9 -47.4 -45-3 -36.0 -35.6 -41.5 -40.7 -28.9

©1992 Dataquest Incorpoiated August—Reproduction Prohibited

DataQuest

Dataquest Research and Saks Offices:

Dataquest Incorporated

1290 Ridder Park Drive

San Jose, California 95131-2398

United States

Phone: 01-408-437-8000

Facsimile: 01-408-437-0292

Dataquest Incorporated

Dataquest/Ledgeway

550 Cochituate Road

Framingham, Massachusetts 01701-9324

United States

Phone: 01-508-370-5555

Facsimile: 01-508-370-6262

Dataquest Incorporated Invitational

Computer Conferences Division

3151 Airway Avenue, C-2

Costa Mesa, California 92626

United States

Phone: 01-714-957-0171

Facsimile: 01-714-957-0903

Dataquest Australia

Suite 1, Century Plaza

80 Berry Street

North Sydney, NSW 2060

Austraha

Phone: 61-2-959-4544

Facsimile: 61-2-929-0635

Dataquest Europe Limited

Roussel House, Broadwater Park

Denham, Uxbridge

Middlesex UB9 5HP

England

Phone: 44-895-835050

Facsimile: 44-895-835260/1

Dataquest Europe SA

TourGalli6ni2

36, avenue du General-de-Gaulle

93175 Bagnolet Cedex

France

Phone: 33-1-48-97-3100

Facsimile: 33-1-48-97-3400

Dataquest GmbH

Kronstadter Strasse 9

8000 Munich 80

Germany

Phone: 49-89-930-9090

Facsimile: 49-89-930-3277

Dataquest Germaqy

In der Schneithohl 17

6242 Kronberg 2

Germany

Phone: 49-6173/61685

Facsimile: 49-6173/67901

Dataquest Hong Kong

Rm. 4A01

HKPC Building

78 Tat Chee Avenue

Kowloon, Hong Kong

Phone: 852-788-5432

Facsimile: 852-788-5433

Dataquest Japan Limited

Shinkawa Sanko Building

1-3-17 Shinkawa, Chuo-ku

Tokyo, 104

Japan

Phone: 81-3-5566-0411

Facsimile: 81-3-5566-0425

Dataquest Korea

Daeheung Building 1105

648-23 Yeoksam-dong

Kangnam-gu

Seoul 135-080, Korea

Phone: 82-2-556-4166

Facsimile: 82-2-552-2661

Dataquest Singapore

4012 Ang Mo Kio Industrial Park 1

Ave. 10, #03-10 to #03-12

Singapore 2056

Phone: 65-4597181

Telex: 38257

Facsimile: 65-4563129

Dataquest Taiwan

Room 801/8th Floor

Ever Spring Building

147, Sec. 2, Chien Kuo N. Rd.

Taipei, Taiwan R.O.C. 104

Phone: 886-2-501-7960

886-2-501-5592

Facsimile: 886-2-505-4265

Dataquest Thailand

300/31 Rachdapisek Road

Bangkok 10310

Thailand

Phone: 66-2-275-1904/5

66-2-277-8850

Facsimile: 66-2-275-7005

Librspyf Corr-orste

SIS-1303236 B! 1

.197-1234

Internal Distribution

13379

I

^s

Hl^^H

Dataquest Vendor Profile

Memories Worldwide

September 14,1992

Micron Technology

BA0iueB)^°9

Corporate Statistics

Headquarters and Facilities Location

Chairman and CEO

President and COO

Hscal Year Ends

Employees

FY1991 Revenue

FY1991 Net Profit after Taxes

Shareholders' Equity

Shares Outstanding

Products

Leadership Area

Boise, Idaho

Joseph L. Parkinson

Steven R Appleton

August 31

4,095 at year-end 1991

$425 million

$5.1 million

$495 million

37.8 million

82 percent DRAMs,

18 percent SRAMs

Largest U.S. domestic producer of DRAMs

For more information on Micron Technology or the memories industry, call Lane Mason at

(408) 437-8120

Micron Technology was founded in 1978 as a design house, but by

1982 it had emerged as a bona fide DRAM producer. Over the past decade, it has faced—and faced down—^iimumerable challenges as it progressed from the 64K to 4Mb DRAM density, and from no annual revenue to a $500 million annual run rate for a wide family of

DRAMs and SRAMs. Micron has been the smallest continuous player in the DRAM business, with total revenue 10 to 100 times smaller than that of its competitors. It has outlasted a host of companies far better financed, including Intel, Mostek, and National Semiconductor. It has had to think smart to siuvive, whether it was in its innovative capitalconserving and cost-reduction methods, its use of the ETC to bring to task the GoUaths of the East that were dumping product in the world's DRAM markets, or in its enticing investment from a major user to accelerate facility expansion.

But as the smallest DRAM-focused DRAM suppUer, Micron now faces the biggest challenges of its decade-long existence. And, being the smallest, perhaps it also shows us a glimpse of what all others will encounter in their tmn.

Table 1 shows several time series for Micron Technology's fincmcial performance since its inception. The table shows the financial roller

DataQuest

n n acompanyof

MmMM ThcDun&BradsttcctcCKporation

MMRY-SEG-VP-9201

This profile is the property of Dataquest Incnrparated. Reproduction or disdoaue in whole or in part to OAICT parties shall be made upon Ihe written and express consent of EbtaqnesI: Itus leport shall be treated at all times as a confidential and jropafiaTy dooiznent for internal use only. The infoimation contaifiEd in this publication is behered to be reliable hut cannot be guaranteed to be comet or compiete.

©1992 Dataquest Incorporated—Repioduchon PriAibited

Dataquest is a rtgistaBd ttademark of A.C Nidsen Coit^>any 0013694

3

E

@

I

Table 1

Micron By the Numbers (Millions of Dollars)

Fiist

Quarter

Ending 12/5

Second

Quarter

Ending 2/28

Third

Quarter

Ending 5/31

Fourth

Quarter

Ending 8/31

FY Ending

9/79-8/82

Revenue

Profit (%)

8/31/83

Revenue

Profit (%)

8/31/84

Revenue

Profit (%)

8/31/85

Revenue

Profit (%)

8/31/86

Revenue

Profit (%)

8/31/87

Revenue

Profit (%)

8/31/88

Revenue

Profit (%)

0.6

-2.01

8.3

2.05

37.2

10.45

5.0

-11.60

18.8

-9.68

43.2

8.43

2.2

-1.39

12.3

2.91

18.2

2.82

9.4

-9.78

20.1

-10.95

58.3

16.94

4.3

0.34

29.4

13.1

14.4

-5.75

14.4

-6.71

22.8

-3.72

85.6

29.28

6.0

0.43

37.4

10.91

6.1

-7.37

20.0

-5.84

29.5

1.42

113.4

43.33

91.2

-22.93

300.5

97.98

Year

4.8

-7.3

13.1

-2.63

87.4

28.97

75.9

0.15

48.8

-33.93

Net

FEE R&D

NA 2.9

18.4

77.0

104.5

97.7

84.3

117.4

0.2

2.7

6.6

2.9

5.3

1

^AdC^

(T

^.,

8

! ! ^

®

Table 1 (Continued)

Micron By the Numbers (Millions of Dollars)

FY Ending

8/31/89

Revenue

Profit {%)

8/31/90

Revenue

Profit (%)

8/31/91

Revenue

Profit{%)

8/31/92

Revenue

Profit (%)

NA - Not available

•Dataquest estimate

Source: Micron Tectinology

First

Quarter

Ending 12/5

110.4

32.18

66.5

0.04

80.3

-9.27

111.8

0.63

Second

Quarter

Ending 2/28

Third

Quarter

Ending 5/31

Fourth

Quarter

Ending 8/31

113.8

29.18

77.5

0.01

94.5

-2.24

' 128.2

1.46

119.2

28.78

84.1

1.81

126.8

7.02

131.1

1.66

103.0

15.96

105.3

3.04

123.8

9.57

135.0*

2.2*

Year

446.4

106.1

333.4

4.90

425.4

5.08

Net

PPE R&D

326.0

385.1

389.3

21.4

35.6

35.8

506.1*

5.95*

390.0* 36.0*

¥

3

S"

Memories Worldwide coaster the company has been on. But Micron rose from the depths of

1985-1986 to consistently run more than $130 million in revenue per quarter recently. The reversals of 1990-1991 were not nearly so severe as those of the earlier cycle, and Micron scraped b y with just two quarters of red ink and revenue that dropped only 45 percent from peak prior levels.

In addition, the table tracks technology portfolio measures: R&D spending, acquisitions of product and process technology (PPT), and royalty payments. For the physical plant, both capital spending and net PPE at year-end are also included.

Micron's Forte: Doing a Lot with a Little

Micron has been able to survive, and prosper from time to time, because it has some excellent design skills and an uncanny knack for making the smallest die that requires the fewest mask steps to produce. The former collection of skills make for more gross and net die per wafer, while the latter make for reduced capital requirements for a given level of unit production.

In 1983 and 1984, the early days of Micron's participation in the

DRAM market, the company often was considered something of a joke. It was thought to be too small to be a serious supplier, and it peddled a 64K DRAM plagued with soft errors. Its answer—a. firstpass 256K DRAM with error correction—^was a good idea, but it was probably four DRAM generations before its time.

Even in 1988-1989, when the industry was far short of meeting demand. Micron squeezed millions out of the 256K DRAM market, even as its critics complained, among other things, that its parts cut too many comers and it could not meet demanding systems requirements.

Micron's critics did not fuUy appreciate the changing nature of the

DRAM market during that time frame: The IBMs, HPs, and Digitals, with their big-system specifications and 5- to 10-year MTBFs, were rapidly being replaced by a cost-driven, low-end systems market.

There was another DRAM market, cost-sensitive and without lengthy and exacting qualifications, where Micron found ample room to play

(and where the Koreans were to follow a few years later).

At the same time, despite occasional early rejections from certain key accoxmts. Micron gradually made its way back as its quality, reliability, and device performance improved. Eventually, its account base looked pretty much like that of everyone else, including IBM, Digital, Compaq, Apple, and Acer.

After suffering mightily in DRAMs in 1985-1986, Micron branched into

SRAMs, which now make u p almost 20 percent of its business. It has differentiated the DRAM and SRAM product lines, as well, with dualport DRAMs, three-port DRAMs, Quad-CAS DRAMs, byte-wide 4Mb

September 14,1992 ©1992 Dataquest Incorporated MMRY-SEG-\/P-9201

Micron Technology

and word-wide 1Mb and 4Mb DRAMs, and ^^deoRAMs. On the

SRAM side, it now has 16K-lMb SRAMs, plus latched and S5mchronous 16Kxl6 and 16ICxl8 SRAMs and FIFOs.

StiU, these differentiated products sell into markets that are not large enough to protect Micron from the crushing pressures of a market that has been too far down for too long.

The Micron Way

Table 2 shows several examples of the skills that Micron has brought to cost-reduced DRAMs. The table shows die sizes and mask counts for successive generations of Micron's 256K, 1Mb, and 4Mb DRAMs.

At present. Micron is completing the transition to the "hypershrink"

1Mb DRAM and the "ministack" 4Mb DRAM. Compared with standard industry parts, the 1Mb is about 75 percent as large as the next smallest competitive part, and the 4Mb (present generation) is about the same as other manufacturers, with two more revs to come.

Indeed, faced with escalating costs of new facilities, and process that only becomes more complex, many competitors that earlier turned up

Table 2

Micron Technology DRAM Mask and

Die Size Progression

Density

256K

1Mb

Device Name

Production 1

Production 2

Shrink 1

Shrink 2

Production 1

Production 2

Shrink

Supershrink

H j ^ r s h r i n k

Die Size

40.00 sq. mm.

32.49 sq. mm.

23.39 sq. mm.

19.10 sq. mm.

56.8 sq. nun.

44.74 sq. mm.

36.67 sq. mm.

24.26 sq. mm.

17.64 sq. mm.

Introduction

Date

7/84

12/85

4 / 8 7

5/91

10/87

10/89

10/90

3/91

12/91

4Mb Production 1

Production 2

Ministack

Shrink

Supershrink

100.10 sq. mm.

75.87 sq. mm.

61.29 sq. mm.

48.32 sq. mm.

46.45 sq. mm.

3/90

6/91

3/92

TBD

TBD

16Mb

Prototype 140.39 sq. mm.

TBD = To be determined

Source: Micron Technology, Dataquest estimates (September 1992)

2/92

Masks

9

9

7

7

13

11

10

10

10

13

12

12

11

10

16

MMRY-SEG-VP-9201

©1992 Dataquest Incorpoiated September 14,1992

Memories Worldwide

their noses at Micron's capabilities are taking another look. Micron has the following collection of proven methods:

• Extending the useful life of its equipment

• Reducing mask counts and thereby increasing throughput without compromising performance

• Making the smallest die in the industry

Technology Laggard

Though Micron has working samples of 16Mb DRAMs and several advanced development programs u p into the 64Mb stratosphere, traditionally it has been a technology laggard. It t57pically was late to market with generation after generation (preferring to make millions on last year's part). It has only sparingly invested in the distant future, choosing instead to concentrate on refining the present money generation by cost reduction. This habit, too, is gaining Micron some attention in the industry, as companies see much of their far-advance investment coming to naught.

At the same time. Micron has been an innovator—not always mindful of the market, but still with a nice portfolio of innovative DRAM and

SRAM designs. Forget for the time being its 64K soft-error problems and its 256K and 1Mb forays into ECC, and pay attention instead to the fact that, like Samsung, it has risen from a virtual nonplayer to a substantial position in the DRAM market in a decade, but with immeasurably smaller resotirces at its disposal. Today it is the largest domestic producer of DRAMs (No. 8 worldwide) and the No. 7 supplier of fast SRAMs. Micron was early with a Quad-CAS 1Mb DRAM, won accolades for its three-port DRAM, and impressed all with its

SRAM successes.

Still, Micron's profits today are marginal, which reflects both on the tough market envirorunent and its own continuous die revision upgrades that have been played out over the past two years as it moved from producing one die revision for 1Mb and 4Mb DRAM to the next. Now that Micron is sold on its last 1Mb revision, it can focus on yield improvement. But its 4Mb has two more revisions to go.

The following analysis provides a pro forma rollout of the revenue r u n rate that Micron may be able to generate. Micron is running about

12,000 ISOmm wafers per week through its facility and is at near capacity, given its mix of SRAM and DRAM products. It is generating about $10 million per week in revenue, or about $800 per wafer.

With the reduced mask-count 4Mb DRAMs now being input into the line. Micron may be able to keep the number of wafer outs at about the same level while shifting the product output more to 4Mb

DRAMs. Micron's newest 4Mb device requires the same number of masks steps as its 1Mb product.

September 14,1992 ©1992 Dabquesl Inconjorated MMRY-SEG-VP-9201

Micron Technology

1Mb DRAM Potential

At its peak, in 1993-1994, Micron may b e able to yield 600 to

650 net die per wafer from its hypershrink 1Mb DRAM, and generate revenue of about $1,600 to $1,700 per wafer.

4Mb DRAM Potential

The present incoming ministack version is a die size of about

61 square millimeters, which has about 210 gross die per wafer; the outgoing "production die" has 170 gross die per wafer. The supershrink 4Mb die, which will be Micron's production vehicle in 1994, will have about 280 gross die per wafer. With a 75 percent line yield, upward of 200 net die per wafer may be jdelded, for revenue of $1,200 to $1,300 per wafer in 1994.

To achieve these potential improvements. Micron must stabilize production around a single die iteration and concentrate on yield improvement. It is not inconceivable that Micron could produce u p to $800 million from its existing facilities in the 1994 time frame, compared with its estimated $506 million for FY1992.

Still, lacking at upturn in demand and profitability. Micron will likely be unable to fund the next increment of expansion, which must be p u t in place in the next two years to be ready for

1995-1996.

The Twin Peaks: Capital and inteiiectuai Property

It has been known for some time that participation in the DRAM business requires immense amounts of capital. Of all the fables p u t forth over the past decade about the DRAM market—cannot sit out a generation, need of a captive user, bi-rule and pi-rule, increasing PPB from generation to generation— holding most true is that massive sums must be expended to participate.

Though the capital is ultimately recovered through depreciation, at some point in the cycle it must be made available by someone, and in large sums. Micron's earlier building programs in 1983-1985 and again in 1988-1990 combined the cydical profits of the DRAM industry with infusions of equity funding from outside in three secondary placements diulng 1986-1987, and in a $76 million investment from its largest European customer, Amstrad pic, in 1989, plus equity offerings for an additional $170 million.

This timie, the down cycle has persisted longer than might ordinarily have been expected, keeping recent profits low. Over the 12 quarters of P^1990 to FY1992, Micron's after-tax earnings have been about

$16 million on sales of $1,164 million—hardly enough to go the next roimd.

At the same time, the capital requirements to fund the next round of capacity expansion and substantially expand capacity for new 4Mb production and the early 16Mb DRAM market are immense even in comparison to the sums expended in 1989-1990. It takes about

$350 million to get 9,000 monthly starts using 200mm wafers on a

MMRY-SEG-VP-9201 ©1992 Dataquest Incorporated September 14,1992

Memories Woridwide

0.6-nm design today, which is enough to stay competitive through about 1996-1997. But without a profit bubble as seen in 1983-1984, or again in 1988-1989, there is small hope of either funding the expansion or enticing investors to part with their money.

Micron certainly needs capacity to grow; it is n o w running at near capacity and has its finest "small die" and "few masks" designs already into production. If the market is to grow better than 50 percent by 1994, as many expect. Micron must have the capacity in place or miss a great opportunity to gain market share.

Micron's board has been reluctant to issue more stock (indeed, the recent price, at $15.75, is 25 percent lower than what Amstrad paid in

1989). It is not known how well the market might receive such an offering, given the state of DRAM profits recently. Micron is said to be actively seeking a partnership that will help it with the capacity upside.

Intellectual Property

Like everyone in the semiconductor industry. Micron's consciousness of the importance of intellectual property has been raised dramatically since 1987, when Texas Instruments renewed its patent licensing agreements with its licensees. Indeed, Micron said the following in its 1987 and 1988 Form lOKs:

• In 1987: "The Company has received notice of infringement of patents from certain semiconductor manufacturers with respect to certain aspects of the Company's processes and devices. If any infringement has, in fact, occurred. Micron is of the opinion that any necessary licenses or other rights under patents could be obtained on conditions which would not have a materially adverse effect on the Company."

• In 1988: "While the Company intends to seek patent protection on as much of its technology as possible, due to the rapidly changing technology in the semiconductor industry. Micron believes that its future success will be dependent, in large measure, upon the technical expertise and creative skills of its personnel."

The license agreements it had in place at the close of FY1988 were with Shell Development Company, Motorola, Standard Microsystems,

ATT, and Intel. All were modest in their financial consequences as originally written. But the deal with Intel was to explode less than

15 months later, resulting in a "renegotiation" of the original agreement and a $50 miUion licensing settlement with Intel for DRAM,

SRAM, and VRAM technologies being used by Micron in its products.

Also, at that time. Micron had already received notice from Texas

Instruments that it was believed to be infringing TI's patents. Though

Micron resisted settlement on TI's terms, it had already set aside a reserve of $17.6 million in the event of an unfavorable resolution.

Eventually, in May 1989, Micron settled with TI for $38.2 million.

September 14,1992 ©1992 Dataquest Incorpofated MMRY-SEG-VP-9201

Micron Technology

So, while Micron maintained outwardly the position that technology was changing fast enough that, with its own creative powers, it could avoid major impacts from others' technology positions, like many others it had small imderstanding of what was to come. Intellectual property rights (IPR) emerged in the late 1980s as the most important source of competitive advantage, income, and profits.

This period m a d e for a realization of the immense value and costs of proper treatment of IPR at Micron. It doubled its efforts to achieve a patent portfolio for itself that it hoped would absorb the brunt of the impact from IPR heavies bearing d o w n on them, all seeking a king's ransom for their own intellectual property.

As of year-end FY1992 (August 31,1992), Micron had been granted about 192 U.S. patents (see Table 3).

One can imagine that many of the critical MOS IC, DRAM, and SRAM patent structures and circuits are already daimed by companies that were in the market before Micron came into existence. Indeed, Intel,

TI, and IBM have proven to be the big patent winners in the IPR wars over the past five years. So, despite the rapid rate of technical change that Micron hoped would save it from pain, the MOS pioneers, for the time being, are reaping vast sums in royalty and licensing fees from the new DRAM makers. Micron included.

Though Micron is rapidly building u p its own patent war chest, and the patents of the pioneers are slowly expiring. Micron is still d u e to pay to play for the next several years. For now. Micron has put in place a series of licensing agreements that give it access to the essential technology to participate in the SRAM, VRAM, and DRAM markets (see Table 4).

Micron Technology has paid out more than $270 million for acquisition of product and process technology and for annual royalty payments since FY1988. This compares with Micron's direct R&D expenses over the same period of about half that amount ($138 million) and its after-tax profits of $220 million.

Table 3

Micron Technology Patents at Year-End 1992

Year

1986

1987

1988

1989

1990

1991

1992

Source: Micron Technology and U.S. Patent Office

Patents

1

1

1

12

44

106

192

MMRY-SEG-VP-9201 ©1992 Dataquest Incorporated September 14,1992

3

Table 4

Micron's Technology Licenses

Company

Shell Development Company

ATT

Intel

Motorola

Texas Instruments

Standard Micro Systems

IBM

Wang Labs

Hitachi

Samsung

Sanyo

Item

Basic MOS patents

XLC (Memory plus sensors)

Tech XLC; DEIAM, SRAM, VRAM

XLC

XLC

MOS patents, XLC

4Mb DRAM technology

Tech XLC, joint technology development

SIMMs license

Tech XLC

EEPROM, SRAM rights, XLC

Micron Lie 64Kxl6

Source: Mfcron Technology Form 10K, Dataquest (September 1992)

Date

1985

10/86, 1/89

3 / 8 8 , 1 / 9 0

1988

5/89, 9/92?

3/88

11/89

Tenns

Paid in full

Fee plus ongoi

$50 million plu

Ongoing

$38.2 million p

$9,3 million sto

$50 million

12/91

7/89

6/86

10/89

Per-unit fee

Fee plus per-u

Samsung buys

Royalty to uT

CO

S3

I

Micron Technology 11

Micron capitalizes its purchases of product and process technology and amortizes those costs over the patent term, the useful life of the technology, or the term of the agreement, whichever is shortest. Royalties paid and amortization of capital costs are ascribed to production and R&D costs. Costs incurred to establish patents are also capitalized.

Micron's 1989 agreement with Texas Instruments is set to expire soon, and Micron warns in its recent third-quarter interim report that "... the

Company's cross-license agreement with Texas Instruments, Inc. expires September 3, 1992. There can be no assurance that the crosslicense agreement can be renewed on acceptable terms." If this year's royalty income for Texas Instruments is any indication of an increase in the aggressiveness with which it pursues favorable cross-license agreements. Micron may not get off as well as it did in 1989. Tl's royalty income is u p 45 percent to $218 million in the first half of

1992, compared with $150 million in the first half of 1991.

Alliances

In addition to licensing technology. Micron has engaged in m a n y alliances over the years to acquire and develop technologies deemed necessary to carry out its business. It is a founding member of Sematech, though it has announced plans to withdraw at year-end.

In 1989, Micron signed an agreement with Sanyo for Sanyo to buy

Micron's DRAMs both for its own use and to resell into the Japanese market. About a year later, this agreement was expanded to provide for Sanyo to actually produce for use and sell Micron's 64Kxl6 DRAM into Japan.

More recently, on July 15, 1992, Micron and NEC announced a joint cross-OEM arrangement to seU each other's SRAM and DRAM products under their o w n brand nam.es. It was offered as a rationale that this would reduce the product development cycle and cost.

However, there may be more to this agreement than first was made public.

Micron's Options: Capacity Expansion

Given its reluctance to float more stock (at least for now). Micron is said to be actively seeking a partner that could help it gain access to additional wafer fab capacity. The options are rather limited, but, as in the past. Micron is certain to strike a creative deal that wiU serve the interests of all parties.

Customer-Funded Fab

The first option is a Texas Instruments-like partner-funded front end. Since 1988, TI has gained essentially a n entire new ftiont end, and control over many times that amoimt, through the use of creative ventures with its customers and others interested in getting into the semiconductor business. Both Dallas DMOS 4.2 and Avezzano, Italy were built, in part, using advance payments from key Tl customers (plus subsidies from the Italian government). KTI, a joint

MMRY-SEG-VP-9201 ©1992 Dataquest Incorporated September 14,1992

12 Memories Woridwide venture with Kobe Steel, was essentially funded b y Kobe Steel, with TI providing the technical wherewithal. Its joint venture with

Acer, now coming u p with the 4Mb DRAM, was more than half funded by Acer, and another 24 piercent of the stock is held by a variety of Taiwan interests. TECH Semiconductor is a joint venture among HP, Canon, and the Singapore government.

Such a venture is appealing to Micron because it would limit

Micron's monetary contribution, as well as its equity share. But, for the most part, it controls the output.

Micron may have an opportimity to take advantage of the same concern that drove Acer to invest in TI: dependence of the

Taiwanese computer businesses on imported DRAMs, largely from other Asian companies. Also, because Micron is (so far) a paid-up licensee of TI (and others') patents, this could be another advantage for Taiwan interests choosing to partner with Micron instead of another DRAM maker.

Foundry

Another option is to use foundries to make its product. For Micron,this could be difficult because the manufacturing-intensity of the product means that Micron itself needs to have tight control over the process and facility to maintain its technical advantages. This would almost be impossible in a facility shared with strangers, because the processes must be compatible.

Shared Facility

A third option is a joint venture with another semiconductor manufacturer operating in a shared facility. This could be a partner that wanted to gain from some of Micron's low-cost manufacturing techniques, or to reduce its own exposure to royalty payments by benefiting from Micron's patent portfolio, which is more filled out.

An example of such a venture is the Altera-Cypress facility in

Round Rock, Texas.

Lease a Fab

A fourth option is to lease an underutilized facility from another party and run it on a contract basis. Such an arrangement would allow Micron to make lease payments out of current revenue and avoid ownership and equity dilution, but gain access to additional capacity. An example is Alliance Semiconductor's aborted leasing of the ATT facility in Lee Summit, Missouri in mid-1989.

Lease options are attractive and a trend in the semiconductor industry. All companies must decide where to apply their capital to greatest effect. As in the airline business, a group of capital/capital equipment providers may spring u p to support the technology providers in the semiconductor industry. As also was the case in the airline industry, banks may be reluctant to extend loans to companies in such an expensive, competitive business as DRAM manufacturing.

September 14,1992 ©1992 Oataquest Incoiporated MMRY-SEG-VP-9201

Micron Technology 13

Bargain-Basement Facility

Cypress picked u p a quality facility from CDC last year, at a very attractive price. One rumor floating around concerning Micron has it tied up with IBM's Manassas, "V^ginia facility.

The prospect of Micron having access to NEC's Roseville, California facility is attractive, as well, and its existing deal may grow to include such a tie-up.

Micron's Options: Intellectual Property Rights

Few^ companies can pay out 10 percent of sales for royalty paymentSf plus another 7 percent for their own R&D, and still p u t money in the bank at the end of the day. As Micron feverishly expands its patent portfolio to gain leverage in its negotiations with oilier patent traders, there are recent court rulings that might help it avoid such excessive pa5rments.

Last month, the initial Cyrix ruling held that, since SGS-Thomson was fully cross-licensed with Intel (via Intel's earlier agreement with

Mostek, which STM bought in 1985), STM could foundry the 387 for

Cjrrix without violating Intel's patent rights. The ruling is being appealed by Intel. Texas Instruments may have similar designs with its own agreement with Cyrix over Cjnix's 486.

In a 1991 ruling, the courts also prohibited SMSC from transferring its full cross-license rights with Texas Instruments to third parties. The limits of this ruling could also have an impact on Micron's ability to reduce its royalty payments.

Significant cross-licensing umbrellas may be available to Micron to reduce its liability to Texas Instruments in particular, or any of its licensers, such as Intel, IBM, or others.

Dataquest Perspective

What might w e expect? Over the next 6 to 12 months, we can expect

Micron to push ahead in its traditional cost-reduction program. We expect to see significant results as the pace of die revision transition slows. At the same time, we can also expect Micron to use its increasingly valuable technical assets to establish a partnership with another party that will supply expanded capacity for Micron at a far reduced cost to Micron than it would face were it to do it by itself.

Finally, the aversion of Micron's management to additional funding may be temporary. Securities analysts—and probably Micron management—^look for improved earnings from Micron as the business improves over the next 6 to 12 months as part of a cyclical upturn. Improved earnings mean improved stock price, so the longer

Micron can wait before any equity offering, the greater the yield. But so excessive was capacity as 1992 began that a year with 75 percent

DRAM bit growth, significant by recent standards, so far has failed t o absorb all excess and arrest DRAM price declines.

MMRY-SEG-VP-9201 ©1992 Dataquest Ixoiporated September 14,1992

14 Memories Worldwide

But this cannot go on forever. Japanese companies have announced cutbacks in capital spending of about 30 percent for the present year, the Micron-initiated antidumping petition and similar EC nilings are less than a month away, and the summer quarter has been mild by comparison with earlier years. Micron could be positioned quite well for 1993 and beyond if it can strike the right deal with the right partner and reduce both its capital bxirden and its royalty burden at the same time.

September 14,1992 ©1992 Dataquest Incorporated MMRY-SEG-\/P-9201

Dataquest Perspective

Memories Worldwide

MMRY-SE6-DP-9204

Do Not Remove

December 14,1992

In This Issue...

Market Analysis

Market Analysis

DRAM Maricet Trying to Keep Afloat in

DRAM Market Trying to Keep Afloat in Choppy Waters

Clioppy Waters

With the handing down of the Preliminary Dumping margins in Micron Technology's antidumping

As a result of last month's ruling concerning

Korean DRAM makers dmnping their product in suit against Korean DRAM makers on 21 Octob^, the U.S market, coupled with a similar ruling in the DRAM market entered choppy waters not the European Community (EC) a month earlier, likely to subside until several important issues are the DRAM market is currently in a minor state clarified further. U.S. DRAM users should watch developments closely, as their interest has yet to be expressed in the proceedings.

By Lane Mason Page 1

of turmoil. Although the U.S. Department of

Commerce (DOC) wHl not make its final ruling imtQ March, at which time it will set forth the final dumping margins, it is already clear that

Forward Alliances: Look for Improved DRAM

User-Vendor Relations

the prelinttnaiy rulings are having an impact, uitrodudng a significant element of uncertainty into the market.

Forward alliances, w^hich are alliances between semiconductor suppliers and their customers, can be an attractive alternative to using the market to gtiide capital investment, production and procurement strategies in the high-fixed-cost and highly volatile DRAM market.

By Lane Mason

Strong demand had already absorbed much of the excess capacity existing at the begiiming of

1992; therefore, the recent ruling came on top of what was a natural tightening of the market

Page 7

with some price stabihty. The ruling introduced a high noise element into the changing market, making it more difficult to discern what is happening in the larger supply-demand balance.

Still, the maimer in which this particular episode has been played out leaves much to be desired.

In the midst of many conflicting impressions, advice, and analysis, Dataquest offers the following commentary on the events of the past month.

The Effort to Find the Truth

Micron filed ite petition on April 22, 1992;, but the preliminary dumping margins were not made public imtil October 21—a full half year later. The EXDC reported that Samsung's preUminaiy dumping margin was 87 percent. This was based on "best information available" rather than on the data Samsung submitted, which was at least in part rejected by the DOC.

Samsvmg has shipped more than $1 billion in

DRAMs in 1992, and few really beUeve that

Dataoyest"

n m acompanyof

MmMM TheDunsTBradamtCorpontion

Dataquest is a registered trademark of A.C. Nieisen Company..

File behind the Perspectives tab inside the binder labeled

Memories Worldwide

©1992 Dataquest Incoiporaied, Reproduction PiDhibited 0014199

Memories Woridwide

SasfiSimg lost ti^e $800,n^o4i5h,|)RAMs as suggested by E>OC's preliminary dumping penalty ruling. The niling was a surprise to both

Samsimg and GoldStar. Why the DOC couldn't have worked more closely with all Korean companies to iiisure that the amounts listed in the preliminary ruling were more accurate is hard to understand. While DOC accountants have just returned from Korea after examining the books of the three defendants, there is no reason they couldn't have been there during the simuner, to ensure that the preliminary ruling was close to accurate. of the recoristructed costs. The Koreans should have known at all times what their costs and sales prices were (on the DOC cost basis) in every region. They should have been able to produce DOC-acceptable cost and price data on a moment's notice, thereby avoiding the uproar that has resulted.

Furthermore, it is clear that GoldStar and

Samsung were not altogether cooperative with the DOC's preliminary investigation, failing to produce adequate cost documentation even as the six-month deadline was about to expire.

The other logic behind antidumping laws is that companies dump product to drive the competition from the market, after which they raise prices and reap immense profits.

Third-Country Prices

While it is written in U.S. law that the absence of adequate data in the home market is siifficient reason to use comparable third-country data (as was the case of using Hyundai's

Singapore sales data), it certainly puts Samstmg and GoldStar at a disadvantage because each is reqmred to use its own Korean sales data for price comparison. Singapore is the essence of a highly competitive (low price) market, while Korea has the significant element of controUed-access.

By failing to do so, and by reporting as high a number as it did, the DOC has caused imnecessary turmoil in the market at a high cost to users, and it has imposed imnecessary costs on

DRAM makers as well.

But the DOC can't take all the blame. Those who are party to the complaint are not entirely blameless.

Afterthought: Where Was the Korean

Forethought?

Micron first rattled its saber concerning antidumping charges in late 1990, and its threat has been intense since early 1992 when GoldStar and Hyundai began to make great efforts to increase their U.S. market share. Micron has stated, in a press release to its DRAM customers, that it also met with representatives of all three

Korean DRAM makers, as well as with U.S. officials in Washington, D.C., three times in the prior two years in an attempt to halt the alleged dumping—obviously without the desired result.

It is hard to imderstand why the Koreaits didn't substantiate their costs relative to the reconstructed cost formulations. Surely the Koreans knew of the Japanese experience, the resulting fair market value agreements, and the intricacies

Logic of Antidumping Legislation

On a global scale, the Korean market offers a vanishingly small opportunity when compared with the rest of Asia, the United States, and

Europe. Dataquest estimates that 1992 Korean consumption was about 3 percent of the world

DRAM market. A logical argument in favor of antidumping laws is that companies subsidize their dimiped export sales with high domestic prices. Given the external sales ratios of the

Koreans (about 90 percent), 4Mb DRAMs would have to cost $100 in Korea, without loss of sales volumes, to make up for the lost profits resulting from londerpriced export sales.

The other logic behind antidtmiping laws is that companies dump product to drive the competition from the market, after which they raise prices and reap immense profits. Certainly, the profits reaped by Japanese DRAM makers in

1988 and 1989, after U.S. competition exited the market, were immense. But Samsung was probably the single most important DRAM supplier to help restore equilibriimi in the market beginning in 1989. And it was Hyundai and GoldStar that were the acknowledged "price aggressors" from

December 14,1992 ©1992 Dataquest Incorporated MMRy-SEG-DP-9204

Memories Worldwicle the fourth quarter of 1991 through the third quarter of 1992, responsible for dropping 4Mb prices from $16.00 to $10.00 and gaining market share on a monthly basis.

A third, and important, reason for dumping laws is to he able to protect domestic industry from unfair trade practices,

Koreans pose far more of a threat to Japanese consimier electronic equipment makers than to any European or U.S. electronic equipment

makisx. Indeed, if it weren't for Siemens in

Europe and Micron Technology in the United

States, U.S. and European policymakers might be well-advised to allow dumping so that the lowest possible DRAM prices exist for their systems makers. Although TI is nominally a U.S. maker of DRAMs, almost aU of its production is in Japan, Taiwan, Italy, and in the near future

Singapore. The rationale of "DRAMs for process driving" is also being scrutinized as never before—one can drive process technology without DRAMs.

Today, with the Japanese stiU controlling 55 percent of the DRAM market, there is no w^ay

"Korea, Inc." can raise prices without giving u p market share to the Japanese (and to Texas

Instruments, Micron, Siemens, and Motorola). As much as Micron needs to be protected by the antidumping laws, which have a valid basis in economic theory and have been historically applied in other industries, the DRAM industry structure does not lend itself to either of these two logical antidumping argvmients.

As the events of 1985 and 1986 show, such concern is not without merit. Mostek, Intel, Inmos, and several others dropped from the DRAM market during this period because of severe financial losses, in part because of each company's failvire to recognize the commitment necessary to be a long-term DRAM player. The market today is much different, to a large degree because of the Korean presence and Korea's willingness to match Japanese suppliers DRAM for DRAM. There currently is no anticompetitive hegemony among producers, and DRAM users

(as long as they are not also producers) may have the best position available today—^immense companies witih substantial technical and financial resoxuces competing for market share gains in the DRAM market.

But with the financial and economic problems in Japan, the perceived threat of a vertically integrated Japanese industry conquering all that stands before it is vastly diminished. Judging from recent public positioning, Japanese semiconductor companies have erhbraced the new rehgion of profitabiUty. Korean companies, on the other hand, are for less capable of damaging established systems businesses worldwide because they hold such a snnall position in PCs, notebooks, mainframes, telecommunications, and other electronics products. In fact, the

Injury to the Domestic Industry

A third, and important, reason for dumping laws is to be able to protect domestic industry from imfair trade practices. In this case, the matter is complicated by the fact that the domestic industry is a complex fixture. Micron is the only fully domestic merchant supplier.

IBM is making noise about moving from behind its captive curtain to enter the merchant market, but so far it can only be a beneficiary of excessively competitive merchant pricing practices. Texas Instruments has been losing money in DRAMs for some time, but most of its present product is from its own manufacturing in Miho (Japan), Avezzano

Otaly), and from Hyimdai, which has acted as a DRAM foundiy for TI for several years.

Motorola makes 1Mb DRAMs in the United

States, but receives its 4Mb DRAMs from

Toshiba and Tohoku SC in Japan, as well as from GoldStar. Motorola also has an OEMinto-Japan arrangement with Mitsubishi. It is therefore not easy to determine, for purposes of the law, what constitutes the domestic industry. Apparentiy, three of the six International Trade Commission (TTC) commissioners wanted to consider domestic SIMM module makers as a part of the domestic industry because they compete against Korean-made

SIMM makers, who may have an unfeir transfer price advantage.

Finally, in the irony of aU ironies, all Japanese manu^cturers that now have facilities in the

United States (including Hitachi, Fujitsu,

Toshiba, NEC, Mitsubishi, and Matsushita) are now a part of the domestic industry that is protected by U.S. antidumping legislation. Of course these companies were all signatories in

MMRY-SEG-DP-3204

©1992 Oataquest Incorporated

December 14,1992

Memories Worldwide the original 1986 Semiconductor Trade Agreement that sought to halt dumping of DRAMs in the United States. Back then, however, they were the ones agreeing not to dump DRAMs.

What constitutes the "injured domestic industry" is not entirely clear at this point, except it is clear that Micron Technology is definitely a part of the domestic industry. Micron has, under the terms of the law, a valid claim.

Whether the others do, or whether they can truly be considered in making the injury case, depends to a degree on the amount of added value that occurs in the United States.

More on Origins of the Law

Having large financial resources and a willingness to lose money to achieve market dominance has for many years been considered an unfair advantage under U.S. law.

This is not the case, however, under the laws of several of the United States' important trading partners. Specifically, the particular legal basis for Micron's antidumping claim can be traced through the Super 301 of the mid-1980s, back through the 1974 trade legislation, and back further to the original law embodied in the Smoot-Hawley Tariff of 1930. Elements of the cost calculation can be traced even further back to 1921 legislation, passed during the sharp post-World War I recession in which the

U.S. GNP dropped 20 percent (and recovered) in the space of 18 months. What was true then is still true today—economic contractior\s lead to price competition, which leads to protective legislation.

Silence of the Lambs

Petitions were filed by U.S. DRAM users in 1985 and 1986 in an attempt to exempt certain SIMMs from tariffs, allowing them to be shipped dutyfree into the U.S. market. SIMMs were ruled to be DRAMs for purposes of the law. The result was the same this time.

The present antidumping proceeding has offered an opportunity for DRAM users and other interested parties to express any concerns they might have about the outcome and impact of the present course. According to a DOC spokesman, DRAM users have filed no such statements.

For the time being, the Computer Systems Policy

Project (CSPP) is mute. Remember that DRAM users, because of a lack of an organized response structure, were shut out from the discussions and decisions that resulted in the Semiconductor Trade Agreement in 1986. DRAM users are on record in a formal March 1990 statement as being opposed to allowing the sale of dumped DRAMs in the U.S. market. Accepting this provision was likely a result of some arm-twisting by the Semiconductor Industry

Association because DRAM users surely showed no reluctance to buy 256K DRAMs in 1985, when prices dropped below $2.00, forcing U.S.

DRAM makers to exit the market one by one.

DRAM users hopefully recognized that they really did need a healthy U.S. semiconductor industry.

Only those with very short memories cannot remember the

Japanese DRAM hegemony as the market moved from surplus to shortage in 1987 and 1988.

Likewise, the Computer and Business Equipment Manufacturers Association (CBEMA), which took a position during the 1987 dumping crisis, is silent. By its own account, it also is not paying much attention to what is happening in the dumping discussion.

Among the various scenarios that may result from the current upheaval is one that would put

U.S. DRAM users, already imder immense price pressure for their own system-level products, at a significant disadvantage in the procurement of

DRAMs that are competitively priced with those from non-U.S. makers. A two-tiered market— with high U.S. prices and low Asian prices—is a conceivable outcome given the present direction of the proceedings.

At the same time, board-stuffing and proairement operations of some U.S. systems companies are now located in Asia, which is currently outside the jurisdiction of U.S. trade law. In addition, Asian motherboard output, including sales into the United States, is specifically excluded from the dumping provisions.

Those Who Cannot Remember the Past Are

Condemned to Repeat It

Only those with very short memories cannot remember the Japanese DRAM hegemony as the

December 14,1992 ©1992 Dataquest Incorporated

MMRY-SEG-DP-9204

Memories Woridwide

market moved from surplus to shortage in 1987 and 1988. The situation was not alleviated by an upwelling of domestic DRAM makers in the

United States. Notably, National Semiconductor was called, but it dedined to reenter the DRAM market. MegaRAM, a business plan for a venture start-up, fell on deaf ears. Intel was preoccupied with microprocessors and later called its departure from the DRAM market the toughest but best decision it ever made. U.S. Memories was stillborn as the market became more balanced in 1990. Alliance Semiconductor, a little start-up running a leased facility in Missouri, bloomed then wilted within months. Eventually, after two years of market tightness, prices began to break late in 1989 and an equilibrium was again possible. they are not stupid. To remove Korean DRAM capacity at this juncture would create a mess in a market that is naturally getting closer to balance on a daily basis. Micron's president, in remarks made at the recent Electronika trade show, acknowledged the late summer upturn in the market, and titie Korean contribution to worldwide DRAM supply is a matter of public record. Such a drastic cure would surely cause far more damage than any alleged dumping has caused, especially since Micron's survival is not threatened. Micron now has six consecutive quarters of profitability, and it is poised to benefit greatly from the market upturn independent of the antidumping outcome.

However, without Samsimg in the DRAM business in 1988 and 1989, market forces would have taken far longer to bring the market into equilibriiun. It has also been Korea that has kept the pressiire on prices over the past two years, maldng it difficult for all DRAM makers—^U.S.,

European, Japanese, and even Korean—to make much money during that time.

Dataquest believes that there are many reasons why a comprehensive worldwide antidumping agreement makes sense.

Settlement Scenario Issue Number

One—Offshore Production

Any immediate resolution that places Korean

DRAM production off limits for sale into the

United States is destined to be only temporary, and ultimately counterproductive. One has only to build DRAMs within the jurisdictional walls of the affected country. Japanese and U.S. producers that perform diffusion in Europe are exempt from the antidumping provisioris of the reference price (RP) system, fri an interesting limitation of the law, Japanese producers in the

United States (notably NEC in Roseville) are not covered by the Semiconductor Trade Agreement, but TI's Miho plant is covered by the agreement.

DRAM makers, without another layer of special considerations, are likely to expand production within the walls of Fortress USA in an attempt to circumvent the present rulings.

Settlement Scenario Issue Number Two—Is

Korean DRAM Capacity Out of Play?

The possibility that the Korean DRAM capacity will be removed from the market is almost nil.

Some of the present driving forces behind the antidumping movement may be intemperate, but

Those who claim that the dumping was a direct consequence of the fact that the Koreans massively overbuilt DRAM capacity in the past three years should look back on the expansionary period of 1984 and 1985 for some perspective.

In 1984 and 1985, Japanese capital spending reached levels (in yen) that still haven't been matched eight years later.

Settlement Scenario Issue Number Three—How

Big Are the Markets?

The Asian market was less than 8 percent of the world DRAM market in 1986, although it was a major outlet for Japanese and U.S. DRAMs and perhaps the most price-competitive region in the world. But with the booming PC business and weakness in both the Japanese and European markets, Asia outside of Japan is running ahead of both Europe emd Japan as a DRAM-consuming region, according to World Semiconductor

Trade Statistics data for the past nine months.

Not only is there a substantial indigenous computer business in Asia, but more than 75 percent of all PC motherboards are now made in Asia.

The tide of U.S. companies moving their boardstuffing operations to Singapore, Taiwan, Hong

Kong, and now the People's Republic of China has proceeded imabated. While some may argue that this is temporary and will reverse as the

Japanese economy rebounds and European

MMRY-SEG-DP-9204 ©1992 Dataquest incotporated December 14,1992

Memories Woridwide

absorption of Eastern Evirope moves further along, other smart money is moving into the Far

East. For Korean DRAM makers, Asia is their fastest growing market, even faster than the U.S. market. are free to sell their product for whatever price they choose. Within the U.S. jurisdiction,

NEC RoseviUe is exempted from the antidvmiping agreements, but it can be covered imder predatory pricing laws.

Why a Comprehensive Agreement Makes Sense

Dataquest believes that there are many reasons why a comprehensive worldwide antidumping agreement makes sense, including the following:

The big profits of 1989 have gradually disappeared into single-digit profitability, or worse, for DRAM makers.

• No significant regional price differentials can be tolerated. Such differentials encourage relocation of productive facilities in response to an economic tilt and support regionalism over globalism at the expense of economic efficiency.

• Asia must be included in the agreement. The only way to do this is through imposed or volimtary (negotiated) restraints on Korean price levels for products it sells into Asia.

The final DOC/ITC judgment is still probably six months away, by which time market forces will probably have driven the DRAM market into a condition in which the outcome is almost a moot point. The larger question lies further ahead to when Japan consumption resumes and the accelerating growth of all the economies of all the world increases demand beyond the current effects as a result of the i486 and

Windows 3.1. •M The opinions and interests of worldwide major DRAM users must be recognized and incorporated into the final resolution. Users must recognize that their interest is vital in the resolution of this matter.

• There needs to be a comprehensive global umbrella agreement, perhaps under the auspices of the General Agreement on Tariffs and Trade, that replaces the reference price system in the EC and U.S. DOC/ITC rulings in a fashion similar to either the present U.S.-

Japan agreement or the EC-Japan reference price agreement.

• Every effort must be made to insure that the formula by which costs are calculated is well known by all participants and reflects a reasonable consensus of all interested parties, including present DRAM suppliers, Taiwanese would-be participants, and major OEMs that use DRAMs. There is currently a host of different laws in place, and many candidate cost formulas that distort trade and misplace financial incentives for both makers and users.

Dataquest Perspective

Summary and Outlook for 1993 and 1994—Profit

Bubble or War of Attrition?

The big profits of 1989 have gradually disappeared into single-digit profitability, or worse, for DRAM makers. TTierefore, it was with some surprise that Hjomdai's preliminary dumping penalty was less than 6 percent.

Hyundai even expressed the belief that the penalty should have been lower. Given the visible pain that Siemens, TI, Micron, and the

Japanese DRAM makers have been in for some time, it is surprising that real profits were obtained at all in this extended period of an interisely competitive market. (The reconstructed cost formula requires an 8 percent profit, implying that Hyundai still made a profit of 2 percent on DRAM sales.) The cost formula used is severe and includes more cost elements than most DRAM makers would use in considering the profitability of their own product line.

• It may even be reasonable to extend the scope of the law to include pricing actions within each of the major trading blocks—EC, Asia, and the impending North American Free

Trade Zone. Japanese and U.S. companies operating within the walls of Fortress Europe are already exempt from the RP system and

When and if the final cost data for GoldStar and Samsung are released, we will get more insight into the true costs of DRAM production. If the final dumping margins are modest, and if the Korearis remain intent on continuing to add capacity to address growing

December 14,1992 ©1992 Dataquest incorporated MMRy-SEG-DP-9204

Memories Worldwide demand, then this profit cycle may not be as robust or as long-lasting as any of the past three profitable periods (1979-1980, 1983-1984, or 1988-1989). Such profit pressure would certainly stress some of the present DRAM makers. It has been a difiiciilt three years, and already we are seeing some of the more marginal suppliers drop from the himt: Oki is in trouble. Sharp has abandoned post-16Mb

DRAM development, and Siemens has shifted strategically and allianced into its future.

Matsushita, after a significant effort to rise into the higher tier in 1988 and 1989, appears to have lost some of its enthusiasm, if not the need to supply DRAMs into its own systems business.

The market has recentiy been further muddied by the antidumping actions in the European

Commimity (EC) and the United States that will have em imcertain impact on the market, but that Dataquest believes will be small compared to the development of the supply-and-demand balance over the coming three quarters. The consequences of these actions may be to keep capacity off the market entirely (highest impact case), shift the regional availability of product, or merely stiffen the tendency of price declines that began in earnest in early 1992. Regardless of this newest twist, much of what follows remaiixs applicable in an environment that is not entirely market-driven.

All DRAM makers need a secular uptiim in profitability to fund the next stage of expansion and product development. If the dumping margiiis are low, and if Korean manufacturers have a cost structure, the financial resources, and a strategic will to keep the pressure on the market, then we may see more DRAM makers recorisider their position and presence in the market.

By Lane Mason

Forward Alliances: Look for Improved

DRAM User-Vendor Relations

Even with minimal help from a weak Japanese market, the DRAM market is expected to grow

75 percent this year in terms of bits shipped.

This is the strongest growth since 1988 and is a clear resporise to the 80 percent reduction in

DRAM prices per bit that has occurred since

DRAM prices began their most recent descent late in tiie summer of 1989.

A New Business Option for a Steady DRAM

Market

Now might be a good time for DRAM users to coiisider making a special effort to ensure volume supplies in a tighter market. Specifically, many of the user/vendor agreements put in place in 1988 and later deserve some review and scrutiny related to their successes and short comings.

Dataquest believes that there can be substantial economic advantages to some more complex supplier/user agreements when compared to deals that are made by sitting across a table negotiating price and delivery on a quarterly or monthly basis. Many creative programs, with equity investments, forward price and quantity guarantees, purchase commitments, and advance product payments have helped moderate the market volatility and reduce the risks for both makers and users that are associated with the fraditional arms-length negotiations between independent DRAM producers and DRAM users.

DRAM makers lost roughly

$4 billion in 1985 and 1986, but they made similar profits during

1988 and 1989.

However, the substantial excess of 0.8nm capacity that existed at the beginning of the year is rapidly vanishing and capital spending cutbacks of 20 to 40 percent in fiscal year 1992 by

Japanese suppliers increases the probability that the downward competitive price spiral will slow as we head into 1993 and on into 1994.

The Background: First 1985-1986, then

1988-1989

Just as World War II is often viewed merely as a continuation (after a pause) of World War I, the supply shortage of 1988 and 1989, and its resolution, had its origin in the demand shortage of 1985 and 1986. DRAM makers lost roughly

$4 billion in 1985 and 1986, but they made similar profits during 1988 and 1989.

The proximate cause of these financial swings tells much about the problems inherent in participating in a high-fixed-cost market, one that

MMRY-SEG-DP-9204 ©1992 Dataquest Incorporated

DecemtKr 14,1992

Memories Woridwide

exhibits price volatility and marginal cost pricing as well as has a high rate of technical change (and obsolescence). DRAM makers lost money not only because bit growth slowed in

1985 (a demand shortfall, with bits shipped up only about 25 percent compared to 1984), but because the excess capacity led to a wide practice of marginal cost pricing, plvimmeting prices

(dumping), and horrendous losses due to imderutilized capacity. A major portion of the capacity built in 1984 and 1985 was never put to the test, was obsolete before used, and had to be written off. the year, but faded by the end of the year), and again at the end of 1987 and early in 1988. As long as the controlling interests in the market were not expanding capacity, capacity in the aggregate would n m behind demand, keeping prices high and allowing recovery of some of the losses suffered from 1985 to 1987.

The shortages of 1988 created problems of their own, this time for users who couldn't get product and whose increasing demand had to be fulfilled on the aftermarket or spot market at high prices.

This was a period of textbook free-market economics, with users continuing to push for the lowest prices from their suppliers. During this time there were virtually no strategic considerations implemented by DRAM users, and not one iota of collective actions on the part of users to preserve a viable, balanced DRAM supplier base. Makers presented themselves better in 1988 and 1989, allocating not so much by price but by relationships. They were certainly better in recognizing that there were mutually beneficial opportimities to enter into user/maker alliances that served the long-term interests of both parties.

This time, however, the user commtmity was forced to reach out and enter into a host of supply-assurance agreements with makers who now were in a controlling position to dictate terms. Although U.S. Memories failed to pass muster and slipped into ignominy in January

1990, some of the user/vendor relations that had their origins during this time frame are just now coming into fruition.

Much has been learned over the years. Perhaps most important is that forward alliances, if properly structured, can work to the benefit of both parties for the duration of the "silicon cycle" and not just in times of severe demand or supply shortage.

Forward alliances, if properly structured, can work to the benefit of both parties for the duration of the "silicon cycle" and not just in times of severe demand or supply shortage.

The consequences of the semiconductor/DRAM losses from 1985 to 1987 impacted the performance of parent companies and forced them to rethink, at the highest levels of the corporation, their role and risks of participation in the

DRAM business. Texas Ir\struments' Board of

Directors in 1985 put explicit limits on the

DRAM exposure it would allow the company to face in the future, and it placed a ceiling on TI's futiire capital investment.

Because of these losses, due in large part to excess capacity, makers were imderstandably cautious about re-expanding their lines in 1986

(when DRAM demand grew in the early part of

The Problems DRAM Makers Face

Full Capacity Utilization-—Build It, and They

Will Come

Full capacity utilization is an important factor in the cost of production equation. For leading edge DRAM manufacturing, facilities depreciation costs are about 25 to 30 percent of the total cost of production over tiie active life of the line. A fab rurming at one-half capacity utilization will have costs that can be about

15 percent higher than a fully utilized facility, other things being equal.

Economies of Scale

If one assumes that investment in process and product design are made oiUy once, then there are vast scale opportunities that await higher volume producers. More importantly, there are the experience-curve advantages resulting from ever-higher volume production. The downside of this, which has tended to limit manufacturers peak run rates, is exposure to price

December 14,1992 ©1992 Dataquest Incorporated

MMRy-SEG-DP-9204

Memories Woridwide attrition, a fell-off in demand, and the prospect of vrnderutilized capacity. Big market presence has meant big risk. So while Hitachi increased its 64K DRAM production to 10 million units per month, it topped out at about

5 to 6 million for the 256K and 1Mb DRAMs and is just now moving u p to 5 million per month for the 4Mb generation.

Having the Money at the Right Turn

The DRAM business is cyclical, generating profits diuing the 18 to 24 good months per cycle, and trying to hold on to as much as possible during the competitive phase of the cycle. One problem many DRAM makers face is they need to bvdld capacity during the counter cycle when cash is short and the future is uncertain. Once the upturn hits, it is almost too late to expand to chase profits and meet demand. The success of the Japanese in the late 1970s and early 1980s has often been attributed to their ability to build countercydically, thus having excess capacity when the market cycles back toward strong demand, enabling them to gain market share in the expanding market.

Demand Assurance

One of the principal motivations for the emergence of more complex user/vendor arrangements was the weak enforceability of long-term commitments. Commercial law and the specifics of the actual pturhase contract allow for either party to renegotiate or nullify such contracts under a variety of circumstances. As users sought secvire suppliers during the heat of the shortages in 1988, they were willing to offer long-term commitments at high prices in return for asstued delivery today.

For a number of reasons, a purchase contract in the semiconductor industry is like no other contract.

better divided the costs and risks of DRAM production and use.

Long Investment Paybacic Horizons

A recognized problem with DRAM production, similar to virtually any advanced technology development, is the extraordinarily long payback horizon. Last summer, for example, IBM,

Siemens, and Toshiba entered into a $l-biIlion

256Mb DRAM development program that is not expected to yield product imtil about the turn of the century. The uncertainty in the future market has proven to be a major deterrent to steady investment for all but the largest and technologically most competent companies.

The Problems that Users Face

Steady Supplies and Competitive Prices

On the other hand, users face problems that are quite different. Until now, most users would say that they wanted assured deliveries at a competitive price. Increasingly, users are benchmarking their DRAM purdtiases against the industry. They want to match the best price available to what their systems competitors are paying. However, a disruption in tiie steady flow of product can lead to delayed system shipments or shipments that are suboptimaUy configured. Rarely do DRAMs offer significant competitive advantages in systems— the vast majority of the DRAM market is undifferentiated, commodity DRAM.

Needs Assessment-Forecasting One's

Own Demand

In 1988, when most big-user volume purchase agreements (VPAs) were maintained at rather stable prices, upsides in demand resulted in severe spot market premiums. Worse still, barring any availability from the established supplier base, the user was forced to enter the aftermarket and pay extortionate, auction prices for products of uncertain history. The aftermarket was a profit boon for new entrant

Samsimg, and a lifeline for second-tier users without established and stable relationships with major DRAM suppliers.

Vendors, certain that these commitments wotild be forgotten G?reached) as soon as supplies loosened u p and prices declined, asked for stronger guarantees and often received them.

A variety of new^ arrangements arose that

Similarly, Apple Computer and other companies had to eventually write down DRAM inventory that was bought at high prices on the aftermarket when the market turned and product became more available.

MIVRY-SEG-DP-9204 ©1992 Dataquest Incorpotated

December 14,1992

10

Memories Woridwide

Not Enough DRAMs

For users, the risk was the downside of not being able to ship product because of a lack of DRAMs. As the saying goes, "For want of a nail, the shoe was lost; for want of a shoe, a horse was lost; for want of a horse, a rider was lost...."

For smaller users or fest growth companies outside the top tier of the user base, it was doubly difficult to even get modest allocations, as md^ers worked hard to accommodate the requirements of their key accounts.

Both Parties Must Face the Purchase Contract

For a nvunber of reasons, a piirchase contract in the semiconductor industry is like no other contract. Prices are routinely contractually renegotiated. Orders are frequently canceled, either by buyer or seller and often for the flimsiest of reasons. Allocation, built into the U.S. commercial codes as a standard practice, is in place during much of the cycle as makers control their oider books and production levels to manage production and prices and control the amount of product in the aftermarket that may come back to compete with new product. reduced risk for suppliers, provided greater s u p ply assurance for users, provided investment capital to suppliers when needed (before they earned their cyclical profits), and ultimately lower costs of production and product delivery.

The following are five examples of these relationships: a Amstrad pic - Micron Technology. In September 1988, Amstrad pic bought 4.0 million newly issued shares of Micron Technology stock at $21.50 each, for a net to Micron of

$77 million (after fees). Amstrad's then

9.2 percent equity stake entitied it to buy u p to 9.2 percent of Micron's DRAM output at the market price. This gave Amstrad an assxired source of supply and an independent equity investment in the volatile DRAM business. Micron got new cash to expand its facility, but not a guaranteed sale.

(Amstrad decided to sell its stake in May

1990, but later retracted its offer. Micron's stock was selling at $13.50 per share. Today, after a recent jump of $5.00 over the past month. Micron stock is selling for about

$20.00.)

Forward alliances clearly are a means for independent semiconductor companies to take advantage of a close supplier-user relationship.

• Micron Technology - Take or Pay Contracts.

In Jvme 1988, Micron annoimced that about

20 customers had agreed to enter into "take or pay" agreements that had the following conditions:

• They were long-term purchase agreements, running u p to 24 months

• Prices were referenced to the pricing at the time of the agreement, about $4.05 for 256K

DRAMs

In tight markets, users promise to order products for long periods just to get the product today. They later cancel oniers, renegotiate prices, or downsize their order when the time comes to take delivery after the market has slacked. Vendors do just the opposite. In slack markets they promise users that their needs will get first consideration when the market tightens up.

What the period from 1985 through 1989 demonstrated was that the traditional user/ supplier market resulted in relationships that involved unacceptably high risks given the experiences of 1985 and 1986 for vendors, and

1988 and 1989 for users.

It was from these experiences that companies began to look at innovative relationships that

• Prices were allowed to decline or rise as the market dictated, but by no more than

10 cents per quarter

• Users agreed to take purchased amoimts at the agreed price dictated by the formula, or pay anyw^ay

Micron was thus able to improve its achievable price many quarters out, after the market price might have dropped outside the formvila price band. Even with these contracts, some litigation ensued as users sought to renege on their purchase/price commitments. Micron still fared better than if it hadn't set u p such arrangements.

December 14,1992 ©1992 Dataquest Incorporated

MMRY-SEG-DP-9204

Memories Worldwide

11

Take-or-pay deals are common in other high-fixed cost industries (electrical utilities, natural gas, and water) as a means of assuring a sufficient revenue stream to cover the cost of fixed investments, regardless of the variable amount of product delivered.

Texas Instruments. Texas Instruments has been the foremost practitioner of externally funded facilities expansion and strong customer supply relationships. In mid-1988, TI began attracting major DRAM users that were concerned about the building concentration of the industry's submicron capacity in Japan.

Eventually, this group of users funded TI's accelerated capacity expansion with more than $100 million.

16 percent each. About $80 million in equity was supplied by the owners, and loans for

$160 million were negotiated.

In the TECH Semiconductor, TI-Acer, and KTI

(Kobe Steel's JV with IT) joint venture arrangements, the investors ane both equity holders in the venture (thus desirous of high prices and profits) and users (thus wanting low prices).

Investor-users are not required to take output, but may take a percentage up to their equity shares, usually at the market price. (The KTI venture is different because Kobe Steel doesn't need semiconductors for its own use, but needs the facility as a lever for its entry into the merchant semiconductor market.)

What was begun in a time of panic, however, proved to have broader appeal, and user investment continued after DRAM prices dedined rapidly in late 1989. In these deals,

TI got what were essentially advance payments for DRAMs, pa5rments that were ttien rebated on a pro rata basis as investors later bought their DRAMs from TI.

Why Such a Relationship Can Work to Benefit

Both Parties

There are several significant costs associated with using the market that are not always apparent, nor readily measurable. They are, nonetheless, real costs that must be recognized from time to time. The following are some of those costs:

Texas Instruments got the facility when it needed it. Users got supply assurance, and eventually got or will get their investment returned. By requiring advance pa5Tiients, TI assured the commitment of its partners to take the designated amount of product.

• Underutilized capacity. Makers face the cost of underutilized capacity in the event that the industry enters a supply-excess condition.

Fixed facility charges that had to be spread over a third or a quarter of the anticipated demand were the proximate cause of the massive losses in 1985 to 1986.

Texas Instruments-Acer. In May 1989, Acer joined TI in a joint venture to biuld a fab in

Taiwan. Acer wanted DRAM supply assurances and greater independence from

Japanese DRAM sources. TI was anxious to move ahead on its facility expansion program, almost on a risk-free basis.

TI-Acer is now nearing volume production of

4Mb DRAMs (1 million per month) on the joint venture's O.Sjxm line. This facility also has 16 percent of its shares owned by the

China Development Corporation and a special issue of stock sold to several financial institutions.

TECH Semiconductot In April 1991, Canon,

Hewlett-Packard, and the Singapore Economic

Development Corp. (SEIXZ) joined Texas

Instruments in establishing TECH Semiconductor in Singapore, Canon and HP each own a 24 percent share, SEDC and TI own

• Information costs of using the m a r k e t

Although DRAMs sell in high volumes and have a lower marketing and selling cost than other products (as a percent of selling price), there are stiU some marketing and selling costs. Volume discounts to big customers reflect the savings that are achievable by selling products in large blocks. Traditional user/ vendor relationships also lower the costs of defining the market with every new product, an advantage to the established suppUers.

• Inventory costs. Uncertainty of demand or of supply can cause the supplier and user to bear excessive inventory costs.

• Supply disruptions. Many medium and small-sized DRAM users suffered mightily during 1988 and 1989 when small allocations from DRAM suppliers either forced them to ship products with suboptional DRAM configluations or held u p systems shipments entirely.

MMRY-SEG-DP-9204

©1992 Dataquest Incorporated

December 14,1992

12 Memories Woridwide

• Price volatility. Price volatility works both ways. In tight markets, users pay premium prices by acquiring incremental product on the spot or in aftennarket piirchases. In slack markets, suppliers have to move quantities down to fair market value or reference price levels, guaranteeing inadequate long-term profits.

The arrangements set forth beginning in 1988 and 1989 occupy the middle ground between traditional VPAs and captive production.

Attempts are made to reorganize, assess, and assign a real value to various risk elements and design an agreement to minimize their cost impact on the two parties involved.

Potential Cost Savings Quantified

An example of a fully implemented 16Mb relationship between one or more users and a supplier that has already developed a given DRAM product and process is presented in Table 1. One can see the power of this idea, which offers the following benefits:

• Incremental product and process development costs are nil

• The facility can run at an average higher capacity utilization

• Marketing and selling costs are reduced

• Risk is reduced significantiy but is also difficiilt to measure

Therefore, because of the potential for reduced costs, these relations can work to the benefit of both parties by reducing the overall costs of participating in the market. The user can be a profit-making investor in the joint venture and get competitive DRAM prices at the same time.

Forward Alliances Compared to Truly Captive

Production

One might ask how forward alliance agreements differ from the captive relationship that exists between NEC Semiconductor and NEC's computer business. The most obvious difference is that a relationship such as NEC's is not available to U.S. companies, which are largely not vertically integrated. Forward alliances dearly are a means for independent semiconductor companies to take advantage of a close supplieruser relationship. In a forward alliance, the relationship (contract) is also negotiated between two or more independent entities, not separate divisions operating within the same company.

The risk/price/investment analysis is more welldefined and is performed by different parties who absolutely have different interests in mind.

There is no top-level dictate that guides investment and production decisions. Additionally, forward alliances are flexible, temporary partnerships (though surely renewable) allowing a partnering strategy that enables either party to adapt more easily to changing business requirements.

Finally, forward alliances can be hedged with one another so that a company can develop a portfoho of forward alliances with a range of partners across the industry. When one is tied to a single biggest customer, certain degrees of freedom are reduced. For example, every fab that

Table 1

Potential Cost Comparison for 16Mb DRAM with Different User/Vendor Relationsliip

FadHty Cost (Millions of Dollars)

Product/Process Development (Millions of Dollars)

Average Capacity Utilization (Percent)

Variable Costs (Millions of Dollars)

Total Cost (Millions of Dollars)

Total Die (Millions)

Average Cost Per Die ($)

Mark-up

Price ($)

Source: Dataquest (December 1992)

16Mb

Normal

350

200

75

244

794

112.6

7.05

2x

14.10

16Mb

Forward Alliance

350

0

85

288

638

127.6

5.00

1.85x

9.25

December 14,1992

©1992 Dataquest Incoiporated

MMRY-SEG-DP-9204

Memories Worldwide

13

Motorola has could have a different partner, each serving a different and changing requirement of the time.

Market price-based systems can tie the transfer price to the seller's lowest, average, or best prices achieved with other DRAM customers.

Or, the transfer prices can be related to the best, average, or highest prices that the user has achieved during the same period.

Forward alliances, like any private activity in a commodity market, will lead to other distortions in the overall market.

If done properly, however, captive relationships similar to the ones that all Japanese companies have could be made to function in the same feshion as forward alliances. NEC Computer could be a major forward alliance partner of

NEC Semiconductor if the deal was structured similarly. But as IBM, General Motors, and

Digital Equipment are discovering, traditional internal business relationships have an inertia of their own and are often hard to change.

What U.S. industry, and not just the semiconductor indvistry, has fotind is that a truly captive relationship under a single management umbrella Wdes defects in the allocation of resources and assumption of risks. Such an organizational arrangement also limits the user from accessing extemaUy-developed technologies and from taking advantage of price changes from the competitive outside market. The mvdtipUcity of conflicting business objectives in an integrated operation is inade explicit by disengaging and renegotiating interdivisional agreements as distinct business interests. Integrated wholes are replaced by smaller business units coimected by a network of strategic alliances.

Forward Alliance Issues

The Price Problem

Transfer pricing for products sold \mder these types of arrangements has proven to be a difficult matter to resolve. Generally, prices are cost-based (plus a fixed percentage), market price-based, or reference price-based, where the price is tied to an external reference price.

Reference price systems can tie the transfer price to an external number such as the historical rates of cost reduction (the experience ciirve), the EC's own cost-based reference prices, reconstructed costs such as those used in fair market values, or concurrent costs from companion, or sister, facilities.

If one believes that future costs are more or less predictable over long periods (the experience cxirve), one should then be able to agree on an equitable forward pricing airangement for u p to fotir or five years. Such an assurance would help to both guarantee the makers of a positive revenue stream and challenge them to beat the curve and make substantial profits.

Forms of Investment

In the examples outlined in this article DRAM purchasers have used several forms of investment, but in aU cases something was obtained in return. We have talked about loans, equity investments, advance payments, and more firmly fixed forward price relationships. In return, users and investors either received product or options to buy product at a price specified in tiie agreement.

Impact on the Larger Market

Forward alliances, like any private activity in a commodity market, will lead to other distortions in the overall market. If more of the industry's production moves through such preordained channels, the remainder of the market is made even more volatile. If 50 percent of the product flow is fixed, an unexpected 25 percent increase in demand creates an apparent 50 percent increase in the half of the market that remains outside the forwardalliance contracts. This is not to say that individual companies cannot benefit from such agreements, nor to say that individual companies that can forecast their own demand won't be entirely better off than those that cannot.

Indeed, such agreements can insulate good forecasters from the ravages created in the marketplace by unforeseen events arising from a host of different catises.

Dataquest Perspective

Outlook for the Future

Big producers can undoubtedly produce for less. The DRAM production economies of scale

MMRY-SEG-DP-9204 ©1992 Dataquest Incorporated

December 14,1992

14

Memories Worldwide are immense. What traditionally, and undesirably, has gone hand in hand with large-scale economies are risk and exposxire to market

Tuicertainties, as well as the sheer capital requirements of making the investments to produce on a massive scale.

Forward alliances address both of these problems for the DRAM producer by allowing producers to receive financing from their customers and by having a guaranteed outlet for the products once they are produced, \^^th next-generation process development and fedHties costing $700 million to $1 billion, it is clear that a more effective institutional arrangement among the various elements necessary to make u p a market (technology providers, manufacturers, users, and financiers) is possible.

Furthermore, by enabling makers to expand countercydically the market itself can be steadied over time, thereby reducing price volatility and giving better forward price visibility and availability of product.

The opportunities for more effective user/vendor relations that reduce risk, uncertainty, and cost are just beginning to be explored. The examples outlined in this article relate specifically to high-voliune commodity memory products with high fixed costs of production, large-scale economies, and high price volatility.

Dataquest believes that there is significant potential for improved price performance in forward DRAM and memory pricing that can be achieved through resource-sharing alliances such as those described here.

The uncertain future that both makers and users face today as we look ahead to 1993 and 1994 may provide the incentive to again explore new user/vendor arrangements that provide both supply and demand assurance and price predictability at a substantial cumxilative cost savings over a generation of

DRAMs, or over a four- or five-year silicon cycle.

By Lane Mason

December 14,1992 ©1992 Dataquest Incorporated MMRY-SEG-DP-9204

16 Memories Worldwide

For More information

On the topics in this issue Lane Mason, Director/Principal Analyst (408) 437-8120

About upcoming Dataquest conferences (408) 437-8245

About your subscription or other Dataquest publications (408) 437-8285

Via fax request (408) 437-0292

The content of this report represents ovu' interpretation and analjrsis of infonnation generally available to the public or released by responsible individuals in the subject companies, but is not guaranteed as to accuracy or completeness. It does not contain material provided to us in confidence by our clients. Individual companies reported on and analyzed by Dataquest may be clients of this and/or other Dataquest services. This information is not furnished in connection 'with a sale or offer to sell securities or in connection with the solicitation of an offer to buy securities. This firm and its parent and/or their officers, stockholder, or members of their families may, from time to time, have a long or short position in the securities mentioned and may sell or buy such securities.

December 14,1992 ©1992 Dataquest Inconxvated MMRY-SE6-DP-9204

Dataquest Perspective

^ Not Remove

Memories Worldwide

MMRY-SEG-DP-9203

August 24, 1992 in Tliis issue...

lyiarlcet Analysis

Market Analysis

Indusby Trends: WHI the 1Gb DRAM Be a Reality?

Industry Trends: Wll the 1Gb DRAM Be a Reality?

This artide provides a long-term analysis of cost and returns on 1Gb DRAM development. Our

(Note: This article is derived from a speech given by

Lane Mason, Dataquesfs principal analyst ^ semi-

analysis points to the need for new ways to think

conductor memories, at the recent Semicon West

about, and fund, deep product development

By Lane Mason Page 1

Equipment Trade show in a session sponsored by

Dataquesfs Semiconductor Manufacturing and

Applications group.)

IBM/Siemens/Toshiba 256Mb DRAM Venture Breaks

New Ground in Industry Cooperative Undertaking

Presuming a straightforward extension of the

DRAM technology that has existed from the 4K

The $1 billion joint venture to develop the 256Mb

DRAM and accompanying 0.25(un process breaks new ground in collective technical efforts and poses new challenges to industry competitors. to 16Mb levels, die technical parameters of the

1Gb DRAM can be specified today, a full decade before we may be expected to see amy sort of

"engineering samples." But there is growing

By Lane Mason Page 8

concern in industry drdes vtrhether such a product will ever become a reality, and whether the massive advance investment required to bring the product to market will ever reap a return. Basically, the uncertainties sunoimding this question are economic: size of investment, lengthening advance development requirements,

' and tincertain character of the memory market when such products could be expected to generate revenue and profits.

Dataquest

aoMnpanyof

The Dun & Biadsticct Ccxpoiation

Dadaquest is a isgistNed bademaik of A.C. Nielsen Company.

The Case for Technology

For generation after generation of DRAMs, pessimists have pointed to the impossibility of solving the technical problems they were sure to face as successive DRAM generations came. One after another, these "Maginot Lines" were overrun, and prices came down further than anyone had anticipated.

As early as the 64K generation, some companies had a hard time shnnking their die to get it into a SOO-mil package, but now all 4Mb devices can accommodate it. Power per unit has given up ground very slowly, rising from about 300mW at the 4K level, to 550 to 600mW at the 4Mb level

Soft errors were going to kill us as early as the

64K generation, but we hear almost nothing

File behind the Perspectives tab inside the binder labeled

Memories Worldvnde

©1992 Dataquest Incoipoiated, Repradudian PiohibitBd 0013613

J0«m

fe;.."" ».-Wii^^

s * ^ '

Memori^ Worldwide

about that today, a cJecade ^ ^ F A n o t h e r kiUer, test, has itself ctied a death of ignominy. Ultimate yield wovild never approach the 90 percent levels we had become accustomed to at the 64K generation, but advanced 1Mb hnes now routitiely have probe yields exceeding that amount.

Redundancy, "the old man walking with the cane," solved all that, and was incorporated into the bag of tools rather effortlessly. Furthermore, companies are consistently able to ramp u p to such yields fer more quickly than in the early days. Masking layers and other measures directly translatable into capital costs would outstrip profit potential. But Micron Technology will have ite 10-mask 4Mb up and running by year-end. IBM's ECC, shown at the 1990 ISSCC, goes further still in showing the ability to quickly get high yield on advanced chips. IBM recently showed a single transistor, suitably sized for a 4Gb DRAM.

If history tells us anything about the technical barriers, it is that they pose weak resistance for the can-do DRAM makers that have promised and delivered thousandfold improvements in price per bit over two decades, and promise continued similar gains into the 1990s, as well.

Some noteworthy developments and costreduction methodologies will fuel the upcoming phase of advancement and cost reduction.

Figure 1 shows Micron Technology's 1Mb product roU-out, showing five generations of device, ending with its now-in-production

"hypershrink" 1Mb DRAM, which at 17 square mUlimeters is about 35 percent smaller than the next-smallest 1Mb DRAM die in the industty.

Micron's 10-mask 4Mb DRAM, to cite another of its cost-reduction strengths, will be in fuU volume production by year-end 1992.

Figure 2 shows WaferScale Integrations' technology waiting in the wings for licensees NSC and

AMD to bring high-volume production. For

NVM proponents, it should be noted that the achievable EPROM cell size using the AMG is less than half the size of a typical 4Mb DRAM using similar design rules.

These two near-term examples show the types of capabilities that have achieved the economies we have today. Still further out, we are reminded that fully functional 64Mb DRAMs were shown at the 1990 and 1991 ISSCC, and that key elements of the 256Mb DRAM have already been demonstrated. Extrapolating such trends into the

21st century, when tiie 1Gb DRAM will or won't be a reality, is far more difficult, and requires that other technical issues either be dealt with or ignored. We are confident in the beUef that when the problems are stiffidently weU-defined, they wiU be solved.

The following statements, which appeared in

Sematech trade press advertising a few months

Figure 1

1Mb Possible DPW and Die Size Comparison

Possible DPW

900

Original

(A10A)

Source: Micron Technology

August 24, 1992

Fast Shrink

(AIDS)

Second Shrink

(A10U)

1 Meg Generation ~ Mask Set

Super Shrink

(D12A)

Hyper Shrink

(D16X)

G20003B7

©1992 Dalaquest Incoipofdsd

MMRY-SEG-DM203

Memories Woridwide

Figure 2

Cell Size Comparison

Industry-standard EPROM technology versus WSI's patented EPROM technologies ff^^'lV staggered

Virtual

Ground

10.8M

4.25M

Source: WaferScale Integration

02000388

ago, show the company's view of the future for the next 15 to 20 years:

• Over the past 35 years, the cost of integrated circuits has decreased an average of 25 to

30 percent per year. The accompanying increase in productivity of more than one million times is expected to continue for at least two more decades, well into the 21st Century.

• In 1990 an integrated duxniit manufacturing facility capable of starting 20,000 wafers per month cost $400 million. By 1995 the cost will exceed $1 billion. By 2005 projected costs are in excess of $2 biUion.

So why the concern that we will not be able to replicate the same progress, for as far as the eye can see? The argument is built aroiind several lines of discussion, all of which are fundamentally financial and economic in nature. Indeed, if we are trying to build a single 1Gb DRAM, we can almost do that today. Even today, a wafer full of advanced 4Mb DRAMs has about 1 billion bits on it already. Obviously, we need to tighten up the question a bit, in order to imderstand the direction the industry takes as it increasingly ibces tough issues later in the decade. We need to understand tiie economics of investment and production, the market development, the likelihood of profitability, and other trends taking place throughout the industry.

Sematech might be called the "Inertial Optimist" that foresees continued cost improvements of

ICs of the order of 25 to 30 percent per year for the next two decades, which is at least 300-fold improvement in cost by 2012. This is optimism!

At the same time, it foresees Polities costing at least $2 billion by 2005.

From a technology viewpoint, these concltosions are all quite reasonable. IBM has already denxonstrated the single transistor for a 4Gb DRAM.

There appear to be no obvious device technical barriers that cannot be overcome.

A better question is the relationship between the process employed (and suitable fadlity employing it) and the price per bit of the DRAM device

. b ^ g made. Prices govern the substitution of older products with newer ones. It is absolutely necessary that we achieve a successively lower price per bit to have the 64Mb replace the 16Mb, the 256Mb replace the 64Mb, and the 1Gb replace the 256Mb.

But at the same time, to encourage continviing advanced development of later generations of

MMRY-SBrDP-9203

©1992 Dataquest Incoiporalad August 24, 1992

Memories Worldwide

DRAM, to make all the investment worthwhile, we need to believe in the prospect of getting a return on our investment. Putting technology aside for the time being, individual companies need to see that the massive sums required to gain the high ground in the 1Gb generation will eventually pay their way.

But there are many trends in the indvistiy besides those that see the 1Gb DRAM as the natural outcome of an additional 10 years' and

$1 billion investment. There are trends that heavily impact the prospects of achieving adequate financial return, which are almost certain to deter many "wannabes" from actually being there when the time comes.

Process Development Costs

Costs to develop the basic 0.2}Jm process for the 1Gb DRAM can be estimated to be from

$800 million to $1 billion, expended over seven to eight years in advance of the first prototype development of the product. Both the time and the money spent in advance are lengthening, and the total amotmt for each is becoming greater. A considerable portion of this expenditure can be saved through the proper time-phasing of process development, through burden-sharing with development partners, and through a broad amortization of the cost across a wide revenue base of products.

Facility Costs

It comes as no surprise to anyone that each generation of fedlity costs more than the prior generation, holding the number of wafers or wafer starts the same. Historically, these increased costs have been more than justified by the increase in productivity. This is the im.portant measiire of whether the new facility was worth it: Did the depreciation and process amortization per bit decline as a result? So far, that has dearly been the case. Will it be the case into the foreseeable future? We don't know. It depends on the lifetime of the equipment, and the depredation rates. Figure 3 shows Mitsubishi's estimates of the investment required for successive generations of DRAMs, although others believe that the curve is actually steeper as we move to the right.

Profit Prospects and the Balanced Marketplace

Companies need profits to naove to the next leveL In theory, this must also be trae on a product-by-produd basis, though the accounting practices that determine DRAM profitability are hardly dear and dean, except in the smallest, most tightly focused DRAM producers, such as

Micron Technology. Historically, DRAM makers have profited significantly in about two of every four or five years—enough to charge ahead with process development and expand facilities for the next-generation produrt. The most recent

Figure 3

MitsubishPs Estimated Investment for Successive DRAI\/I Generations

Investment (3 Million Units/Month)(Billions of Yen)

10,000-

1,000-

100 •

10-

^—-rt^.-Lii**

t^l^^H^i^*^^

1 i i f T ^ i _^

^^g^f,- ••'"•••;'< ^^irfB—-"

<!:j5>ii i ^ i j j i i i i - " ' ^

10K

Source: Mitsubishi

August 24, 1992

100K

1M

Bit Size

10M

©1992 Dataquest incorporated

100M

1G

G2000989

M«RY-Sa3-OP-S203

Memories Worldwide

period of high profitability was during the

1988 to 1989 shortage. Prior to that, DRAM makeis made good profits in 1983-1984, and in

1979-1980. But if the period from 1990 to 1992 was typical of the margins achievable over the entire silicon cycle, there would be far fewer

DRAM suppliers today.

It may be that a shortage, or at least demand growing faster than supply, is needed every three or so years for the DRAM business to succeed. However, with the Korean DRAM makers keeping the pressvue on the dominant Japeinese suppliers, it may be harder to get an out-ofbalance situation in the market in the future.

Every reluctance to follow the prices down has resulted in incremental gains in market share by the Koreans. So long as there is no hegemony, prices and costs of production wiU travel down in lockstep fashion, denying the opportunity to make the kinds of cyclical profits that appear to be necessary to fuel the continued prosperity of

DRAM makers. for technical development. While the absolute size of the industry is greater, the scope of its technical effort is also broader. Relatively smaller investment leqviirements and greater future opportunities in the past led to a "sprint to tomorrow" attitude tiiat has since been moderated by slower growth, tougher competition, and greater financial risks. Returns on investment must be scrutinized carefully.

Absolute Market Size

The stuns required by major players in the industry are immense. The capital demands by the semiconductor division are no longer payable out of petty cash. The semiconductor division management must work harder to justify net cash flows from the parent com.pany.

One has only to look at the profitability of the semiconductor groups' impact on total corporate earnings, and the recent poor performance of

Japanese companies, to see the impact of the next-generation process and facility costs.

However, with major positions in DRAMs held by Koreans and Japanese companies today, there exists a reasonably close balance of supply and demand, making a significant shortage condition—and its period of high profitability— virtually impossible.

Intellectual Property

Intellectual property is another factor that has raised its head in tiie past five years. There is now an embedded cost of production, in terms of royalty pajrments to the laesy DRAM patent holders, of about 10 percent of sales. This amount is now being paid by most of the recent entrants, and smaller amounts by the larger, more established producers such as the major

Japanese DRAM makers. Wiih thin margins, few companies can afford to pay out yet another

10 percent of sales to Texas Instruments, Hitachi,

STTM, and Toshiba for royalties. Already, these payments have impacted the rate at which the price per bit has declined.

Where the Industry Has Found the

Best Profitability

The industry has seen its profitability impacted severely since the supply-demand balance was restored in late 1989. Ultimately, companies have to fund their growth out of profits. But aside from major x86 monopoly profits at Intel, aftertax returns for the top 20 companies constituting

95 percent of total industry sales have probably beCTi in the 2 to 3 percent range. The top five

U S . companies (Intel AMD, NSC, TI, and

Motorola) have 58 percent of total U.S. company sales; profits for 1991 were about $1.3 billion on sales of $13.4 bUlion. But if the monopoly x86 profits are excluded from Intel and AMD, that

10 percent drops to 3 to 4 percent, with NSC and TI suffering substantial losses. This hardly gives solace to those that woiild like to invest hundreds of miUions of dollars in new DRAM product and process development. Smaller companies playing in design-ridi niches fared much better.

Slowing Semiconductor Marlcet Growtli

The industry as a whole has seen its revenue growth slow from about 15 percent during the

1980s to an anticipated 10 percent during the

1990s. Historically, growth has opened u p new market opportunities, and with tiie profits growing at the same pace, provided new resources

The big three European companies are worse still. SIM has been losing money almost since its inception ftnit has reaped substantial royalty payments for the Mostek and Inmos patents).

Siemens has rarely been profitable. Philips' semiconductor group has lost large stuns for more than two years. Other European companies have done no better.

MMRY-SEG-DP-gaOS

©1992 Dataquest Incorporated

August 24, 1992

6

Memories Woridwide

In Japan, the aforementioned corporate results attest to semiconductor division and DRAM profitability.

In the slow-growth market of 1990 to 1992, profits have been foimd in design-rich products, proprietary architectures, cop)nighted microcode, and proprietary algorithms, and not in commodity memories such as DRAMs. Although the financial losses in memories have been nowhere near as severe as in prior soft markets, the profit potential increasingly appears to be in the proprietary parts of the market.

Growth needs the fuel of profitability, especially as the trend away from continued "parental support" and semiconductor group accountability gathers momenttun. Today, the semiconductor divisions of all European companies and several

Japanese companies are under do-or-die dictums because of their negative impact on the parent companies' performance.

Who needs a 1Gb DRAM, and wUl four imits of

256Mb DRAM do as well? Or will the interests in the manufacturing commimity shift from manufacturing excellence that confers no sustainable competitive advantage? Already we have seen that, absent a shortage, returns on intellectual property (via the Intel or the TI method) and design excellence are the two highprofit techniques for the 1990s.

Evidence of PPB Rate of Decline

The floor price at which a fully depredated facility can run a given density of DRAM is increasing from generation to generation. This marginal cost covers only recovery of variable costs. Today, 256Ks, running in near-zero-cost facilities, sdl for about $1.60, beheved to be about as low as the part can be made. 1Mb

DRAMs sell for about $3.00, and almost no one beUeves that this part will drop much lower and still allow the maker to recover all costs.

Slowing Bit Growtti Rate

Those that believe in the experience curve must conclude that the bit growth rate, which has slowed since 1985, will have the certain effect of slowing the rate at which costs of bit production wiU occur. From the mid-1970s through the mid-1980s, DRAM bits grew at more than

100 percent per year. Since 1985, however, bit growth has averaged just 70 percent per year, and the consensus is for a continued slowing for the remainder of the decade, at a bit growth rate of 50 to 60 percent.

Therefore, in just one generation, the floor has risen by more than 75 percent. Most forecasts for the 4Mb DRAM, now selling for about $11.00, are that it will vjltimately reach $5.50 to $6.00.

This is an anticipated, not achieved price. One can only guess at the floor price of the 16Mb

DRAM, but we have never failed to exceed our expectations and a priori analysis of where the lowest price can be. Careful analysis for the

1Mb, done in 1988, concluded that about $3.80 would be the limit. This was reduced to near

$3.50 late in 1990, and spot prices in April were as low as $2.50 for inventory sell-ofts. Today, prices are about $3.00 and appear stable.

Where Will Demand Come From?

One problem that is clearly in evidence is the make-up of demand for the coming years, even to the extent of increasing bit shipments at a heretofore modest 50 percent per year. Software such as VN^idows 3.1 has replaced hardware as the primary memory driving force. Software is a new force in DRAM demand, and is subject to production, distribution, and utilization patterns of its own. In its most DRAM-intensive form, even HDTV will use only 8MB, or just fotir

16Mb chips in the 1995 to 1996 time frame. The ability of memory piwiucers to produce dense

DRAMs is running far in advance of users to make use of them.

Payback Period

\^^th advanced development taking large sums five to seven years in advance of major returns, the interest rate become more important in evaluating the return on investment. Figure 4 shows Hitachi's view of the increasing lag between process/product development (read investment) and appearance in the marketplace.

Higher interest rates at which capital is diverted into advanced process development mean that future prices necessary to achieve an adequate return wiU necessarily be higher. Japan is no longer the coimtiy of free capital

August 24, 1992 ©1992 Dabquest Incorpoisted MMHY-SE6-DP-9203

Memories Woridwide

Figure 4 iHitachi's View of Process/Product Development versus Maricet Appearance

Bit Size

1G

256M

64M

16M

4M

1M

256K

64K

Announcement

(Research Result)

y^ ^ ^

r>n

1980

y y

/ 5 Years

^ 4 . 5 ^^^^

j& 1 *^^ — —

1990

Mass Production

(1 Million Pieces/Montli)

2000

Source: Hitachi

G2000990

Dataquest Perspective

mz\ Are We to Do?

In the face of such forbidding economic prospects for future generations of DRAMs and other products, what options are available to industry participants to postpone the day of reckoning and keep the tedmology juggernaut running one or two more generations before slowing down?

Although they do not make the future any less expensive in the aggregate, many trends now in evidence do reduce individual companies' costs and promise a better return on investment for them. across product lines. Processes are more transferable, and fadlities are more flexible in their application, which reduces costly incompatabilities, and improves their ability to transfer processes from one facility to another and always run the highest-revenueper-wafer product in the fabs.

• Defend IPR: TI gains a return on its

$500 million-per-year R&D investment through embod3dng that technology in its own products, and through licensing others to use it as welL This cost recovery has proven to be a valuable means of getting an adequate return on that R&D investment.

Collective Actions for Cost-Sharing

Collective actions are also attractive. They reduce redundancy of effort and make the collective industry expenditure for R&D and facilities more efficient

TI highlights three practices that are a part of its strategic effort to mitigate the escalating cost of advanced development and improve its return on investment, as follows:

• Harmonization: Define processes to have the highest possible degree of commonality

• Customer-funded facilities: TI has gained about $1 billion in capital costs through its innovative programs that allow it to partner with its customers to build facilities in advance of demand. Indeed, this transfers risk to others and also brings in some capital from the future.

Joint ventures and coUective actions, such as

Sematech and any of several joint development agreements in advanced DRAMs, are attractive

•ways of reducing the overall costs of future product development.

By Lane Mason mm-SB^p^m

©1992 Dataquest incorpoiated

August 24, 1992

8

Memories Worldwide

IBim/Siemens/Toshiba 256Mb DRAM

Ventme Bivaks New Ground in Industry

The joint venture among IBM, Siemens, and

Toshiba announced July 2 is another step in the indTistrj^s trend toward massive cooperative ventures to lower the cost and risks associated with advanced process development. In one stroke, it cuts the private cost of process development for the 0.25nm process by two-thirds or more and poses a significant economic challenge to companies that fancied going at it alone. devices are now in production at the existing

IBM facility at CorbeU-Essones, and are being marketed by Siemens. IBM and Siemens also have a 64Mb development program in place.

Apparently seeking to dispel notions that

Siemens was re-evaluating its positioning in

DRAMs or semiconductors, Siemens President and CEO Karlheinz Kaske commented that the joint venture "contributes to future applications in telecommunications, and assures our customers of otjr engagement in microelectronics."

This article discusses the particvilars of this most recent megaventure, along with its significant implications for the development of ttie industry for the remainder of the 1990s.

Toshiba and Siemens have a relationship that began with the 1Mb DRAM in 1985, which transferred the Toshiba 1Mb design and process to Siemens in exchange for a fee and continuing technical support.

Basic Tenets of the Agreement

The basic agreement calls for Toshiba, Siemens, and IBM to collectively develop a 256Mb DRAM design and the 0.25|Jm process on which it can be manufactured. The limit stojw at the end of the development stage and at present has no provision for manufacturing (which might run into a host of antitrust objections). The group estimated that the program would entail aggregate expenditure of more than $1 billion to develop ihe 256Mb DRAM and qualify it for production late in the decade.

Also, IBM and Toshiba within the past few months have negotiated a technology agreement to develop solid-state files (SSFs) iising Toshiba's

NAND Flash technology and IBM's advanced controllers and interface technology.

Clearly, the prior arrangement between Siemens and IBM and the addition of a 256Mb agreement will make it easier to keep the process and product program on a steady path.

IBM's Advanced Semiconductor Technology

Center in East FisWdU, NY will be the principal initial focus of development activity, with supporting projects being undertaken independently by Toshiba and Siemens. The program is expected to employ more than 200 researchers from the three members at its peak.

Earlier Agreements

At present, IBM has an agreement to produce

16Mb DRAMs in France with Siemens. These

Financial Risk and Cost—The Prime Mover for Alliances

All other reasons aside, the prime mover for this agreement is cost and risk As underscored elsewhere in this issue, the calculus of return on investment on deep process development is horrendous. One has only to look at IBM's massive investments in X-ray lithography to see the difficulty of the problem: year aiter year, tens or hundreds of millions of dollars were invested to try to catch a receding goal It is small solace to

IBM to be the X-ray leader. It has cost close to a billion dollars, without appreciable return.

According to IBM, each participant will also be allowed to resell the technical fruits of the joint venture, making it possible that any of these companies could r^luce its net financial commitment significantly by achieving a royalty stream to compensate for the immense development costs. In addition, though the process wiU initially be developed for the 256Mb DRAM, each peirty is free to enhance and modify the common process and apply it to other products, including logic devices.

Development of a 256Mb technology is a sirrtilar program, requiring significant years of investment in advance of any return, fraught with timing uncertainties of market development and pushing into the luiknowns of technology development. In sheer magnitude, it is on the same scale, and no one, not even IBM, is rich or smart enough to go it alone. The risks are too great and tiie costs are too large. Just as oil companies formed the Aleyska consortium to seek oil on

Alaska's North Slope in the early 1970s, the semiconductor industry is grouping together to

August 24, 1992

©1992 Dataquest Incorporated

MVRY-SE6-DP-g203

Memories Worldwide

create advanced process knowledge and to share costs.

Global Technology

While this present agreement has a partner in each of the world's markets, in fact the global element of this venture is weak compared to the finance and risk elements. StiU, IBM cements its position as a "European" electronics company and lends a hand to Europe's leading supplier of commodity memory chips and arguably its leading semiconductor technology house.

Becavise there are no manufacturing or marketing plans as a part of the compact, however, most of the trade issues are sidestepped or avoided and knowledge will flow freely through the porous borders of the United States, Japan, and Europe. movers are still private companies pursuing what they perceive as their own best interests.

The U,S. govemirvenf s subsidy of Sematedh^ at

$200 million per year, is about 5 percent of the

U.S. industry's R&D budget, and is comparable to tihis single program. JESSI is of similar scope and magn&ude. TTiis undertaking will be financed, at least superficially, by the industry participants themselves.

There are probably lessons here, as weU, for managing multiparty development undertakings that require substantial investment and provide returns to each participating party. Deciding the quid pro quo and the research program among disparate parties with similar interests is a formidable problem. How can one be sure that the benefits derived by each party are commensurate with its contribution? There is every incentive to minimize financial and hiunan resource inputs and maximize technology outcomes.

Perhaps the more important global aspects of this venture may be any difficulties that arise from conducting research in three widely separated locations. Although research tasks can be well defined and divided up, there is certainly a high value in the incessant communication taking place among the research staff. Whether we like it or not, geographic separation has its high overhead costs and inefficiencies.

Increased Pressure on Other DRAM Makers to Do Likewise

Another likely outcome of this annoimced venture among Toshiba, Siemens, and IBM wiU be forcing other aspirants to the 0.25pin or 256Mb

DRAM realm to find similar means of remaining cost-competitive later in the decade. No independent, go-it-alone DRAM producer can hope to be competitive in future generations while spending three times as much as other participants to develop the process. To date, we have seen three 64Mb/0.35nin deals (NEC/ATT,

IBM/Siemens, and Hitachi/Texas Instruments).

Already the 256Mb development costs are getting steep enough that they need to be shared.

NEC annoimced earlier in the year that it woiild spend $150 miUion for development of the

256Mb DRAM in 1992.

Who Is Driving the Industry?

Such a transnational arrangement serves to refocus the industry's attention on the fact that, despite virtually imiversal government participation in the semiconductor industry, the prime

Common Process—Core and Differentiators

The formidable costs faced by companies for development in the subquarter-micron range have been rather cleanly divided into a "common process pool" that has appeared to pose the most significant barrier to 21st-centuiy industry development, and "other," which includes manufacturing costs, marketing, non-

DRAM product definition, and specialty process development costs. Process development costs are where the biggest dollars are spent, but they do not provide proportionate profits or valueadded in today's marketplace.

From another view, just as TI tries to feed as much revenue as possible off a common set of piocess tools, equipment, and recipes (both independentiy and with the Hitachi joint ventures), these three companies seek Hie same broad amortization across a massive range of product: not only their own product lines (which in 1992 were about $10 billion), but also to others through resale of the technology allowed under the terms of the agreement.

By dramatically reducing the costs of forward piocess development, "process" pushed back ttie hierarchy of differentiating capabilities, because these three companies, and likely others latec can build off the same core capabilities. Many observers of the industry have critidzed the intense focus of the industry on manufacturing

MMW-SEG-DP-9203 ©1992 Dataquest IncoiporalBd

August 24, 1992

10 Memories Worldwide

and money, on "process," and on the fine-line capabilities best exemplified by Japanese progress in MOS memories during the 1980s, instead of on looking at where the performanceenhancing opportunities are in silicon-consuming systems.

In Future Issues

Look for articles on the following topics in the next Memories Worldwide Dataquest Perspective:

u ECL I / O SRAMs

Today, "process" appears to be becoming an enabling capability, necessary but not sufficient for semiconductor companies' profitability. Value to the customer and sustainable market advantage are increasingly given by proprietary architecttires, products well-defined to fit applications, and software. One can read in this agreement then that, provided this 0.25|im capabiUty is made available to parties outside the three principals, a tilt toward design-intensive

U.S. companies and a-way from market domination through process excellence will lesvilt. It reduces, though hardly eliminates, the advantages achievable through sheer financial resource.

• WideDRAJVIs

Dataquest Perspective: A New World Order?

This megaventure quite likely is the largest and most recent fixture in the emerging semiconductor industry structure. In this view, basic technologies will be developed in common, widely shared, and differentiated by each individual practitioner. Fully 20 separate 0.8- and O.Zjxm processes were developed for 4Mb DRAM generation. For future generations and the 0.25]m\ level, as a result of this common development pact, we may see just four to five basic processes offered by groups of collaborators, reducing redundancy and unnecessary process development and freeing industry resources to concentrate on the highest value-added (and, for the maker, profitable) chip design issues.

Process development may be even further separated from production and design in the future, just as the equipment industry, formerly a part of the semiconductor industry, has evolved into a separate standalone industry offering standard products to all device manufacturers.

By Lane Mason

August 24, 1992

©1992 Dataquest Incorporated

MJSfi-SBrOP^SOi

12 Memories Woridwide

For More Information

On the topics in this issue Lane Mason, Director (408) 437-8120

About online access (408) 437-8576

About upcoming Dataquest conferences (408) 437-8245

About your siibscription or other Dataquest publications (408) 437-8285

Via fax request (408) 437-0292

The content of this report represents our interpretation and analysis of information generally available to the public or rdeased by responsible individuals in the subject companies, but is not guaranteed as to accuracy or completeness. It does not contain material provided to us in confidence by our clients.

Individual companies reported on and anafyzed by Dataquest may be clients of this and/or other Dataquest services. This infonnatian is not furnished in connection with a sale or offer to sell securities or in coimection with the solicitation of an offer to buy securities. This firm and its parent and/or theii officers, stockholders, or members of their feanilies may, from time to time, have a long or short position in the securities mentioned and may sdl or buy such securities.

August 24, 1992 ©1992 Dataquest Incoipoiated MMRY-SEG-DP4203

Dataquest Perspective

Memories Worldwide ^ Not RemoVe

April 6,1992

MMRY-SEG-DP-9202

In This Issue...

Market Analysis

Market Analysis

PwcessorSpeciHc SRAMs: A Maiket

Processor-Specific SRAMs: A Market Slow to

Slow to Catch On

Catcli On

Most medium- to high-performance com.puting systems are now equipped with cache memories.

R'ocessors' clocks are increasing to speeds that make cache designs extremely difficult. Cachespecific SRAMs being offered by some fast SRAM manufacturers support performance that might otherwise be impossible to achieve. In this artide, we examine these parts by the processor types they support

By Jim Handy Page 1

These are interesting times for static RAM vendors and users alike. Suddenly, nearly all medium- to high-performance geneial-pvupose computing systems are equipped with cache memories. Fast (25ns) SRAMs have recently gained such broad acceptance for use in cache memories that a flood of SRAM vendors has joined the fast SRAM market in anticipation of gaining market share in a high-margin business.

Augmented by the current recession, this rush has had the opposite effect and has caused an overabundance of fast SRAMs, driving compres-

Product Analysis

sion of the average selling prices (ASPs) to the point that the premium for a 25ns part is not

Tile Future of the SfiAM Market

This artide ecamines the future growth of the

SRAM market, focusing on high-speed SRAMs, that significant in comparison with the price of a 100ns part. which are expected to be a major growth segment

Although this shoiild be wonderful news for the to match faster MPUs.

By AJdra Minamikawa Page 12

system designer, who needs fast SRAMs to construct a cache memory, it falls short of the mark.

Inquiry Summary

Dataquesf s Semiconductor Memories inquiry summary is designed to inform dients of commonly asked questions and Dataquesf s respective answers. No confidential information provided by our dients is induded in this materiaL The information contained in this publication is believed to be reliable, but it cannot be guaranteed to be correct or complete.

Processors' clocks and bus inter&ces are now being pushed into speed ranges that make such cache designs extremely difficult. Certain timing specifications vary with processor dock frequency increases, while others do not (see Kgure 1). This acts as a lever between the processor and the SRAM used to implement the cache, forcing the SRAM used to triple or quadruple in speed for every doubling in speed of the processor clock.

• Is flash replacing EPROM?

• What is the difference between flash EPROMs and flash EEPROMs?

Some designers are taking advantage of new cache-specific SRAMs being offered by some fast SRAM manufacturers and are able to design

• Who offers what technology? for performance that might otherwise be impossible to achieve. These parts are basic static

RAMs vsdth some new twists. Features include

• Name a couple of good applications for flash memory. bank switching, wide data paths, synchronous interfaces, and burst counters. We will examine

By Nicolas Samaras Page 14

these new SRAMs in this article.

DataQuest

n n a company of

MMMM The Dun &Bradstitct Corporation

Dataquest is a registered tademark of A.C. Nielsen Company.

File behind the Perspectives tab inside the binder labeled

Memories Worldvnde

©1992 Dataquest Incoipoiated, E^roduction Piohitiited 0012928

Memories Worldwide

Figure 1

.-:^i"4-4

4 '-^

SRAM Access Time Compression for Faster CPU C\(Kk Rates ns

140

25 33

CPU Ctock (MHz)

G2a00DS1

Source: Dataquest (April 1992)

One diffictdty in producing a report like this one is the question of where to cut off the realm of the processor-specific SRAM, or more precisely, which parts should not be counted as processorspecific SRAMs but as integrated cache chips.

Tlie definition we will use here is as follows: If there is a significant lever causing the manufacturer of the cache controller to disallow other manufacturers from participating in the SRAM business that should go with that cache controller, then the cache data RAM device wiU be counted as a part of a cache controller chip set rather than as a processor-specific SRAM. Therefore, we do not count the Intel 82490, which is tightly coupled to the 82495 controller, nor the

MOSel MS443, which requires that company's

MS441 cache controller to operate. The Intel

82395, the Motorola XC88200, and other all-inone cache and controller chips also will not be coxxnted. Likewise, we do not include parts that are general-purpose SRAMs but have been screened to better match the specifications of a certain microprocessor, because the success of these parts depends entirely upon the lack of availability of fester parts—a short-term situation at best.

The survey of parts in this article will describe processor-specific SRAMs in terms of their markets or the microprocessors driving the sales of each SRAM. The parts are also categorized by

CPU and vendor in Table 1.

Parts, by Processor Supported

The 1386

The most widely available processor-specific

SRAM to date, and quite possibly the oldest, is the one designed to support Intel's 82385 cache controller for iiie 1386 microprocessor.

Originally conceived by Vitelic (now MOSel/

"N^tdic), the part is unusual in three areas.

First, the data path is 16 bits wide, with separate chip enable inputs for each byte of the

16-bit word. Second, it incorporates a flowthrough latch on the address input pins. Third, the device is broken into two banks, each of which is 4K words deep and has its own write enable and output enable pins (see Figure 2). A special "mode" pin allows the two banks to be stacked as one, with an additional address pin (offered by Micron as either latched or unlatched) to select which of the two to use, rather than the two sets of enable pins. This address pin is about twice as fast as the other two. Tlie device is referred to by three different organizations: 8Kxl6, 2x4Kxl6, and 4Kxl6x2. We wiU use the name 2x4Kxl6 in this article. The device is available only in a 52-pin PLCC package, and 18-bit-wide

April 6, 1992

©1992 Dataquest incorporated

MMRY-SEG-DP-9202

o

-p

Table 1

Processor-Specific SRAM Offerings

Processor

Intel 1386

Intel i486

Motorola 68040

Sun SPARC

Sun Viking

MIPSR3000

MIPS R4000

Moto DSF56001

Organization Features

2Kxl6x2 For C&T 307

4Kxl6x2

AO-All Latched

A0-A12 Latched

4Kxl8x2

AO-All Latched

A0.A12 Utched

8Kxl6x2

4Kxl6

Address Latd)

Address Latch

4Kxl8x2

4Kxl8x2

32Kx9

64Kx9

128Kx9

16Kxl6

32Kx9

16Kxl6

For Intel 82485

Burst Cnt/ST \lftite

Burst Cnt/ST WHte

Burst Cnt/ST Vftite

Burst Cnt/ST Vftite

Burst Cnt/ST Write

128Kx8

128Kx9

3ZKx9

8Kx20x2

16Kx9(10)x2

4Kxl6

8Kxl5(16)x2

64Kx4

256Kx4

32Kx8

8Kx20

2 Address Latches

2 Address Latches

Address Latch

2 Address Latches

I Fast Address Input

Synchronous

1 Fast Address Input

AT&T

ATr7C183

ATr7C184

Burst Cnt/ST Write

Addr/Data Latches

Addr/Data/CS Latch

ATT7C157

Self-Tuned

Self-Tuned

Self-Timed

Cypress Hitachi IDT

CY7C183

CY7C184

CY7B173

CY7B155

CY7B174

CY7C157

HM62A168

HM62A188

HM62A932

HM62A8128

HM62A9128

HM62A2016

IDT71586

IDT71589

IDT71B229

IDT71586

Logic

Devices

L7C157

iirmo

Micron

Mr56C0816

Mr51C3816

MT56C0818

MT51C3818

M

MS

MT51C2818

MT58C1289

a.

sr*

• &

^

1

%

1

S'

R

Table 1 (Continued)

Processor-Specific SRAM Offerings

Fiocessor

Intd t386

Intel i486

Motorola 68040

Sun SPARC

Sun Viking

MIPSR3000

MIPS R4000

Mote DSP56001

Organizatteft PeahircB

2Kxl6x2 For C&T 307

4Kxl6x2

AO-AU Latched

4Kxl8x2

A0-A12 Utched

AO-All Latched

A0'A12 Latched

8Kxl6x2

4Kxl6

4Kxl8x2

4Kxl8x2

32Kx9

64Kx9

128Kx9

16Kxl6

32Kx9

16Kxl6

Address Latch

Address Latch

For Intel 82485

Burst Cnt/ST Wite

Burst Cnt/ST Write

Burst Cnt/ST Wite

Burst Cnt/ST Write

Burst Cnt/ST VWte

Burst Cnt/ST Wite

Addr/I>ata Latches

Addr/Dafa/CS Latch

Self-Timed

128Kx8

128Kx9

32Kx9

8Kx20x2

16Kx9(10)x2

Self-Timed

Setf-Timed

2 Address Latches

i y a i S n i t ^

PSM44259

PSM44029

PSM44659

PSM44039

4Kxl6

8Kxl5(16)x2

64Kx4

256Kx4

32Kx9

8Kx20

Source: Dataquest (April 1992)

2 Address Latches

Address Latch

2 Address Latches

1 Fast Addieas Input

PSM44298

Syndtfonous

1 I ^ t Address Input

PSM44028

P4C214

; p 4 C ^

P4C215(6)

{Haatatif Quality Samaimg S G S '

Son

PI2C2589

PI2C2157

PI2C2158

QS88160

QS88163

QS88180

QS88183

QS88181

QS83291

QS83283

KM78C80

KM78B86 MK62486

KM78B40 MK62940

KM741006

CXK

CX

CX

=5

(n

5!

%

•P

Memories Worldwide

Figure 2

2x4Kx16 SRAM Block Diagram

Mode

A 1 2 ^

D0-D7/Dp0

CCCCOOQQO-

I/O Buffer

D8-D15/Dp1

IA3 Buffer

BSO 6 -

• ^ i " 9 -

Soutx^e: Toshiba Corporation Gaoooosz versions in the same package with a similar pinout are offered by some companies.

A slower SRAM process can be made to appear faster to the CPU by integrating the address latch onto the chip. Internal bank switching through the separate output enable pins also makes the device appear to operate fester than commodity SRAMs. The beauty in this part, though, is not tiiat it solves speed problems so much as the board space it saves.

Intel recommends the use of either of two

SRAM configurations with its 82385 cache controller. four 8Kx8s with two octal address latches, or sixteen 4Kx4s with four tristatable octal buffers, two octal address latches, and an

OR gate (27 chips total). The lower chip cotint solution offers slightly lower performance than does the high chip count version, because the two versions support direct-mapped and twoway set-associative cache policies, respectively.

Nobody chose to use the device in its directmapped mode, so the high chip coimt alternative was the design of choice. By providing the unique 2x4Kxl6 architecture, Vitehc was able to reduce designs from 27 ICs to simply

2—a considerable savings in board space.

The part also fits ideally with Chips &

Technologies Inc's 307 cache controller, which operates in about the same manner as the

Intel 82385. MOSel designed a specialty SRAM to support the Chips & Technologies cache controller, which ironically was nearly the

MMRY-SEG-DP-9202

©1992 Dataquest Incorporated

April 6,1992

6

Memories Woridwide same as the VlteUc part, but half as deep.

MOSel's MS82C308 works with the Chips cache controller, but is too small to be used with the Intel controller. MOSel made few inroads with its part, yet it is stiU available.

Compaq Computer Corporation first circulated the Vitelic specification in 1987, in order to accumvdate alternate sources for the part.

Compaq was the first PC manufacturer to use the Icitel cache controller, and aU other PC manufacturers, including IBM, were expected to follow suit. The prospective business looked astounding. Sales in 1989 were expected to exceed 4 million units, at an ASP of about

U.S.$10. At its peak, about 13 manufacturers had agreed to manufactiue the '^^telic SRAM.

Then the tables turned. level of purchases. Some of these purchases continue today.

Currentiy, the 2x4Kxl6 cache RAM is offered for sale only by Cypress Semiconductor,

Micron Technology, Samsung Electronics, Sony, and Toshiba. Of tiiese manufacturers, only

Micron and Toshiba ship appreciable volume.

Dataquest estimates that the 1991 worldwide market for the 2x4Kxl6 was just over 1 million units. Shipments are in dedine (see

Figure 3). Today^s ASP for the part is about

U.S.$8.50 for the 25ns version, about 85 percent of the price of four 8Kx8s (as would be used in the IBM cache), and about 30 percent more costiy than a 25ns 32Kx8 commodity

SRAM, which is twice as dense.

First, while the alternate sources and Compaq were changing the specification in such a manner as to force Vitelic into a redesign,

Integrated Device Technology beat the other vendors to the market with a four-chip solution, a simple 4Kxl6 latched SRAM. This product was available early, was widely merchandised, and stiU looked a lot better than the 27-part alternative, so it ate significantiy into the market.

SPARC

The second most widely available processorspecific SRAM is a 16Kxl6 part from Cypress

Semiconductor, Logic Devices, AT&T, and

Pioneer Semiconductor, all of which introduced versions in the order listed. The device includes registers on the data I / O and on the address and write enable inputs (see Figure 4).

On the rising edge of the clock input, tiie address is captured in the address register; on the falling edge, the data input and write enable inputs are sampled. An output hold register maintains output data after a new address has been clocked into the address register.

This is the only s5nichronous SRAM available with this sort of timing.

Next, the awaited IBM PC based around

Intel's cache controller stirprised everybody by not using the controller according to Intel's recommendations. Rather, it doubled the cache size by performing some unobvious tricks with the address pins. Suddenly, everybody else was forced into imitating the IBM design in order to offer a larger cache also. A strange twist is that the IBM approach also reduced the chip cotmt, not by vising the Vitelic part, but by using eight 8Kx8s, two octal address latches, and a small bit of random logic, to consume sUghtiy more than ten devices, aU of them commodily products, and all of which came in much snialler packages than the

52-pin PLCC. Meanwhile, the first deliveries of the Chips & Technologies cache controller were slipping farther and farther away, and the product was losing its design wins.

The device was made in response to a specification for a 32Kx8 sjmchronous SRAM circulated to several SRAM vendors by Sun Microsystems in 1987. The unusual latch configuration is an ideal fit for the cache/memory management tmit chips (CMMUs) offered by

Cypress, Fujitsu, and LSI Logic. Surprisingly enough. Cypress loses to its competitors some

SRAM sockets in boards in which its own

CMMU and integer unit are used. Of course,

Fujitsu recentiy annovmced that it would build workstations aroimd Cypress's lU and CMMU, so we shouldn't be surprised, should we? The world is sometimes a strange place.

Suddenly, the prospective market for the

•\^telic part nearly disappeared. Compaq had signed contracts witii the first three manufacturers to commit to make the part, so it was unable to back out of a certain

What makes the 16Kxl6 part salable, and what drives the market? First, all of Sun's systems use this SRAM to implement a 64KB cache. Every system Sxin offers has the same

y\pril 6, 1992 ©1992 Dataquest Ixorporated

MMRY-SEG-DP-9202

Memories Woridwide

Figure 3

Worldwide Sales of 2x4Kx16 SRAIVIs

G2000053

Source: Datquest (April 1992)

cache size, so Sun uses quite a few of the

16Kxl6 cache SRAMs. Dataquest estimates

1991 worldwide SPARC sales of 291,000 tmits

(see Figure 5), mainly from LSI Logic, Fujitsu, and Cypress, each of which had about a third of the market. Stm has taken harsh measures to ensure that clone manufacturers have a hard time attaining good sales channels, and the effect has been to reduce the number of

16Kxl6 unit shipments to an amoimt almost identical to Sun's consumption. Dataquest estimates worldwide 1991 unit sales for the

SPARC-compatible 16Kxl6 to have been about

300,000 units. ASPs are about U.S.$10 for a

25ns part, down from about U.S.$30 a year ago. We expect sales to ramp slightly in 1992, then to taper off in 1993 as Sim converts designs to the new Viking processor.

The i486

More attention has been focused on processorspecific SRAMs for the Intel 486 than for any other processor, and as a result there is more variety in this ntarket than in any other.

Table 1 shows six different configurations designed for 486 applications. Sales of

486-spedfic SRAMs are expected to grow considerably from 1992 through 1994, but success will probably be Umited to one or two t3rpes of devices. StUl, the market for these chips is difficult to tmderstand. Many of the

486-specific parts have nothing in common with each other.

The majority of the 486-specific SRAMs offer either or both of two features: bursting and self-timed write cycles. Bursting SRAMs help the 486 to refill its internal cadie lines using the fastest refill mechanism available on the

486 processor, the bvirst read cycle. In a burst read cycle, the processor outputs an address, and the memory can respond by sending the four words within the same general location, each on successive processor clock cycles. This means that the processor can read as many as four words of data every five clock cycles, a significant improvement over the 386's maxim u m rate of one word every two cycles. Burst refills require the use of address generation outside of the processor, and manufacturers of bursting SRAMs have put this address generation logic into the SRAM chip itself. Burst count sequences are different for different processors, so a chip with a 486 burst address generator will not work optimally with other processors.

Self-timed write cycles are a simple way to conquer the problems of generating clean write cycles. A clean write cycle is nearty impossible to generate at high processor frequencies, especially if the processor has a sjmchronous

MMRY-SEG-DP-9202

©1992 Dataquest Incoiporated

April 6,1992

8

Figure 4

16Kx16 SRAM Block Diagram

Memories Worldwide

S. Data-in Latch

(7:0)

WEI —

— C

Reg

d

V

i

1

WEO

—c

Reg

t>

1 '

Self-

Timed

Write

Reg

-4y

t T

. Output

Hold

Delay

n r 1 _

S Data-out Latch

(7:0) s^ Data-Out Latch

^ (15:8)

7

1

L T T

i

1

T T

I/O0-I/O7 1/08-1/015

GSOOOOM Source: A T & T Semiconductor Inc.

Array

16Kx8

Self-

Timed

Write i

S Data-in Latch

(15.8)

Array

16Kx8 interface, as does the 486. The problem is barely surmoimtable if the processor's timing is forgiving enough, but this is not the case with the 486.

The most widespread 486 support RAM comes in a 32Kx9 organization, which is supplied in two competing pinouts. This sorry state of affairs came about after Compaq circulated the specification for Integrated Device Technology's

32-pin part to attain wide second-soiircing, but failed to tell anybody that TUT was close to sampling the product. A division came about when several sources had already rationalized that a 44-pin package was imperative to the manufacture of their versions of the part. The camp is divided into the 32-pin contingent

(IDT, Pioneer, and Quality), and the 44-pin contingent (Motorola, Cypress, SGS, Hitachi,

Paradigm, Samstmg, and others later). The stances grew firmer when the 44-pin device was standardized by ihe Joint Electronics

Device Engineering Cotmdl (JEDEC) in spite of a patent pending to IDT. This should be interesting to watch in a few years. To date, sales of ttie part are ramping sharply, with the bulk of the market controlled by Motorola and

IDT. Unit shipments in 1992 should exceed

1 million units worldwide, at an ASP just imder U.S.$20, which compares very favorably against the U.S.$6 ASP now seen for fast

32Kx8 commodity SRAMs.

Other 486 SRAMs featuring a burst counter and self-timed write include a 2x4Kxl8 organization to be offered by Quality Semiconductor,

April 6, 1992 ©1992 Dataquest Incorporated

MVIRY-SEG-DP-9202

Memories Woridwide

Figure 5

SPARC Processor Market Share by Units

1990

BIT 3.1%

Wettek 5.2%

1991

BIT 0.3%

9

Total = 160,000 Units

Source: Dataquest (April 1992)

Total = 291,000 Units

GS000055

a 16Kxl6 from Cypress, and a 128Kx9 proposed by Paradigm. Not one of these is altemate-sourced, so Dataquest does not expect them to become well accepted in the market.

Another 2x4Kxl8 without a burst counter is sole-souTced by Micron to Intel (quite a coup for Micron), and is excltisively used in Intel's

82485 cache module (also known as the TurboCache and the C6). Volumes of this SRAM are reported to be healthy, despite the fact that the 82485 was a year and a half behind schediile once it finally shipped, and as a result was designed out of ttie majority of its original design wins. (Even Intel's systems group is rumored to be attempting a designout of the 82485 for cost reasons.) Still, of all the processor-specific SRAMs being offered in support of the 486, the two 32Kx9 organizations are expected to lead the market by a wide margin.

So, will every 486 system use two or more of the 32Kx9? No! Although secondary caches are viewed as necessary difiFerentiators for 486based systems, their performance is seldom an issue. OPli, the leading supplier of 486 chip sets, has built its strength upon a nonoptimized cache architecture that incurs wait states on all cache cycles. The success of this cache controller has proven that the majority of PC consumers buy cache by size, not by performance. Bursting 32Kx9s are a relatively costly means of obtaining the highest performance from a 486, so the designs tising tiiis part will necessarily be those aimed at the more sophisticated PC buyer—the one who buys on benchmarks. Dataquest expects high performance SRAMs to be sold to only about

5 percent of the overall 486 market.

Note that the features fotmd on the 486specific 32Kx9 are also put to good use in systems based on other processors, despite the feet that the burst cotmter is optimized for the

486. With a little care, good results can be obtained through the tise of these parts in 860,

960, 68030, 68040, and even high-speed

386-based systems. This feet is not as important as it may seem. AH of these processors except the 386 are mainly used in closed systems. Designers of closed systems can avoid the need to use costly external caches when performance increases can be realized through redesigning the bus or the software. Such systems ^ o tend not to use cache as a buzzword, and it is not as much of a differentiator

MMRY-SEG-DP-9202

©1992 Dataquest Incorporated

April 6,1992

10

Memories Woridwide as is outright performance. These applications will hardly make a difference to the sales of the 32Kx9 bursting self-timed SRAM. We also do not expect 386 systems to widely use this part, because 386-based desktop system design is viewed by system designers as a mature art, and they are not exploring new horizons. offered by NEC. To make matters confusing,

IDT now offers a 16Kx9x2 in a 32-lead,

300-mil SOJ package, a package that allows the design of far smaller boards, but the x9 organization requires the designer to play certain unobvious tricks to keep the package coimt down to six devices.

ne 68040

A different version of the 32Kx9 bursting selftimed SRAM in a 44-pin PLCC has had its count sequence optimized for use with the

68040 processor. The part is currently available from Motorola and Cypress, soon to be followed by Paradigm, SGS, Samsung, and possibly others. As just mentioned, few 68040 system designers choose to augment the 68040's

8KB on-chip primary cache with a costly secondary cache, so the market for this chip is nearly nonexistent.

We do not expect to see any improvement in the market for this chip over the Ufe of the

68040.

Other devices are also touted as R3000 support RAMs, even though they were designed with other applications in mind. The IDT71586

4Kxl6, which was designed for 386 applications, is being promoted by IDT to be a reasonable R3000 cache support chip, and several designs now use Motorola's MCM62990

16Kxl6 general-pvupose synchronous SRAM.

Toshiba's and NEC's 15ns 64Kxl6s are also popular in both R3000 and R4000 applications.

The Viking

Bolstered by the sales of the SPARC 16Kxl6, several SRAM vendors hope to make a pretty penny on the 128Kx9 support chip promoted by Sim to support its Vildng processor. This chip is a very standard synchronous architecture, with the only difference being that common I/O is tised, so the part has a dead cycle when moving from a write cycle to a read cycle.

MIPS R3000

MIPS Computer has recently reversed a prior stance in which it claimed that there was no reason to use anything but commodity SRAMs in an R3000 system. As a result, the company has succeeded in generating interest in processor-specific SRAMs to cater to its R3000 architecture. MIPS' reversal probably can be attributed to the fact that current versions of the R3000 can operate at clock frequencies considerably faster than those originally anticipated by the processor's designers. The

RSOOO's SRAM interface causes design difficulties at clock speeds higher than 20 MHz. The

R3000 requires interleaved banks of latchedaddress SRAM at widths of up to 60 bits per bank.

Rumors are that Sun promised guaranteed minimum-volume contracts to the first three vendors to commit to manvifactiue the part.

The 'Wldng-spedfic SRAM is now sampling from Sony and Paradigm, and is being advertised by Micron. Dataquest expects many sources to follow. The product's simplicity lends its use in other applications, so this architecture could take off in a number of non-Wdng applications.

Although MIPS originally proposed an

8Kx20x2 organization, to satisfy the two-bank scheme, and to provide 64KB of combined instruction and data cache within three 68-pin

PLCC packages, semiconductor manufacturers responded to inputs from system designers to implement a six-package version that would offer twice the size of cache, or a total of

128KB. Manufacturers currently shipping the

8Kx20x2 are NEC and Hitachi The deeper part, a 16Kxl0x2 organization, is now only

MPS R4000

AU eyes are watching the ACE initiative, because its success or failure will determine the health of the SRAMs used to support the

R4000 processor. Intel's mere presence in the

ACE consortium puts the fate of the R4000 into question.

MIPS has taken a cautious path again with the R4000. This processor contains a small on-chip primary cache, as well as the control logic to support a much larger off-chip secondary cache. This cache control logic can

April 6,1992

©1992 Dataquest Incoiporated

MMRY-SEG-DP-9202

Memories Woridwide

11 support caches of varying sizes manufactured from industry-standard asynchronous SRAMs of various speeds. The cache configuration is communicated from a PAL or a PROM into the CPU during the system reset.

A hook was placed into the R4000's reset vectors to support a fester configuration of cache, in which one or two address pins are twice as fast as the SRAM's overall access time. A 15ns

SRAM might have a single address input that exhibits an address access time of 7ns. Several

SRAM vendors have expressed an intent to manufacture this part, with the consensus being that 64Kx4 should be the appropriate size and organization. A general agreement was reached to accelerate the address on pin

11 of this device.

Dataquest Perspective

Processor-specific SRAMs have so far been relatively slow to catch on. This appears to stem from a reluctance on the part of system designers to take advantage of these parts owing to worries that the manufacturers will not be as price competitive as they would be with generic asynchronous SRAMs. This has been augmented by the fact that processor-specific SRAMs sometimes miss the target density of their end applications. Examples are the 2x4Kxl6 for the

386 and the 8Kx20x2 for the R3000.

Even the highest-volume processor-specific

SRAMs do not sell well. The 2x4Kx32, the bestseller of all processor-specific SRAMs, moved about 1 million units in 1991, and runner-up

16Kxl6 only sold 300,000 units. These are not major volumes. Although volumes should increase significantly with the availability of high-speed processors, do not look for these products to displace any important volumes of standard SRAMs. It will probably be late 1992 before sales of processor-specific SRAMs reach an annualized sales rate of well over $10 million. With this in mind, it is not surprising that these products are of decided interest to smaller

"boutique" SRAM manufacturers, whose bottom line can be significantly improved through the addition of a million-dollar product. These manufacturers also aim for products whose

ASPs stay high, affording better margins than would commodity products.

Meanwhile, one of the largest users of the highest-speed version of the R3000—and MIPS' new parent company—^Silicon Graphics Inc.

(SGI), proposed an altogether new scheme, wherein it plans to use a fast synchronous

256Kx4 organization. It floated tiie specification about a year ago, and has gotten several responses. The part will soon be provided by

Sony, Paradigm, and Samsung, but appears not to be sampling yet.

One unfortimate aspect of the R4000 is that the only version to support external cache is the one in the extraordinarily high pin-coimt package. The R4000 in the 179-pin package does not support external cache. The part that does support external cache comes in a hefty

447-pin package, which certainly will factor into design decisions, especially because no plastic package is currently being promised.

Which package will be used by the majority of ACE systems? Time wiU tell, but Dataquest favors the less expensive alternative, even if the system designer will be forced into using an external cache controller.

Threats are also being heard from another front.

DRAM manufacturers are eyeing the ASPs of processor-specific SRAMs and are trying to figure ways to divert that revenue into their own pockets. From such thinking come Mitsubishi's and Ramtron's "cached DRAM" approaches, or

Rajnbus's special 500-MBps proprietary DRAM interface. Meanwhile, expect on-chip caches to become bigger and better, reducing the performance advantages to be gained from the addition of external caches.

Another problem is the need to educate the busy system designer on the need for these products.

First you need to get their attention, and then you need to know exactiy what to say to sell ihe part. Few SRAM sales organizations are structured this way. SRAMs are commodity products, and educational sales are not commodity approaches. This type of sale is best approached by the manufacturer of the processor, not the manufacturer of the SRAM.

Semiconductor vendors selling processor-specific

SRAMs have a lot of work ahead for a relatively small return. Is it worth it? Some seem to think so, but Dataquest does not expect the world to suddenly welcome these parts with a warm embrace.

By Jim Handy

MMRY-SEG-DP-9202

©1992 Dataquest Incorporated

April 6,1992

12

Memories Worldwide

Product Analysis

The Future of the SRAM Mmket

The static RAM (SRAM) market has grown to its current size because of high-speed access and low power consumption, which has more than compensated for rdatively low density. However, the emergence of high-speed DRAMs and flash memories is eroding the traditional competitive edge for slower SRAMs. It is a little easier to enter the SRAM market than the DRAM market because of facility capacity. DRAM requires veiy advanced technology, and the

DRAM business is very risky. The number of

SRAM suppliers is larger ttian that of DRAM suppliers. Therefore, in order to survive in the

SELAM market, SRAM suppliers compete with one another through competitive prices. Price erosion is the result.

High-speed SRAMs are used mainly for cache and main memories of computers. In particular, cache memories use very fast 64Kb to 256Kb

SRAMs with access times of 5 to 35ns, and they are becoming increasingly important to govern the entire system performance. The 5ns bipolar

SRAMs are generally used in supercomputers and mainframes, while 10 to 35ns versions are used in minicomputers and workstations.

Recentiy, workstations have required 10ns or faster SRAMs because of the increasing use of reduced-instruction-set computing (RISC) CPUs.

Similarly, more and more PCs use cache memories to consume 25 to 35ns 64Kb and 256Kb

SRAMs. Applications are further extended to caches for external storage (on disk), where

100ns parts are used. Cache memories for workstations and PCs often use multibit SRAMs to minimize board space, whereas main memories of supercomputers constime a large amoiint of xl or x4 versions with large capacities.

High-Speed SRAM Market in 1990

Dataquest estimates that the worldwide highspeed SRAM market increased at a compound annual growth rate (CAGR) of 6.8 percent in

1990 to reach approximately $1,076.7 million

(see Figure 1). This is healthy and strong growth compared with the negative growth of the MOS memory market, which was down 20.1 percent from the previous year to $12.5 billion. In fact, only high-speed SRAMs (access time of 70ns and faster) and flash memories recorded growth among MOS memory products in 1990 (see

Table 1).

Higli-Speed SRAM Maricet Trends

Recently, high-speed SRAM demand has been growing rapidly for use as cache memories for

33-MHz or faster complex-instruction-set computing (aSC) and RISC MPUs. On the other hand, profitability has deteriorated recentiy because of competitive pricing by a large nvaaix- ber of vendors. The 33-MHz 386, 486, and RISC

MPUs use caches and 256Kb SRAMs used for

33-MHz SPARC and R3000 require access times of 30ns or fester. In particular, R3000 demands

Figure 1

Worldwide Fast SRAM Forecast

Millions of Dollars

3500

^ Very Fast SRAM

3000

^ Fast SRAM

2500

2000-

1500-

1000 -

500 -

1987

Source: Dataquest (April 1992)

1990

1991 1995

0200X156

April 6, 1992 ©1992 Dataquest Incorporated MMRY-SEG-DP-9202

Memories Worldwide

13

Table 1

Worldwide MOS Memory Market (Millions of Dollars)

Device

DRAM

SRAM (>70ns)

SRAM (<70ns)

EPROM

ROM

EEPROM

Flash

Total

Source: Dataquest (Apiil 1992)

1989

8,968

2,364

1,008

1,808

1,221

318

11

15,698

1990

6,830

1,684

1,077

1,446

1,157

314

35

12,543

Growth Rate (%)

1989-1990

-23.8

-28.8

6.8

-20.0

-5.2

-1.3

218.2

-20.1

SRAMs with access times of twice the frequency—15ns or faster. Demand for highspeed SRAMs also comes from emerging microprocessor imits (MPUs) with dock frequency of 50 MHz or higher. These facts point to rapid expansion of demand for SRAMs with very high speeds. In fact, these SRAMs have boosted titieir share of the high-speed SRAM market from 35 percent in 1987 to 39 percent in

1990; their share is expected to grow to 41 percent in 1991 and then over 57 percent in 1995.

Nevertheless, there are some hxardles to be cleared before these goals are attained. First, delay due to external standard logic ICs would affect system performance significantly in the

^high-speed operating environment at the

50-MHz level. One solution to improve performance is to integrate logic circmts into SRAMs to reduce delay by 2ns to 3ns, which is then allocated to memories. In practice, some SRAM

S5^tems incorporate the address latch circuit between MPU and memories. This solution is designed to implement appUcation-spedfic memories optimized for different types of microprocessors. For instance. Motorola Incorporated and NEC Corporation are marketing cache memories dedicated to R3000, SPARC, or i386.

Customized SRAMs offer large bit width, integrating the address latch circuit and other functions to reduce chip count for cache system com^pared with general-purpose high-speed

SRAMs. Availability is a major problem, however, necessitating the securing of second sources.

Another problem is that the price remains at a relatively high level because of the small ntimber of suppliers. To secure a stable supply, some U.S. MPU manufeicturers are looking for

Japanese SRAM makers—a seemingly mutually beneficial deal.

Although an attempt is being made to incorporate cache into MPUs, it is not technologically feasible in the short run to integrate ttie secondary cache into a single chip^ paitly because of chip size. Instead, the multichip module is receiving increasing attention as a solution to avoid delay due to external manory and data input/output. By mounting the MPU, cache memories, and other devices on a single module, wiring impedance can be minimized and operating frequency in the module can be increased.

On the other hand, even the SRAM with transistor-transistor logic (TTL)-level interface requires 10ns access time or less. To achieve such high speed, while dealing with an accompanying noise problem upgrading firom TTLlevd to emitter-coupled logic (ECL)-level interfece may be required. For this purpose, the

BiCMOS process must be suitable for both TTL and ECL levels and there must be commercialization of the 3.3V system, which allows speed to increase while maintaining compatibility with

TTL. The ECL process is also a potential solution for implementation of high-speed versions, but high cost and power consumption are likely to limit its application to some very high speed products. FLn^y, improvement is expected in packaging. Compared with the conventional package in whidi the power sotirce and GND pins are arranged at the comers, very high speed SRAMs will have them at the center of the package in order to minicnize impedance in lead frame.

MMRY-SEG-DP-9202

©1992 Dataquest incorporated

April 6,1992

14

Memories Woridwide

Dataquest Perspective

As the increase in processing speed of MPUs leads to the increase in operating speed of workstations and PCs, Dataquest expects cache memories to play an increasingly important role.

While the Wgh-speed SRAM market is encroached upon by high-speed DRAMs and the slow SRAM market faces a threat from flash memories, we expect very high speed SRAMs to become a growth center in the SRAM market.

Clearly, high speed as well as low power consumption are keys to the future prosperity of the

SRAM market. At the same time, Dataquest sees that SRAM maniifacturers must survive through the development of cache memories optimized for different MPUs—jointly with capable microprocessor makers—^to build u p a rdiable supply capability. In this sense, ttie SRAM market is about to enter an industry-wide restructuring period characterized by strategic alliances.

By Akira Minamikazoa

Inquiry Summary

Semiconductor Memories Inquiry

Higliiiglits

Q: Is flash replacing EPROM?

A: Even though the majority of flash memory

ICs conform to Joint Electronics Device

Engineering Council (JEDEC) standard pinouts, making them pin-for-pin compatible with

EPROM devices, flash memories are not replacing EPROMs directly. The primary reason is, of course, cost. The 1Mb flash EPROM costs about

$10, whereas the 1Mb EPROM costs $4. Even if the significant cost differential is ignored, replacing the EPROM in an existing board design with a flash memory does not make sense because the systems usually are not designed to exploit flash's main advantage over EPROM: the ability to alter the stored data without removing the device from its socket/board. As a result flash is used mainly in new designs, where the electrical rewriteability can be designed in. The in-socket reprogrammability that flash offers may more than offset at times its cost disadvantage.

Automotive applications offer plenty of examples. If there is a need to change the data stored in an EPROM, a board along with the subassembly must be physically removed from the automobile; then the EPROM must be physically removed and exchanged. That is a very expensive option. If flash memory were used, the data could be easily altered by connecting the board to a computer using just a cable.

Q: What is the difference between flash

EPROMs and flash EEPROMs?

A: Simply stated, flash EPROM requires a 12V supply for programming whereas tiie flash

EEPROM requires a 5V supply. Both EPROMand EEPROM-derived flash need only a 5V supply for read operations. Flash EEPROM is derived from the full-featured EEPROM (twotransistor cell). The smallest die size is achieved when only bulk-erase capability is desired (electrically erasing all memory locations). But this makes the flash EEPROM eqtiivalent to a flash

EPROM from a feature standpoint while incurring a die size/cost penalty of the two-transistor cell. As a result, the only advantage that flash

EEPROM with bulk erase capability offers is being a 5V, single-supply device. TTiis in itself would be a significant advantage if flash

EEPROM could be produced at a cost parity with flash EPROM. By subdividing the flash memory into sectors (for example, 4Kb or 16Kb each), a finer granularity is achieved. This is desirable for solid-state disk and memory card applications (at a cost penalty, as the die size increases).

Flash EPROM is derived from basic EPROM technology (one-transistor cell). It offers the smallest die size (for standard JEDEC products) and thus the lowest-cost flash products. The disadvantage is that, like the EPROM, aU data must be erased at the same time QovSk erase). However, unlike EPROM that requires a timeconsvuning off-system UV erase, flash EPROM devices can be quickly erased in-system, electrically.

Q: Who offers what technology?

A. Intel leads the EPROM-derived flash camp that also includes AMD, NEC, Hitachi, Mitsubishi, SGS-Thomson, Exel, Catalyst, and Old.

Toshiba and ATMEL offer EEPROM-derived flash. Intel offers a flash EPROM with hmited sector erase capabihty. The 1Mb 28F001BX is segmented into four sectors: one 8KB, two 4KB, and one 112KB. The 28F001BX is quite popular in BIOS appHcations. Both Hitachi and NEC

April 6,1992

©1992 Dataquest Incorporated

MMRY-SEG-DP-9202

Memories Woiidwide

15

plan to offer 4Mb flash with block erase capabilities. The Hitachi HN28F4001 is divided into 32 blocks of 16KB. NEC's 13PD28F4001 offers a similar organization; the uFD28F4000 is organized as 16K words by 16 blocks. Finally

Toshiba plans to introduce the TC584000, 5Vonly 4MB flash EEPROM with block erase capability (4Kb block size). computer's BIOS. This arrangement allows for easy in-system upgradability of the BIOS code, allowing new features to be integrated into existing systems. The portable PC market seems to be embracing flash technology. It should be noted that tliis market demands low-voltage devices, and in the long run may drive the 5V and 3V flash memory technology. Palmtop PCs wiU use flash for mass storage (solid-state disliO-

Q: Name a couple of good applications for flash memory.

A. Automotive, in engine and transmission management control dectronics. The availability of 12V makes this a good application for the standard flash EFROM products. The fact that this happens to be the least expensive flash comes as an added bonxis in an extremely costconscious industry.

It also should be noted that some flash EEPROM designs are optimizing the overall device die size by sacrificing speed. This is targeted to rigid disk drive applications where speed may be compromised. Here emulation of an existing slow electromechanical system is required (average access time of 10ms to 20ms). The Toshiba

TC584000 flash memory fits this category.

By Nicolas Samaras

In personal computers, flash memory is increasingly used to replace UV EPROMs to store the

MMF(Y-SEG-DP-9202

©1992 Dataquest IncoiporalBd

April 6,1992

16 Memories Woridwide

For More Information

On the topics in this issue Lane Mason, Director (408) 437-8120

About online access (408) 437-8576

About upcoming Dataquest conferences (408) 437-8245

About your subscription or other Dataquest publications (408) 437-8285

Via fax request ~(408) 437-0292

The content of this report represents our interpretation and analysis of infonnation generally available to the public or released by responsible individuals in the subject companies, but is not guaranteed as to accuracy or completeness. It does not contain material provided to us in confidence by our dients.

Individual companies reported on and analyzed by Dataquest may be clients of this and/or other Dataquest services. This information is not furnished in connection with a sale or offer to sell securities or in connection with the solicitation of an offer to buy securities. This firm and its parent and/or their officers, stockholders, or members of their families may, from time to time, have a long or short position in the securities mentioned and may sell or buy such securities.

April 6,1992 ©1992 Dataquest Incorporated MMRY-SEG-DP-9202

Dataquest Perspective F""

Do Not Remove

Memories Worldwide

March 30, 1992

MMRY-SEG-DP-9201

In This Issue...

Market Analysis

Market Analysis

Preliminary 1991 Worldwide MOS

Preliminary 1991 Worldwide MOS Memory Market

Memory Market Share Climates

Share Estimates

Dataquest has com.pleted its preHminaiy 1991

Dataquest has completed its preliminary 1991 MOS memory market share survey analysis. Although

MOS memory market share survey. We mailed a survey questionnaire to more than 150 semiconthe market recovered from a dismal 1990 showing, serious price erosion coupled with lackluster growth in new densities made 1991 anjrthing but a halcyon year. ductor vendors in early November. The respondents provided us with detailed breakouts of their revenue and unit shipments based on a combination of year-to-date data and company-

By Lane Mason, Jim Handy, and Nicolas Samaras Page 1

generated forecasts for the rest of the year. The collected results are published in this article. We

A Year of Transition for DRAMs

The year 1991 was in many ways a transitional one for DRAMs. Korean companies are gaining market share over their Japanese covmterparts and rising production costs from the 1Mb to tiie 4Mb win continue to refine and update the data, and we plan to release our final market share data docvunents on May 31, 1992.

Market Sfiare Highliglits

generation are initiating an increase in anticipated floor costs. Surface-mount packages have taken

The following analysis covers the three areas of over, and several new technologies are now attempting to gain market acceptance.

MOS memory tracked by Dataquest DRAMs,

SRAMs, and nonvolatile memories (EPROMs,

By Lane Mason Page 2

EEPROMs, ROMs, and Flash memories).

SItAM Suppliers Must Run Faster or Fall Behind

Dataquesf s preliminary 1991 market share estimates for static RAM suppliers show little change from 1990. The rankings have barely changed, despite the fact that most manufacturers dramatically increased their unit shipments. Dramatic ASF erosion in the fast SRAM arena has been the culprit.

By Jim Handy Page 6

Nonvolatile Memories: A Year of Bullish Flash Growth

MOS memory grew in 1991 at a relatively moderate 6 percent, in contrast to 1990's disastrous 17 percent decline. Observed in this light, however, MOS memory has made a rather im.pressive recovery, especially considering the fact that the 4Mb DRAM has not taken off as quickly as some had hoped. Toshiba returned its

No. 1 ranking for all MOS memories, shipping more than $1.44 billion. In 1991, Toshiba was followed in order by Hitachi, NEC, Fujitsu, and

Samsimg.

Nonvolatile memory revenue growth in 1991 outpaced that of DRAM and SRAM according to

Dataquesf s preliminaiy estimates. Strong electronic game sales helped ROM shipments. EPROM remained flat as both the automotive and data processing segments were down. Hash memory was the bright new star vrith substantial growth.

By Nicolas Samaras Page 8

Highlights of the 1991 MOS memory market indude the following:

• Overall dollar growth in DRAMs was 5.0 percent; in SRAMs, 5.0 percent; and in nonvolatile memories, 11.7 percent.

DataQuest

n n acompanyor

J i n ThcDun&BiadstrcctCorporadon

Dataquest is a tegistsred tiadematk of A.C. Nielsen Company.

FUe behind the Perspectives tab kiside the binder labeled

Memories 'Worldwide

©1992 Dataquest Incoipoiabsd, Reproduction Prohibited 0012759

-i££]

Memories Worldwide

^ . \ - f - . •'•:

P: ir\y^ r i Q

i * » • ••?

» W

Figure 1

1991 Estimated Worldwide l\/lemory Sales, if/ Type

(Millions of Dollars)

/ static RAM

/ $2,710

A Year of Transition for DRAMs

The year 1991 was in many ways transitional for

DRAMs. Aside from the triennial move from generation to generation, we began to see the long-anticipated emergence of wide DRAMs at the 4Mb level, continued market share gains by

Korean manvifecturers over the stiJl-dominant

Japanese suppliers, and, with the 4Mb ramp-up, the first real impact of the new economics of

DRAM production that promise to raise ultimate floor prices from generation to generation in degrees never seen before.

Dynamic RAM i

I Nonvolatile /

\ $3,296 /

$6,800 §

Prices per bit for 4Mb DRAMs crossed over those of 1Mb parts about midyear, and by yearend, 4Mb DRAMs were generating greater revenue and shipping more bits per quarter.

Source: Dataquest (March 1992)

For the most part, the initial 350-mil 4Mb

DRAM was replaced by the historical standard

300-mil package and DRAM speeds inched d o w n w ^ . The majority of products are now available at 70ns to 80ns. The market for veryhigh-speed optimized designs was fovmd wanting, as MPU speeds made cache systems unavoidable.

• Rapid price erosion decreased the growth of

SRAM sales significantly.

• Micron Technology became a top-10 player in the DRAM market, at No. 8.

• Rankings of the top 10 SRAM and nonvolatile memory manufacturers remained nearly vmchanged.

• Flash memories imderwent the most dramatic change. Increasing unit shipments by

491 percent.

• Japanese vendors took 60.6 percent of the

DRAM market and 71.5 percent of the SRAM market. Japanese market share of nonvolatile memories was far lower at 47.7 percent.

The pie chart in Figure 1 shows the relationship in dollar sales between the DRAM, SRAM, and nonvolatile memory meirket segments.

By Lane Mason

Jim Handy

Nicolas Samaras

By year-end, low pricing in the 1Mb market had encouraged several manufacturers to ramp down production and concentrate entirely on the

4Mb and 16Mb devices.

1991 Market Share Movement

Table 1 shows suppliers' 1991 DRAM revenue.

Despite an estimated 31 percent price-per-bit

(PPB) erosion, the DRAM market managed revenue growth of about 5 percent for the year. Japanese companies, led by No. 1 Toshiba with an estimated $904 million in sales, took 6 of the top

10 places in 1991 DRAM production. No. 2

Samsvmg was the top 1Mb shipper. Micron

Technology, coming off a difficult transition to

1Mb in 1990 that impacted 1990 revenue, showed the top growth rate among the top 10 and had sales estimated at $362 miUion in 1991.

Despite continuing trouble with its 4Mb at several facilities, Texas Instruments was the No. 6 producer, although it dropped about 1 percent in sales to $571 miUion.

As a portent of things to come, Goldstar and

Hjomdai each grew shipments more than

140 percent, but remained out of sight of the top 10, for 1991 at least. Overall, Korean

March 30, 1992 ©1992 Dataquest Incorporated

MMRY-SEG-DP-9201

Memories Woridwide

Table 1

1991 DRAM Revenue (Millions of Dollars)

Company

Toshiba

Samstmg

NEC

Hitachi

Texas Instruments

Fujitsu

Mitsubishi

Micron Technology

Old Semiconductor

Siemens

Motorola

Goldstar

Hyundai

Matsushita

NMB Semiconductor

"V^telic

Sharp

Others

Total

Source: Dataquest (March 1992)

1990 Sales

947

809

669

597

75

131

171

58

72

110

6,475

576

497

503

294

293

306

292

75

1991 Sales

904

900

698

675

571

484

467

362

319

287

264

204

186

116

116

79

62

• 106

6,800 companies garnered virtually all of the aggregate

DRAM market growth, largely at the expense of the more established Japanese suppliers.

Goldstar and H5amdai have become significant enough players to threaten low-end pricing, and in 1991 became bona fide suppliers of 4Mb

DRAMs. Samstmg cemented its position as a leading contender for the DRAM crown, shipping 4Mb DRAMs at a rate within striking distance of market leader Hitachi. Samsvmg's 16Mb parts are said to be on a par with the best in the industry—^no price discounts anjrmore! Samsung will be interesting to watch in 1992 as it challenges for the DRAM lead, as will H5mndai and Goldstar as they upgrade both their products and customer base.

One important advantage that newcomers to the

DRAM market have in 1992 and beyond is the changing channels to end users. A growing fraction of the DRAM business is upgrades for PCs bought at mom-and-pop outlets: Blue-chip accounts, with exacting system requirements and lengthy, expensive qualifications, make up a steadily declining share of DRAM demand.

Add-on memory is an attractive outlet for the price competitive new kids on the block—

Goldstar, Hyundai, and any other late arrivals to the 4Mb race—who want to keep their fabs full and learn the biisiness.

Packaging Trends

The 1Mb generation was the last hurrah for the dual in-line package (DIP), as its share of the business migrated rapidly to SOJ and, it appears, soon to TSOP. DIPs were about 22 percent of 1Mb DRAM unit shipments in 1991, compared with about 4 percent of the 4Mb shipments. ZIPs held steady at about 15 percent of the 4Mb generation. So, after five generations of

DRAMs that went into the DIP, we have gone through two major packaging turnovers in just the past two generations, witii the last yet to fully express itself in the market.

Change(%)

-4.5

11.2

4.3

13.1

-0.9

-2.6

-7.2

23.1

8.9

-6.0

-10.0

172.0

148.0

-11.5

-32.2

36.0

-13.9

-3.6

5.0

MMRY-SEG-DP-9201 ©1992 Dataquest IncoiporatBd Maich 30, 1992

Memories Woridwide

The single in-line memory module (SIMM) market constituted an estimated 40 percent of

DRAM sales at year-end, but was difficult to size because of the prevalence of aftermarket

SIMM packagers. During the year, the SIMM issue was further muddied by the Wang lawsuits, which draw a low royalty wall around

30-pin x9 SIMMs. Manufacturers are watching to see which way the winds blow, pushing the demand over to x36 modules, while begrudgingly paying the 3 to 4 percent royalty to Wang.

As a significant 6:action of the DRAMs in

SIMMs fill in the upgrade memory needs of the

PC and workstation installed base, it looks as though x9 will continue in a major way for

1992. In addition, there appears to be no rush in newer PCs to design expansion slots to accommodate x36 modules, though some are doing so.

Pricing

Pricing for all DRAM densities continued through 1991 in an iminterrupted dedine. There were no significant "spot" shortages in packages, speed grades, or operating modes. Yearend pricing for 1Mb DRAMs was often below

$4, with a bottom of about $3.50 for mainstream parts. This was somewhat lower than many had expected 1Mb bottom prices to be when cost-ofproduction forecasts were made during the 1988 to 1989 shortage.

Prices for 4Mb DRAMs marched down from about $21 in early 1991 to a fourth-quarter average of about $14, with low-end pricing near

$13.50. Off-standard parts, remnants of the

350-mil package inventory, and slow speed grades went for as low as $11.50. Steady pricing erosion continues into 1992.

Wide DRAIVIs

The market development and opportunity for x8, x9, xl6, and xl8 also holds some substantial imcertainties, as applications using only a few megabytes of DRAMs can draw from monolithic

DRAMs, SIMMs, PS RAMs, and, soon, selfrefreshed 4Mb and 16Mb DRAMs. There are many tough market calls, and an equal number of opporttmities. For 1991, fewer than 2 percent of 1Mb DRAMs were organi2ations other than xl or x4; for 4Mb DRAMs, only about 4 percent were wide DRAMs. This fraction promises to grow substantially for 4Mb, but we will likely have to wait until the 16Mb ramps to see a significant fraction of the market in wide organizations.

During the year, 64Kxl6 DRAMs were available from several suppliers, though most DRAM suppliers seemed content to wait for the 4Mb market to enter. Users seem eager to get the wide parts as soon as the xl and x4 appear, but have been disappointed by wide price disparities and narrow supplier bases.

VRAMs

The much-maligned video RAM (VRAM) business appears to be not so bad after aU. An estimated 16 percent of the 256Ks shipped in 1991 were VRAMs, and 6 percent of the 1Mb DRAMs were actually VRAMs. Considering the fact that

VRAMs didn't enter the market until about two years after their standard part cousins, this is not really that bad. As of year-end 1991, only samples of 2Mb and 4Mb VRAMs were available, indicating that both the designs and use were still lagging. Standards remain a problera

For the few hundred thousand of the 16Mb

DRAMs shipped in 1992, prices began the year at about $300 per xadt but declined to about

$210 at year end. A few orders were placed during 1991 for volumes ranging from 10,000 to

20,000 pieces to be fulfilled early in 1992. Even at $200, the price-per-bit premitim is stUl about

4x compared with 4Mb devices. The rate at which 16Mb can come down and be cost competitive with 4Mb DRAMs is severely limited by the reconstructed cost-based pricing dictated by a host of fair-pricing initiatives in Europe and the United States.

Foundries

Foundry arrangements gained in importance in

DRAMs, as Hitachi and Goldstar teamed up (in addition to the long-standing relationship between Hytmdai and Texas Instruments). For its own peirt, TI had continued difficulty bringing up its large production capacity increments into high volume with costs that allowed it to be profitable. It concluded its 1991 year with yet another losing quarter, its sixth in a row.

Other arrangements contributing to total DRAM output for 1991 included Intel buying OEM

DRAMs from Samsimg and the five-year-old

Motorola/Toshiba joint venture for 1Mb and

March 30, 1992 ©1992 Dataquest Incorporated

MMRY-SEG-DP-9201

Memories Woridwide

4Mb DRAMs. Many suppliers are engaged in collaborative alliances at the 16Mb and 64Mb level, and we expect the exorbitant cost of technology development to drive more DRAM makers into each other's arms as time passes.

Foreign Facilities

At year-end, NEC's RoseviUe, California, 4Mb line joined at least eight other DRAM production sites located outside the home base of the parent company. NEC also produces in Livingston, Scotland. As 1992 opened, TI was producing at its facility in Avezzano, Italy; at its fab in

Miho, Japan; and at the joint venture with Acer in Taiwan. Fujitsu has begun prototyping at its facility in Gresham, Oregon, as well as in

Newton-Aycliffe, Scotland. Motorola is in production at its plant in East Kilbride in the

United Kingdom and at the joint venture with

Toshiba in Tohoku, Japan, in the Sendai prefecture. Hitachi, Mitsubishi, and Oki all do some

DRAM assembly and test at their foreign facilities, as welL The global diffusion of leadingedge manufacturing continues apace, perhaps not as fest as was envisioned in 1989 and 1990, but steadily nonetheless. wUl be a major uncertain element in the equation of DRAM supply and demand for the coming years.

ism DRAM Future in 1992

At year-end 1991, there was increased talk about the imminent arrival of the 16Mb

DRAM. Several companies just now bringing up their 0.5- to 0.6-micron 16Mb fabs gave production schedules running up to a few hundred thousand imits per month by svimmer

1992. Here we wiU have a dilemma, for a number of reasons. First, the cost structiire of the 16Mb is richer still than the 4Mb and promises to retard the rate at which the costs can be reduced into proximity of the 4Mb generation. Second, the user community, which has shifted more toward cost-sensitive PC and consimaer applications and away from mainframes, will support early PPB premivmis less than in earUer generations. Third, trade relations between the United States and Japan will force something like fully loaded market pricing for 16Mb DRAMs, making it tough for suppliers to get dose enough to 4Mb pricing to ramp 16Mb very much until true costs, via yield improvements, are achieved and fixed charges for process and fadHty are behind them.

Dataquest Perspective

The coming year promises to be even more exciting. Though demand is lackluster at present, there also is no gaping excess of production capacity. Many issues regarding the market development for differentiated products remain undecided as to market mix, timing, and even standards. Already we have seen some lowvoltage parts enter the market, but there is no clear direction as to whether 5V will be converted to 3V on-chip, or wiU be 3V only. By the time the 16Mb ramps into its own in 1994 or

1995, these are hkely to be settled. But not for today.

For the time being. Rev II and Rev HI of tiie

4Mb DRAMs will take up most of the wafers started on the most advanced lines of the

DRAM leaders. After all, 1992 will actually be the first year that the 4Mb products have had the field to themselves, as 1Mb has finally been pushed into the postmaturity status—still substantial business, but no new design wins, and continued production dedine that began early in 1991. In our opinion, talk of 16Mb

DRAMs ramping in a big way in 1992 is very premature.

We may also see the unfolding of IBM's semiconductor strategy in 1992, which is certain to impact the merchant DRAM market. So fax, we have seen small quantities sold to Hjmndai from

IBM Japan, but no decision one way or the other as to the final strategy. IBM President Jack

Kuehler indicated in recent EETimes and EBN articles that IBM's production capabilities wiU be aimed squarely at the IBM internal systems market, and will ihus remain captive. This disclaimer notwithstanding, we believe that IBM

Quo Vadis

Overhanging the entire DRAM marketplace, and by imphcation the semiconductor industry, are a host of retum-on-investment and accoimting questions. Though cost accoimting comprises a rather clear disdpline in other industries, there is considerable value spillover, long-term investment accounting, and assignment of accrued costs that make determining actual costs of DRAM production difficvilt to determine. \^^th the Semiconductor Trade

MMRY-SEG-DP-9201 ©1992 Dataquest Incorporated

March 30, 1992

Memories Woridwide

Agreement (STA) of 1986 and the requirement that companies maintain similar costaccounting data internally for the STA2 pressing on one side, and the stratospheric costs of process development and facility expansion to produce products of uncertain payback, the

DRAM industry has become conservative and careful in its willingness to proceed apace.

Since the DRAM shortage cracked apart in the fall of 1989, the market has exhibited a remarkable balance of supply and demand for more than two years. Price declines have been significant since that time, steady and measured—^not the roUer coaster we saw in

1985-1986 or 1981-1982. The stakes are so large and the investment requirements so great that they are no longer ignorable by parent companies. Caution prevails in investment, production, and pricing.

By Lane Mason

SRAM Suppliers Must Run Faster or Fall Behind

The 1991 SRAM market tmderwent some changes, in part due to serious price erosion— especially at the high end of the speed range.

Preliminary 1991 data report a paltry 6 percent revenue growth, in spite of a healthy unit shipment growth of 16 percent. Three companies introduced 4Mb monolithic SRAMs, although the 1Mb part continued on its 1990 path of slow growth.

Despite all of this activity, the names on the list of top 10 vendors did not change (see Table 1).

Changes in rank included Fujitsu and Toshiba exchanging their second-place and third-place positions. The rankings of the two companies in question were within a very small percentage of each other; when the final numbers are confirmed by Dataquest, these ranking changes may reverse themselves.

By region, Japanese companies produced

70.2 percent of all SRAM dollar sales followed by North American companies at 19.3 percent,

Asia/Pacific at 7.3 percent, and European suppliers at 3.1 percent. Figure 1 illustrates how these percentages are broken down into speed categories using a new method of data collection initiated by Dataquest this year. Rather than splitting fest and slow parts by a 70ns delineator as done previously, data are now collected in six speed bins: pseudostatic, slow, 70ns to 45ns,

35ns to 20ns, 15ns to 10ns, and 8ns and faster.

(Figure 1 does not show 8ns data.) We expect this method to be useful in predicting market speed trends in a means more usable to our clients. This method will be used in all future versions of Dataquesf s SRAM forecast and in certain forms of analysis.

Figure 1 shows that North American suppliers focus their efforts more on high-speed SRAMs,

Table 1

Top 10 SRAM Vendors (Millions of Dollars)

1991

Rank

1

6

7

8

9

10

4

5

2

3

1990

Rank

1

4

6

5

7

8

3

2

9

10

Source: Dataquest (March 1992)

Hitachi

Fujitsu

Toshiba

NEC

Sony

Mitsubishi

C5rpress

Motorola

Sharp

Samsung

1990

Revenue

(U.S.$M)

427

238

251

224

195

196

124

99

92

91

1991

Revenue

(U.S.$M)

484

271

269

253

201

186

123

122

117

103

1991

Market

Share(%)

17.9

10.0

9.9

9.3

7.4

6.9

4.5

4.5

4.3

3.8

March 30, 1992 ©1992 Dataquest Incorporated

^fflVIRY-SEG-DP-9201

Memories Woridwide

Figure 1

SRAM Sales, by Speed

Percent

5 0 -

•40-

3 0 -

2 0 -

1 0 -

100

9 0 -

8 0 -

7 0 -

6 0 -

' ' ? ^

10,12,15ns 20, 25, 35ns 45.55, 70ns

f~| North America H Japan

Source: Dataquesi (March 1992)

Slow

Other

Pseudo

while European and Asia/Pacific vendors focus on slower-speed devices. Japanese semiconductor vendors approach the entire market more evenly despite speed grade, and the pseudostatic market is owned by a handful of Japanese companies. This arrangement is in keeping with the strategies chosen by the vendors in each of these regions. North American SRAM vendors attempt to unprove their profitability by spending their efforts on fast devices. The market for these devices may be smaller, but their high average selling prices (ASPs) allow for increased margins. Asia/Pacific SRAM vendors are looking to a neglected part of the market—slow SRAMs, with margins that ate shm—^in order to gain market share through manufacturing prowess.

Vendors in Japan are using a "cover liie market" strategy to continue to dominate.

Shifting to a device focus, vinit sales growth by organization continued to follow the trends established by the end of 1990. The 256K density led unit sales of slow SRAMs. In fest SRAMs, sales of the 256K density still followed unit sales of the 64K density but crossed over the unit sales figure for fast 16K SRAMs.

Dataquest Perspective

Although overall SRAM unit sales grew by

16 percent, dollar sales grew by only 4 percent indicating an overall ASP drop of more than

10 percent. The products that were hardest hit were fast SRAMs, which have seen ASPs drop by as much as 70 percent over the past four quarters, bringing their prices down almost to ttie level of their slower counterparts. Despite this severe price erosion, most high-speed SRAM vendors were not ruined, possibly because they ramped u p tmit shipments and focused on improving the speed of existing products.

Overall, 1991 was a year fraught with difficulties in the SRAM market, but s o M efforts to increase unit shipments and continued pursuit of sales of deiwer and fester SRAMs have paid off by saving companies from a destructive downwani pricing spiral. AU manufecturers that left the market appear to have been replaced by startups and new entrants, and growth is continuing for fabless SRAM vendors as well as for resellers of products manufactured by outside companies.

By Jim Handy

MMRY-SEG-DP-9201 ©1992 Dataquest Incorporated

Maich 30, 1992

Memories Woridwide

Nonvolatile Memories: A Year of Bullish

Flash Growth

Worldwide revenue for nonvolatile memories increased by 11.8 percent in 1991. Shipments were 1.1 billion, up 16.2 percent firom 0.95 billion in 1990. Table 1 presents the ranking for the top 10 companies based on preliminary revenue estimates for 1991.

Flash

Unit shipments of flash memories grew 446 percent, from 2.8 million units to 15.4 million units.

Intel remained the leading supplier, with an

85.0 percent unit shipment market share. It should be noted that a total of 10 companies, listed in Table 2, were shipping product in 1991.

The company rankings for the top 10 nonvolatile memory suppliers show very little change.

Hitachi and Toshiba exchanged places in 1991, according to the preliminary revenue market estimates. However, the figures for those two companies are very close. Hash, EPROM, OTP,

EEPROM, ROM, and NV-RAM make up the field of nonvolatile memories.

EPROM

The EPROM category includes OTP memories.

In 1991, EPROM unit shipments grew by a meager 0.3 percent over 1990 as the EPROM market experienced significant pricing pressvtres. Signetics withdrew from this market, and Advanced

Micro Devices (AMD) maintained its No. 1 position (see Table 3).

Table 1

1991 Preliminaty Estimated Martet Share Ranking:

Worldwide Nonvolatile Memories (Millions of Dollars)

1991

Rank

5

6

7

8

1

2

3

4

9

10

1990

Rank

4

6

7

8

1

2

3

5

9

10

Company

Sharp

Intel

NEC

Hitachi

Toshiba

AMD

SGS-Thomson

Fujitsu

TI

National

North American Companies

Japanese Companies

European Companies

Asia/Pacific Companies

Total Market

Note: Some columns do not add to totals shown because of rounding.

Source: Dataquest (Maich 1992)

1,213

1,571

274

238

3,296

1991

Revenue

357

311

310

253

251

239

201

171

167

110

1990

Revenue

322

268

. 253

212

248

209

198

156

155

120

Percent

Change

10.9

16.0

22.5

19.3

1.2

14.3

1.5

9.6

7.8

-8.3

1,095

1,446

290

118

2,949

10.8

8.6

-5.5

101.7

11.8

1991

Market

Share

10.8

9.4

9.4

7.7

7.6

7.3

6.1

5.2

5.0

3.3

36.8

47.7

8.3

7.2

100.0

March 30, 1992 ©1992 Dataquest Incorporated

MMRY-SEG-DP-9201

Memories Woridwide

Table 2

1991 Worldwide Preliminary Flasli Ranking by Unit Sliipments

(Thousands of Units)

4

5

6

7

8

9

10

1991

Rank

1

2

3

1990

Rank

1

2

3

5

4

Company

Intel

AMD

Toshiba

Atmel

SGS

Seeq

Hitachi

Mitsubishi

n

Catalyst

Total

1991

Units

13,137

1,115

415

351

203

100

65

30

22

5

15,443

Note: Some columns do not add to totals shown because of rounding.

NM = Not meaningful

Source: Dataquest (March 1992)

Table 3

1991 Worldwide Preliminary EPROM Ranl<ing by Unit Sliipments

(Thousands of Units)

3

4

5

6

7

1991

Rank

1

2

12

13

14

15

16

17

18

8

9

10

11

19

20

1990

Rank

5

9

7

8

6

10

11

1

3

4

2

14

12

13

16

17

15

18

19

20

Company

AMD

TI

SGS

Intel

National

Mitsubishi

Signetics

Microchip

Fujitsu

Toshiba

Hitachi

Atmel

NEC

WaferScale

C)rpress

Sharp

Oki

Sony

Catalyst

Seiko-Epson

Total

Note: Some columns do not add to totals shown because of rounding.

Source: Dataquest (March 1992)

1991

Units

71,089

64,057

60,180

1,595

7,145

22,715

21,502

19,130

17,790

12,250

11,770

9,698

7,480

7,459

3,757

2,605

2,075

1,400

1,308

220

425,225

5,089

9,055

5,129

2,609

2,175

2,873

1,200

1,116

800

423,992

1990

Units

62,658

58,331

57,347

9,379

38,976

18,036

24,198

23,220

27,270

12,910

11,200

1990

Units

2,413

0

280

0

34

0

2,828

72

0

29

0

Percent

Change

444.4

NM

48.2

387.5

NM

244.8

NM

NM

-35.3

NM

446.0

1991 Market

Share (%)

85.0

7.2

2.7

2.3

1.3

0.6

0.4

0.2

0.1

0.1

100.0

Percent

Change

13.4

9.9

5.0

-13.15

-4.7

26.0

-11.1

44.0

19.8

-27.8

16.7

17.2

-72.5

0.3

-17.6

-34.7

-5.1

5.1

90.6

-17.4

45.4

1991 Market

Share (%)

16.7

15.0

14.1

12.1

8.7

5.3

5.0

0.6

0.5

0.4

0.3

0.1

100.0

4.5

4.2

2.9

2.8

2.3

1.8

1.8

0.9

MMRY-SEG-DP-9201 ©1992 Dataquest IncoiporatBd Maich 30, 1992

10

Memories Woridwide

Table 4

1991 Worldwide Preliminary EEPROM Ranking by Unit Shipments

(Thousands of Units)

9

10

11

12

4

5

6

7

8

2

3

1991

Rank

1

13

14

15

16

17

18

19

20

1990

Rank

3

9

2

1

4

12

7

5

10

8

6

13

11

18

19

14

Company

SGS

Catalyst

Xicor

National

Oki

Hjomdai

Microchip

Mitsubishi

Hitachi

Rohm

ICT

Atmel

SEEQ

Fujitsu

Samsimg

Siemens

NEC

Philips

Sony

AMD

1991

Units

25^00

25,437

24,502

19,097

16,935

15,000

14,649

14,280

5,350

4,700

4,120

15

16

17

20

Others

Total

316

184,793

Note: Some columns do not add to totals shown because of rounding.

NM = Not meaningful

Source: Dataquest (Maich 1992)

3,743

2,728

2,402

2,132

1,394

670

655

540

344

Percent

Change

56.4

548.7

25.8

-8.9

15.8

400

84.9

1.9

58.5

-16.8

-50.3

47.4

-147

900.8

1,0340

-7.1

241

70.1

42.1

NM

-9.2

45.4

2,539

3,200

240

188

1,500

540

385

380

0

348

127,064

1990

Units

16,500

3,921

19,471

20,970

14,625

3,000

7,920

14,012

3,375

5,650

8,300

1991 Market

Share (%)

13.9

13.8

13.3

10.3

9.1

8.1

7.9

7.7

2.9

2.5

2.2

2.0

1.5

1.3

1.1

0.7.

0.4

0.3

0.3

0.2

0.2

100.0

EEPROM

Unit shipmerits for EEPROMs grew 45.4 percent in 1991. This growtii was fueled by ttie everincreasing demand for low-density/low-ASP

(average selling price) serial EEPROM devices used in consumer electronic products. The

EEPROM category includes NV-RAM shipments.

Table 4 lists the preUminazy ranking of companies in the EEPROM market.

ROM

ROM vaxA shipments grew by 20.7 percent in

1991 from 399 million to 482 million units. The

ROM market growth can be attributed in part to strong demand for electronic games and some migration from EPROM/OTP to ROM (cost reduction) for consimier products. Another area for growth was font cartridges for laser printers

(see Table 5).

Maich 30, 1992

©1992 Dataquest Incorporated

MMRY-SEG-DP-9201

Memories Worldwide

11

Table 5

1991 Worldwide Preliminary ROM Ranking by Unit Siiipments

(Thousands of Units)

8

9

10

11

12

13

14

15

16

17

3

4

5

6

7

1991

Rank

1

2

18

19

1990

Rank

1

2

4

5

3

6

9

7

10

11

13

11

13

12

15

14

16

Company

Sharp

NEC

Fujitsu

Ricoh

Toshiba

Hitachi

Samsung

Windbond

Matsushita

Sony

Macronix

Atmel

Gould

Mitsubishi

IMP

Goldstar

Oki

Seiko-Epson

NCR

Total

1991

Units

146,151

62,180

43,260

41,232

39,740

27,820

27,705

17,806

16,575

14,850

7,972

3,743

6,978

5,300

3,492

3,105

1,736

1,180

584

481,851

Note: Some columns do not add to totals shown because of rounding.

NM = Not meaningful

Souice: Dataquest (Maich 1992)

Dataquest Perspective

Overall, nonvolatile memories grew at a rate twice that of DRAMs and almost three times that of SRAMs. EPROM shipments were flat, and the average selling prices suffered. Dataquest believes that the lack of growth in the

EPROM market resulted from ttie declining automotive market and the weak data processing (PC) segment In a recessional environment where cost cutting became necessary for survival of companies and products,

1990

Units

132,150

67,542

33,445

30,300

35,990

22,680

0

11,600

20,160

11,050

452

2,539

7,580

2,950

3,450

0

1,560

1,200

1,047

399,303

Percent

Change

10.6

-7.9

29.3

36.1

10.4

22.7

NM

53.5

-17.8

34.4

1,663.7

47.4

-7.9

79.7

1.2

NM

11.3

-1.7

-44.2

20.7

EPROM/OTP was often replaced by ROM. On the other hand, EEPROM unit shipments grew as more and more consumer electronic products

(TVs, VCRs, camcorders, cameras) began using serial EEPROMs. Flash was a "hot" market in

1991 and certainly lived up to expectations for substantial growtiti. Flash devices are now beginning to replace EPROM/OTP devices and at times ROM devices in applications that benefit from flash's flexibility.

By Nicolas Samaras

1991 Market

Share (%)

30.3

12.9

9.0

8.6

8.3

5.8

5.7

3.7

3.4

3.1

1.6

2.0

1.4

1.1

0.7

0.6

0.4

0.2

0.1

100.0

IVIMW-SEG-DP-9201

©1992 Dataquest IncoipoiatBd March 30, 1992

12 Memories Woridwide

For More Information

On the topics in this issue Lane Mason, Director (408) 437-8120

About online access (408) 437-8576

About upcoming Dataquest conferences (408) 437-8245

About your subscription or other Dataquest publications (408) 437-8285

Via fax request (408) 437-0292

The content of this report represents our interpretation and analysis of information generally available to the public or released by responsible individuals in the subject companies, but is not guaranteed as to accuracy or completeness. It does not contain material provided to us in confidence by our clients.

Individual companies reported on and analyzed by E>ataquest may be clients of this and/or other Dataquest services. This information is not furnished in coimection with a sale or offer to sell securities or in connection with the solicitation of an offer to buy securities. This firm and its parent and/or their officers, stockholders, or members of their femilies may, from time to time, have a long or short position in the securities mentioned and may sell or buy such securities.

March 30, 1992 ©1992 Dataquest InconxHated M\/IRY-SEG-DP-9201

Dataquest Perspective

Memories Worldwide

MMRY-SEG-DP-9201

March 30, 1992

In This Issue-

Market Analysis

Market Analysis

Pnlimnary 1991 Woridwide MOS

Memory Market Share Estimates

Preliminary 1991 Worldwide MOS Memory MarM

Share Estimates

Dataquest has completed its preliminary 1991

Dataquest has completed its preliminary 1991 MOS

MOS memory market share survey. We mailed a memory market share survey analysis. Although survey questionnaire to more than 150 semiconthe market recovered from a dismal 1990 showing, ductor vendors in early November. The responserious price erosion coupled with lackluster dents provided us with detailed breakouts of growth in new densities made 1991 anything but a their revenue and imit shipments based on a halcyon year.

By Lane Mason, Jim Handy, and Nicolas Samaras Page 1

combination of year-to-date data and companygenerated forecasts for the rest of the year. The collected results are published in this article. We will continue to refine and update the data, and

A Year of Transition for DRAMs

we plan to release our final market share data

The year 1991 was in many ways a transitional documents on May 31, 1992. one for DRAMs. Korean companies are gaining market share over their Japanese counterparts and rising production costs £rom the 1Mb to ttie 4Mb generation are initiating an increase in anticipated

Market Share Highlights

floor costs. Surface-mount packages have taken over, and several new technologies are now attempting to gain market acceptance.

The following analysis covers the three areas of

MOS memory tracked by Dataquest: DRAMs,

SRAMs, and nonvolatile memories (EPROMs,

By Lane Mason Page 2

EEPROMs, ROMs, and Flash memories).

SRAM Suppiiers Must Run Faster or Fall Behind

Dataquesfs preliminary 1991 market share estimates for static RAM suppliers show little change from 1990. The rankings have barely changed, despite the fact that most manufacturers dranriatically increased their unit shipments. Dramatic ASP erosion in the ^ t SRAM arena has been the culprit.

By Jim Handy Page 6

Nonvolatile Memorim: A Y^ir of Bullish Flash Growth

MOS memory grew in 1991 at a relatively moderate 6 percent, in contrast to 1990's disastrous 17 percent decline. Observed in this light, however, MOS memory has made a rather impressive recovery, especially considering the fact that the 4Mb DRAM has not taken off as quickly as some had hoped. Toshiba returned its

No. 1 ranking for all MOS memories, shipping more than $1.44 billion. In 1991, Toshiba was followed in order by Hitachi, NEC, Fujitsu, and

Samsung.

Nonvolatile memory revenue growth in 1991 outpaced that of DRAM and SRAM according to

Dataquesfs preliminary estimates. Strong electronic game sales helped ROM shipments. EPROM remained flat as both the automotive and data processing segments were down. Hash memory was the bright new star with substantial giowtti.

By Nicolas Samaras Page 8

Highlights of the 1991 MOS memory market include the following:

• Overall dollar growth in DRAMs was 5.0 percent; in SRAMs, 5.0 percent; and in nonvolatile memories, 11.7 percent.

Dataqyest

n n acompanyof

J i l t TheDun&BradstiEctCocporatian

Dataquest is a registared trademailc of A.C. Nielsen Company.

File behind the Perspectives tab inside the binder labeled

Memories WorlduHde

e l 9 9 2 Oataqueet incofporakd, Reproduction PraNbitad 0012759

Memories Woridwide

Figure 1

1991 Estimated Worldwide iVIemory Saies, tiy Type

(Miliions of Dollars)

/ static RAM

/ $2,710

A Year of ThmsiUott for DRAMs

The year 1991 was in many ways transitional for

DRAMs. Aside from the triennial move from generation to generation, we began to see the long-anticipated emergence of wide DRAMs at the 4Mb level, continued market share gains by

Korean manufacturers over the stUl-dominant

Japanese suppliers, and, with the 4Mb ramp-up, the first real impact of the new economics of

DRAM production that promise to raise ultimate floor prices itoin. generation to generation in degrees never seen before.

Dynamic RAM i

1 Nonvolatile /

\ $3,296 /

$6,800 i

Prices per bit for 4Mb DRAMs crossed over those of 1Mb parts about midyear, and by yearend, 4Mb DRAMs were generating greater revenue and shipping more bits per quarter.

Source: Dataquest (March 1992)

For the most part, the initial 350-mil 4Mb

DRAM was replaced by the historical standard

300-mil package and DRAM speeds inched downward. The majority of products are now available at 70ns to 80ns. The market for veryhigh-speed optimized designs was foimd wanting, as MPU speeds made cache systems imavoidable.

• Rapid price erosion decreased the growth of

SRAM sales significantiy.

• Micron Technology became a top-10 player in the DRAM market, at No. 8.

By year-end, low pricing in the 1Mb market had encouraged several manufacturers to ramp down production and concentrate entirely on the

4Mb and 16Mb devices.

• Rankings of the top 10 SRAM and nonvolatile memory manufacturers remained nearly imchanged.

• Flash memories underwent the most dramatic change, increasing tinit shipments by

491 percent.

• Japanese vendors took 60.6 percent of the

DRAM market and 71.5 percent of the SRAM market. Japanese market share of nonvolatile memories was far lower at 47.7 percent.

The pie chart in Figure 1 shows the relationship in dollar sales between the DRAM, SRAM, and nonvolatile memory siarket segments.

By Lane Mason

Jim Handy

Nicolas Samaras

1991 Market Share Movement

Table 1 shows suppliers' 1991 DRAM revenue.

Despite an estimated 31 percent price-per-bit

(PPB) erosion, the DRAM market managed revenue growth of about 5 percent for the year. Japanese companies, led by No. 1 Toshiba with an estimated $904 million in sales, took 6 of the top

10 places in 1991 DRAM production. No. 2

Samsimg was the top 1Mb shipper. Micron

Technology, coming off a difficult transition to

1Mb in 1990 that impacted 1990 revenue, showed the top growth rate among the top 10 and had sales estimated at $362 million in 1991.

Despite continuing trouble with its 4Mb at several facilities, Texas Instruments was the No. 6 producer, altiiou^ it dropped about 1 percent in sales to $571 million.

As a portent of things to come, Goldstar and

Hytmdai each grew shipments m,ore than

143 percent, but remained out of sight of the top 10, for 1991 at least. Overall, Korean

Match 30, 1992

®1992 Oataquest Incorporated IUMRy-SEG-DP-g201

Memories Woridwide

Table 1

1991 DRAM Revenue (MiHions of Dollars)

Company

Toshiba

Samstmg

NEC

Hitachi

Texas Instruments

Fujitsu

Mitsubishi

Micron Technology

Old Semiconductor

Siemens

Motorola

Goldstar

Hjomdai

Matsushita

NMB Semiconductor

Vitelic

Sharp

Others

Total

Source: Dataquest (March 1992)

1990 Sales

947

809

669

597

576

497

75

131

171

58

72

110

6,475

503

294

293

306

292

75

1991 Sales

904

900

698

675

571

484

467

362

319

287

264

204

186

116

116

79

62

106

6,800

companies garnered virtually all of the aggregate

DRAM market growth, largely at the expense of the more established Japanese suppliers.

Goldstar and Hyundai have become significant enough players to threaten low-end pricing, and in 1991 became bona fide suppliers of 4Mb

DRAMs. Samsung cemented its position as a leading contender for the DRAM crown, shipping 4Mb DRAMs at a rate within striking distance of market leader Hitachi Samsung's 16Mb parts are said to be on a par with the best in the industry—^no price discovmts anymore! Samsung will be interesting to watch in 1992 as it challenges for the DRAM lead, as will Hyundai and Goldstar as they upgrade both their products and customer base.

One important advantage that newcomers to the

DRAM market have in 1992 and beyond is the changing channels to end users. A growing fraction of tihe DRAM business is upgrades for PCs bought at mom-and-pop outlets: Blue-chip accoimts, with exacting system requirements and lengthy, expensive qualifications, make up a steadily decliidng share of DRAM demand.

Add-on memoiy is an attractive outlet for the price competitive new kids on the block—

Goldstar, Hyundai, and any other late arrivals to the 4Mb race—^who want to keep their fobs full and learn the business.

Packaging Trends

The 1Mb generation was the last hiurah for the dual in-line package (DIP), as its share of the business migrated rapidly to SOJ and, it appears, soon to TSOP. DIPs were about 22 percent of 1Mb DRAM unit shipments in 1991, compared with about 4 percent of the 4Mb shipments. ZEPs held steady at about 15 percent of the 4Mb generation. So, after five generations of

DRAMs ttiat went into the DIP, we have gone through two major packaging turnovers in just the past two generations, witii the last yet to fuUy express itself in the market

Change (%)

-4.5

11.2

4.3

13.1

-0.9

-2.6

-7.2

23.1

8.9

-6.0

-10.0

172.0

148.0

-11.5

-32.2

36.0

-13.9

-3.6

5.0

MMW-SEG-DP-9201 ®1992 Dataquest IncnpocatBd

Maich 30, 1992

Memories Woridwide

The single in-line memory module (SIMM) market constituted an estimated 40 percent of

DRAM sales at year-end, but was difficult to size because of the prevalence of aftermarket

SIMM packagers. During the year, the SIMM issue was further muddied by the Wang lawsuits, which draw a low royalty wall arotmd

30-pin x9 SIMMs. Manufacturers are watching to see which way the winds blow, pushing the demand over to x36 modules, while begrudgingly paying the 3 to 4 percent royalty to Wang.

As a significant fraction of the DRAMs in

SIMMs fiU in the upgrade memory needs of the

PC and workstation installed base, it looks as though x9 will continue in a major way for

1992. In addition, there appears to be no rush in newer PCs to design expansion slots to accommodate x36 modules, though some are doing so.

Pricing

Pricing for all DRAM densities continued through 1991 in an vminterrupted decline. There were no significant "spot" shortages in packages, speed grades, or operating modes. Yearend pricing for 1Mb DRAMs was often below

$4, with a bottom of about $3.50 for mainstream parts. This was somewhat lower than many had expected 1Mb bottom prices to be when cost-ofproduction forecasts were made during the 1988 to 1989 shortage.

Prices for 4Mb DRAMs marched down from about $21 in early 1991 to a fourth-quarter average of about $14, with low-end pricing near

$13.50. Off-standard parts, remnants of the

350-mil package inventory, and slow speed grades went for as low as $11.50. Steady pricing erosion continues into 1992.

Wide DRAIVIs

The market development and opportunity for x8, x9, xl6, and xl8 also holds some sul^stantial uncertainties, as applications using only a few megabytes of DRAMs can draw from monolithic

DRAMs, SIMMs, PS RAMs, and, soon, selfrefreshed 4Mb and 16Mb DRAMs. There are many tough market calls, and an equal number of opportunities. For 1991, fewer than 2 percent of 1Mb DRAMs were organizations other than xl or x4; for 4Mb DRAMs, only about 4 percent were wide DRAMs. This fraction promises to grow substantially for 4Mb, but we wiU likely have to wait imtil the 16Mb ramps to see a significant fraction of the market in wide organizations.

During the year, 64Kxl6 DRAMs were available from several suppliers, though most DRAM suppliers seemed content to wait for the 4Mb market to enter. Users seem eager to get the wide parts as soon as the xl and x4 appear, but have been disappointed by wide price disparities and narrow supplier bases.

VRAMs

The much-maligned video RAM (VRAM) business appears to be not so bad after all. An estimated 16 percent of the 256Ks shipped in 1991 were VRAMs, and 6 percent of the 1Mb DRAMs were actually VRAMs. Considering the fact that

VRAMs didn't enter the market imtil about two years after their standard part cousins, this is not really that bad. As of year-end 1991, only samples of 2Mb and 4Mb VRAMs were available, indicating that both the designs and use were still lagging. Standards remain a problem.

For the few hundred thousand of the 16Mb

DRAMs shipped in 1992, prices began the year at about $300 per unit but declined to about

$210 at year end. A few orders were placed during 1991 for volumes ranging fitom 10,000 to

20,000 pieces to be fulfilled early in 1992. Even at $200, the price-per-bit premium is still about

4x compared with 4Mb devices. The rate at which 16Mb can come down and be cost competitive with 4Mb DRAMs is severely limited by the reconstructed cost-based pricing dictated by a host of fair-pricing initiatives in Europe and the United States.

Foundries

Foundry arrangements gained in importance in

DRAMs, as Hitachi and Goldstar teamed up (in addition to the long-standing relationship between Hyimdai and Texas Instruments). For its own part, TI had continued difficulty bringing up its large production capacity increments into high volume with costs that allowed it to be profitable. It concluded its 1991 year with yet another losing quarter, its sixth in a row.

Other arrangements contributing to total DRAM output for 1991 included Intel buying OEM

DRAMs from Samsung and the five-year-old

Motorola/Toshiba joint venture for 1Mb and

March 30, 1992

©1992 Dataquest Incorporated

MMRY-SEG-OF4201

Memories Woridwide

4Mb DRAMs. Many suppliers are engaged in collaborative alliances at the 16Mb and 64Mb level, and we expect the exorbitant cost of technology development to drive more DRAM makers into each other's arms as time passes.

Foreign Facilities

At year-end, NEC's Roseville, California, 4Mb line joined at least eight other DRAM production sites located outside the home base of the parent company. NEC also produces in Livingston, Scotland. As 1992 opened, H was producing at its facility in Avezzano, Italy; at its fab in

Miho, Japan; and at the joint venture with Acer in Taiwan. Fujitsu has begun prototyping at its fadUty in Gresham, Oregon, as well as in

Newton-Aychffe, Scotland. Motorola is in production at its plant in East Kilbride in the

United Kingdom and at the joint venture with

Toshiba in Tohoku, Japan, in the Sendai prefecture. Hitachi, Mitsubishi, and Oki all do some

DRAM assembly and test at their foreign facilities, as welL The global diffusion of leadingedge manufacturing continues apace, perhaps not as fast as was envisioned in 1989 and 1990, but steadily nonetheless. will be a major uncertain element in the equation of DRAM supply and demand for the coming years.

16Mb DRAM Future in 1992

At year-end 1991, there was increased talk about the imminent arrival of the 16Mb

DRAM. Several companies just now bringing up their 0.5- to 0.6-micron 16Mb fabs gave production schedules running up to a few hundred thousand units per month by summer

1992. Here we will have a dilemma, for a number of reasons. First, the cost structure of the 16Mb is richer stiU than the 4Mb and pronuses to retard the rate at which the costs can be reduced into proximity of the 4Mb generation. Second, the user commimity, which has shifted more toward cost-sensitive PC and consumer applications and away from mainframes, will support early PPB premiums less than in earlier generations. Third, frade relations between the United States and Japan will force something like fuUy loaded market pricing for 16Mb DRAMs, making it tough for suppliers to get close enough to 4Mb pricing to ramp 16Mb very much until true costs, via yield improvements, are achieved and fixed charges for process and iadRty are behind them.

Dataquest Perspective

The coming year promises to be even more exciting. Though demand is lackluster at present, there also is no gaping excess of production capacity. Many issues regarding the market development for differentiated products remain undecided as to market mix, timing, and even standards. Already we have seen some lowvoltage parts enter the market, but there is no clear direction as to whether 5V wiU be converted to 3V on-chip, or will be 3V only. By the time the 16Mb ramps into its own in 1994 or

1995, these are likely to be settled. But not for today.

For the time being, Rev II and Rev III of the

4Mb DRAMs will take up most of the wafers started on the most advanced lines of the

DRAM leaders. After all, 1992 will actually be the first year that the 4Mb products have had the field to themselves, as 1Mb has finally been pushed into the postmaturity status—still substantial business, but no new design wins, and continued production decline that began early in 1991. In our opinion, talk of 16Mb

DRAMs ramping in a big way in 1992 is very premature.

We may also see the vmfolding of IBM's semiconductor strategy in 1992, which is certain to impact the merchant DRAM market. So far, we have seen small quantities sold to H)aindai from

IBM Japan, but no decision one way or the other as to the final strategy. IBM President Jack

Kuehler indicated in recent EETimes and EBN articles that IBM's production capabilities will be aimed squarely at the IBM internal systems market, and will dtus remain captive. This disclaimer notwithstanding, we believe that IBM

Quo Vadis

Overhanging the entire DRAM marketplace, and by implication the semiconductor industry, are a host of retum-on-investment and accounting questions. Though cost accoimting comprises a rather clear discipline in other industries, there is considerable value spillover, long-term investment accounting, and assignment of accrued costs that make determining actual costs of DRAM production difficult to determine, \^^th the Semiconductor Tbrade

MMRV-SEG-DP-gaOl ©1992 Dataquest IncofpoiatBd March 30, 1992

Memories Worldwide

Agreement (STA) of 1986 and the requirement that companies maintain similar costaccotuiting data internally for the STA2 pressing on one side, and the stratospheric costs of process development and fadlity expansion to produce products of uncertain payback, the

DRAM industry has become conservative and careful in its willingness to proceed apace.

Since the DRAM shortage cracked apart in the fall of 1989, the market has exhibited a remarkable balance of supply and demand for more than two years. Price declines have been significant since that time, steady and measiujed—^not the roller coaster we saw in

1985-1986 or 1981-1982. The stakes are so large and the investment requirements so great that they are no longer ignorable by parent companies. Caution prevails In investment, production, and pricing.

By Lane Mason

SRAM Suppliers Must Run Faster or FaK Behind

The 1991 SRAM market underwent some changes, in part due to serious price erosion— especially at the high end of the speed range.

Preliminary 1991 data report a paltry 6 percent revenue growth, in spite of a healthy imit shipment growth of 16 percent. Three companies introduced 4Mb monolithic SRAMs, although the 1Mb part continued on its 1990 path of slow growth.

Despite all of this activity, the names on the list of top 10 vendors did not change (see Table 1).

Qianges in rank included Fujitsu and Toshiba exchanging their second-place and third-place positions. The rankings of the two companies in question were withdn a very small percentage of each other; when the final numbers are confirmed by Dataquest, these ranking changes may reverse themselves.

By region, Japanese companies produced

70.2 percent of all SRAM dollar sales followed by North American companies at 19.3 percent,

Asia/Pacific at 7.3 percent, and European suppliers at 3.1 percent. Figure 1 illustrates how these percentages are broken down into speed categories using a new method of data collection initiated by Dataquest this year. Rather than splitting fast and slow parts by a 70ns delineator as done previously, data are now collected in six speed bins: pseudostatic, slow, 70ns to 45ns,

35ns to 20ns, 15ns to 10ns, and 8ns and foster.

(Figure 1 does not show 8ns data.) We expect this method to be useful in predicting market speed trends in a means more usable to our clients. This method will be used in all future versions of Dataquesf s SRAM foreccist and in certain forms of analysis.

Figure 1 shows that North American suppliers focus their efforts more on high-speed SRAMs,

Table 1

Top 10 SRAM Vendors (Millions of Dollars)

7

8

9

10

2

3

4

1991

Rank

1

5

6

1990

Rank

1

3

2

4

6

5

7

8

9

10

Hitachi

Fujitsu

Toshiba

NEC

Sony

Mitsubishi

Cypress

Motorola

Sharp

Samsimg

Source: Dataquest (March 199Z)

1990

Revenue

(U.S.$M)

427

238

251

224

195

196

124

99

92

91

1991

Revenue

(U.S.$M)

484

271

269

253

201

186

123

122

117

103

1991

Market

Share(%)

17.9

10.0

9.9

9.3

7.4

6.9

4.5

4.5

4.3

3.8

March 30, 1992

©1992 Dataquest Incoiporalad IMMRY-SEG-0P-9201

Memories Woridwide

Figure 1

SRAM Sales, by Speed

Percent

100

10.12,15ns 20,25.35ns 45,55.70ns

r~| North America ^ Japan

Source: Dataqueet (March 1992)

X

Slow Psaudo

• Other

while Europeein and Asia/Pacific vendors focus on slower-speed devices. Japanese semiconductor vendors approach the entire market more evenly despite speed grade, and the pseudostatic market is owned by a handful of Japanese companies. This arrangement: is in keeping with the strategies chosen t y the vendors in each of these regions. North American SRAM vendors attempt to improve their profitability by spending their efforts on fast devices. The market for these devices may be smaller, but their high average selling prices (ASPs) allow for increased margins. Asia/Pacific SRAM vendors are looking to a neglected part of the narket—slow SRAMs, with margins that are slim—^in order to gain market sha]% through manufeicturing prowess.

Vendors in Japzm are using a "cover tiie market" strategy to continue to dominate.

Shifting to a de>{ice focus, imit sales growth by organization continued to follow the trends established by the end of 1990. The 256K density led vadt sales of slow SRAMs. In fest SRAMs, sales of the 256K density still followed unit sales of the 64K density but crossed over the imit sales figure for fast 16K SRAMs.

Dataquest Perspective

Although overall SRAM unit sales grew by

16 percent, dollar sales grew by only 4 percent, indicating an overall ASP drop of more than

10 percent The products that were hardest hit were fast SRAMs, which have seen ASPs drop by as much as 70 percent over the past four quarters, bringing their prices down almost to ttie level of their slower counterparts. Despite this severe price erosion, most high-speed SRAM vendors were not ruined, possibly becaiise they ramped up wait shipments and focused on improving the speed of existing products.

Overall, 1991 was a year fraught with difficulties in the SRAM market, but solid efforts to increase tmit shipments and continued pursuit of sales of denser and faster SRAMs have paid off by saving companies from a destructive downward pricing spiraL All nwnufacturers that left the market appear to have been replaced by startups and new entrants, and growth is contintiing for fabless SRAM vendors as well as for resellers of products manufactured by outside companies.

By Jim Handy

IW»«W-SEG-DP-9201

®1992 Dataquest IncoiporabBd March 30, 1992

8

Memories Woridwide

Homfolatlle Memories: A Year of Bullish

Flash Growth

Worldwide revenue for nonvolatile memories increased by 11.8 percent in 1991. Shipments were 1.1 biUion, up 16.2 percent from 0.95 billion in 1990. Table 1 presents the ranking for the top 10 companies based on preliminary revenue estimates for 1991.

Flash

Unit shipments of flash memories grew 446 percent, from 2.8 million units to 15.4 million xmits.

Intel remained the leading supplier, with an

85.0 percent itnit shipment market share. It should be noted that a total of 10 companies, listed in Table 2, were shipping product in 1991.

The company rankings for the top 10 nonvolatile memory suppliers show very little change.

Hitachi and Toshiba exchanged places in 1991, according to the preliminary revenue market estimates. However, the figures for those two companies are very close. Flash, EPROM, OTP,

EEPROM, ROM, and NV-RAM make up the field of nonvolatile memories.

EPROM

The EPROM category includes OTP memories.

In 1991, EPROM unit shipments grew by a meager 0.3 percent over 1990 as the EPROM market experienced significant pricing pressures. Signetics withdrew from this market, and Advanced

Micro Devices (AMD) maintained its No. 1 position (see Table 3).

Table 1

1991 Preliminary Estimated Market Share Ranking:

Worldwide Nonvolatile Memories (Millions of Dollars)

3

4

5

6

7

1991

Rank

1

2

8

9

10

6

7

8

9

10

3

5

4

1990

Rank Company

1

2

Sharp

Intel

NEC

Hitachi

Toshiba

AMD

SGS-Thomson

Fujitsu

TI

National

1991

Revenue

357

311

310

253

251

239

201

171

167

110

North American Companies

Japanese Companies

European Compemies

Asia/Padfic Compemies

Total Market

Note: Some columns do not add to totab shown because of rounding.

Souice: Dataquest (March 1992)

1,213

1,571

274

238

3,296

1990

Revenue

322

268

. 253

212

248

209

198

156

155

120

Percent

Change

10.9

16.0

22.5

19.3

1.2

14.3

1.5

9.6

7.8

-8.3

1,095

1,446

29C

118

2,949

10.8

8.6

-5.5

101.7

11.8

1991

Market

Share

10.8

9.4

9.4

7.7

7.6

7.3

6.1

5.2

5.0

3.3

36.8

47.7

8.3

7.2

100.0

March 30, T992

®1992 Dataquest IncoporalBd

MMRY-SEG-DP'9201

Memories Woridwide

Table 2

1991 Wbridwide Preliminary Fiasii Ranking by Unit Siiipments

(Tiiousands of Units)

6

7

8

9

10

3

4

5

1991

Rank

1

2

1990

Rank

1

2

3 f i

Company

Intel

AMD

Toshiba

Atmel

SGS

Seeq

Hitachi

Mitsubishi n

Catalyst

Total

1991

Units

13,137

1,115

415

351

203

100

65

30

22

5

15,443

Note: Some columns do not add to totals shown because of rounding.

NM s Not meaningful

Source: Dataquest (March 1992)

Table 3

1991 Wbrldwide Preliminary EPROM Ranking by Unit Siiipments

(Tiiousands of Units)

Z

3

4

5

6

7

8

1991

Rank

1

?

10

11

17

18

19

20

12

13

14

15

16

1990

Rank

15

18

19

20

1

3

4

2

5

9

7

8

6

10

11

14

12

13

16

17 *

Company

AMD

TI

SGS

Intel

National

Mitsubishi

Signetics

Microchip

Fujitsu

Toshiba

Hitachi

Atmel

NEC

WaferScale

Cypress

Sharp

Oki

Sony

Catalyst

Seiko-Epson

Total

Note: Some columns do not add to totals shown because of rounding.

Source: Dataquest (March 1992)

1991

Units

71,089

64,057

60,180

1,595

7,145

22,715

21,502

19,130

17,790

12,250

11,770

9,698

7,480

7,459

3,757

2,605

2,075

1,400

1,308

220

425,225

1990

Units

62,658

58,331

57,347

9,379

38,976

18,036

24,198

23,220

27,270

12,910

11,200

5,089

9,055

5,129

2,609

2,175

2,873

1,200

1,116

800

423,992

1990

Units

2,413

0

280

72

0

29

0

0

34

0

2,828

Percent

Change

444.4

NM

48.2

387.5

NM

244.8

NM

NM

-35.3

NM

446.0

1991 Market

Share (%)

85.0

7.2

2.7

2.3

1.3

0.6

0.4

0.2

0.1

0.1

100.0

Percent

Change

13.4

9.9

5.0

-13.15

-4.7

26.0

-11.1

-17.6

-34.7

-5.1

5.1

90.6

-17.4

45.4

44.0

19.8

-27.8

16.7

17.2

-72.5

0.3

1991 Market

Share (%)

16.7

15.0

14.1

12.1

8.7

0.9

0.6

0.5

0.4

0.3

0.1

100.0

5.3

5.0

4.5

4.2

2.9

2.8

2.3

1.8

1.8

MMRY-SEG-DP^201 ®1992 Dataquest IncopoialBd Maich 30, 1992

10

Memories Worldwide

Table 4

1991 Wbrkjwide Preliminaiy EEPROM Ranking by Unit Stiipments

(Tliousands cf Units)

3

4

5

1991

Rank

1

2

2

1

4

1990

Rank

3

9

Company

SGS

Catalyst

Xicor

1991

Units

25^00

25,437

24^02

19,097

16,935

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

12

7

5

10

8

6

13

11

18

19

14

15

16

17

20

National

Oki

Hyimdai

Microchip

Mitsubishi

Hitachi

Rohm

ICT

Atmel

SEEQ

Fujitsu

Samsimg

Siemens

NEC

15,000

14,649

14,280

5,350

4,700

4,120

3,743

2,728

2,402

2,132

1,394

670

655

Philips

Sony

AMD

Others

Total

Note: Some columns do not add to totals shown because of rounding.

NM = Not meaningful

Source: Dataquest (Match 1992)

540

344

316

184,793

5,650

8,300

2,539

3,200

240

188

1,500

540

385

380

0

348

127,064

1990

Units

16,500

3,921

19,471

20,970

14,625

3,000

7,920

14,012

3,375

Percent

Change

56.4

548.7

25.8

-8.9

15.8

400

84.9

1.9

58.5

-16.8

-50.3

47.4

-14.7

900.8

1,034.0

-7.1

24.1

70.1

42.1

NM

-9.2

45.4

1991 Market

Share (%)

13.9

13.8

13.3

10.3

9.1

8.1

7.9

7.7

2.9

2.5

2.2

2.0

1.5

1.3

1.1

0.7

0.4

0.3

0.3

0.2

0.2

100.0

EEPROM

Unit shipmerits for EEPROMs grew 45.4 percent in 1991. This growth was fueled by the everincreasing demand for low-density/low-ASP

(average selling pricej serial EEPROM devices used in consumer electronic products. The

EEPROM category includes NV-RAM shipments.

Table 4 lists the preliminary ranking of companies in the EEPROM market.

ROM

ROM xmit shipments grew by 20.7 percent in

1991 from 399 million to 482 million units. The

ROM market growth can be attributed in part to strong demand for electronic games and some migration from EFROM/OTP to ROM (cost reduction) for consumer products. Another area for growth was font cartridges for laser printers

(see Table 5).

March 30, 1992

(S)1992 Dataquest incorporated

MMRY-SEG-OP-9201

Memories Woridwide

11

Table 5

1991 Wbrldwide Preliminary ROM Ranldng i}y Unit Stiipments

Crtiousands of Units)

1991

Rank

1

8

9

10

11

12

13

14

15

16

17

18

5

6

7

2

3

4

19

1990

Rank

1

2

4

5

3

6

9

7

10

11

13

11

13

12

15

14

16

Company

Sharp

NEC

Fujitsu

Ricoh

Toshiba

Hitachi

Samsung

Windbond

Matsushita

Sony

Macronbc

Atmel

Gould

Mitsubishi

IMP

Goldstar

Oki

Seiko-Epson

NCR

Total

1991

Units

146,151

62,180

43,260

41,232

39,740

27,820

27,705

17,806

16,575

14,850

7,972

3,743

6,978

5,300

3,492

3,105

1,736

1,180

584

481,851

Note: Some columns do not add to totals shown because of rounding.

NM = Not meaningful

Source: Dataquest (March 1992)

Dataquest Perspective

Overall, nonvolatile memories grew at a rate twice that of DRAMs and almost three times that of SRAMs. EPROM shipments were flat, and the average selling prices sviffered. Dataquest believes that the lack of growth in the

EPROM market resulted from tiie declining automotive market and the weak data processing (PC) segment In a recessional environment wliere cost cutting became necessary for survival of companies and products,

1990

Units

132,150

67,542

33,445

30,300

35,990

22,680

0

11,600

20,160

11,050

452

2,539

7,580

2,950

3,450

0

1,560

1,200

1,047

399,303

Percent

Change

10.6

-7.9

29.3

36.1

10.4

22.7

NM

53.5

-17.8

34.4

1,663.7

47.4

-7.9

79.7

1.2

NM

11.3

-1.7

-44.2

20.7

EPROM/OTP was often replaced by ROM. On the other hand, EEPROM imit shipments grew as more and more consumer electronic products

(TVs, VCRs, camcorders, cameras) began using serial EEPROMs. Flash was a "hot" market in

1991 and certainly lived up to expectations for substantial growth. Flash devices are now beginning to replace EPROM/OTP devices and at times ROM devices in applications that benefit from flash's flexibility.

By Nicolas Samaras

1991 Market

Share (%)

30.3

12.9

9.0

8.6

8.3

5.8

5.7

3.7

3.4

3.1

1.6

2.0

1.4

1.1

0.7

0.6

0.4

0.2

0.1

100.0

MMRY-SEG-OP-9201 01992 Dataquest IncorpoialBd

March 30, 1992

12 Memories Woridwide

For More Information

On the topics in this Issue Lane Mason, Director (408) 437-8120

About online access (408) 437-8576

About upcoming Dataquest conferences (408) 437-8245

About your subscription or other Dataquest publications (408) 437-8285

Via fax request (408) 437-0292

The content of this report represents our interpretation and analysis of infbrmatian generally available to the public or released by responsible individuals in the stibject con^anies, but is not guaranteed as to accuracy or completeness. It does not contain material provided to us in confidence by our dients.

Individual companies reported on and analyzed by Dataquest may be clients of this and/or other Dataquest services. This infonnation is not furnished in connection with a sale or otter to sell securities or in connection with the solicitation of an offer to buy securities. This firm and its parent and/or their officers, stockholdeis, or members of their families may, from time to time, have a long or short position in the securities moitioned and may sell or buy such securities.

March 30, 1992 ®1992 Dataquest Inoxpoiated MffiY-8EG-OP^9201

Dataquest Perspective

FILE COPY

Do Not Remove

Memories Worldwide

March 30,1992

Errata

In the article entitled "Nonvolatile Memories:

A Year of Bullish Flash Growth" in Memories

Woridwide Dataquest Perspective issue 9201, which was dated March 30, 1992, Table 3 had incorrect data. We apologize for any confusion this may have caused and reprint the table here with corrected data.

Table 3

1991 Woridwide Preliminary EPROM Ranldng by Unit Stiipments

(Tiiousands of Units)

1991

Rank

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

1990

Rank

5

9

7

8

6

10

11

1

3

4

2

14

12

13

16

17

15

18

19

20

Company

AMD

TI

SGS

Intel

National

Mitsubishi

Signetics

Microchip

Fujitsu

Toshiba

Hitachi

Atmel

NEC

WafeiScale

Cypress

Sharp

Oki

Sony

Catalyst

Seiko-Epson

Total

Note: Some columns do not add to totals shown because of rounding.

Source: Dataquest (March 1992)

1991

Units

71,089

64,057

60,180

51,595

37,145

22,715

21,502

19,130

17,790

12,250

11,770

9,698

7,480

7,459

3,757

2,605

2,075

1,400

1,308

220

425,225

1990

Units

62,658

58,331

57,347

59,379

38,976

18,036

24,198

23,220

27,270

12,910

11,200

5,089

9,055

5,129

2,609

2,175

2,873

1,200

1,116

800

423,992

Percent

Change

13.4

9.9

5.0

-13.15

•4.7

26.0

-11.1

-17.6

-34.7

-5.1

5.1

90.6

-17.4

45.4

44.0

19.8

-27.8

16.7

17.2

-72.5

0.3

1991 Market

Share (%)

0.6

0.5

0.4

0.3

0.1

100.0

16.7

15.0

14.1

12.1

8.7

5.3

5.0

4.5

4.2

2.9

2.8

2.3

1.8

1.8

0.9

For More Information . . .

On the topics in this issue Lane Mason, Director (408) 437-8120

About online access (408) 437-8576

About upcoming Dataquest conferences (408) 437-8245

About your subscription or other Dataquest publications (408) 437-8285

Via fax request (408) 437-0292

The content of this report represents our interpretation and analysis of information generally available to the public or released by responsible individuals in the subject companies, but is not guaranteed as to accuracy or completeness. It does not contain material provided to us in confidence by our clients.

Individual companies reported on and analyzed by Dataquest may be clients of this and/or other Dataquest services. This information is not furnished in connection with a sale or offer to sell securities or in connection with the solicitation of an offer to buy securities. This firm and its parent and/or their officers, stockholders, or members of their femilies may, from time to time, have a long or short position in the securities mentioned and may sell or buy such securities.

DataQuest

n n a company of l E M f The Dun SiBradsticct Corporation

Dataquest Is a registered trademark of A.C. Nielsen Company.

File inside the Dataquest Perspective binder labeled

Memories WorMxmde

®1992 Dataquest Incoiporatsd, Reproduction PnohiljitBd 0012759

\

Dataqyest

n m acompanyof

• Dataquest ^^oS^'^iS " S : ^

Perspective

Vol. 1, No. 2 December 30, 1991

Market Analysts

Those Incredible FIFOs: A Summary of Today's FIFO Business

The FIFO market is interesting. Although it totals only about $120 million, 14 suppliers offer more than 90 devices in a virtual plethora of speed grades and package types. It appears to be a difficult market to earn a living in.

By Jim Handy Page 2

Memory Cards: An Emerging and Potentially Explosive Market

The memory card market is poised for rapid growth as portable computing, electronic photography, and other applications incorporate the memory card as an enabling technology.

By Nicolas Samaras Page 11

\

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0012433

Memories Worldwide

Market Analysis

Those Incredible

FIFOs: A Sumtnaty of

Today's FIFO Business

FIFO Background

First-in/first-out (FIFO) memory devices have existed in the electronics industry for more than

20 years. Strangely enough, the companies that were the first suppliers of FIFOs are not now the dominant FIFO suppliers.

As their name implies, FIFOs are data buffering devices that output data in the same order in which they were input to the device. FIFOs can be broken into two types: register-based and memory-based. The older register-based FIFOs have been around for over two decades and are available mainly in organizations such as

(AxA and 64x5. Memory-based FIFOs were introduced by SGS-Thomson Microelectronics

(then called Mostek) in the early 1980s. Small memory-based FIFOs are usually in the range of 256x9 bits and are proposed (but not yet sampled) to become as large as 64Kx9. Today's most popular and most widely second-sourced

Figure 1

Worldwide FIFO Sales

FIFOs are the 72Qx series of memory-based

FIFOs.

Multiple FIFOs can be used for either depth or width expansion, and modules are available to support multiple FIFOs. Only monolithic FIFOs will be discussed in this article.

The FIFO Market

The estimated worldwide 1991 FIFO market is between $110 million and $120 million and is not expected to increase significantly in 1992 because overall unit growth in the FIFO market is relatively small and ASPs are falling due to the high number of participants (see Figure 1).

Despite its small size, 14 manufacturers share the market. Their standard and specialty FIFO offerings tximbine for 93 different oiganizations

(see Tables 1, 2, and 3). As if this degree of ftagmentation were not enough, it appears that other manufacturers are thinking of entering the market and that existing manufacturers are planning to introduce more sole-sourced configurations in an effort to improve ASPs by adding value,

As is true in any fragmented market, certain devices and certain manufacturers are well ahead of the rest The three major FIFO manufacturers that now dominate the market

Source: Dataquest ODecember 1991)

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-6000 / Fax (408) 437-0292

0012433

Memories Worldwide are Advanced Micro Devices Inc. (AMD),

Cypress Semiconductor Corporation, and

Integrated Device Technology Inc. (IDT). These three companies share about 85 percent of the entire market

Two kinds of devices dominate: register-based

64x4 and 64x5 FIFOs, and memory-based 256x9 to 2Kx9 organizations. Register-based FIFOs are waning in popularity. They are rarely being designed into new systems, not because their function is no longer required (more small

FIFOs are being implemented in system designs than ever before) but instead because systems designers now find it more feasible to include the functions of the small FIFO direcdy onto an application-specific integrated circuit (ASIC) instead of using a separate component. FIFO vendors seem to agree that the doUar sales of the small FIFOs are decreasing, while dollar sales of medium-depth (256x9 to 2Kx9) FIFOs are on the increase. Some believe that the crossover point in dollar sales between these two categories has been reached.

Specialty FIFOS are memory-based designs that lend themselves to a narrower set of applications by incorporating logic features such as serial-to-parallel and parallel-to-serial data stream conversion, bidirectionality, and bus matching.

Specialty FIFOs are not a large enough segment of the overall FIFO market to warrant a separate analysis.

The FIFO sell is a design-in and not something that can be pursued at the buyer's desk. Suppliers with strong direct sales forces and fieldapplication engineering teams tend to do better, both at getting FIFOs designed in and at winning designs for proprietary specialty FIFOs.

Specialty FIFOs command higher average selling prices (ASPs) and can give a supplier a lever on the sales of other components in the system.

Since most systems use a small number of

FIFOs, even high-ASP devices contribute a very small percentage of the overall system cost.

Buyers tend to focus their negotiating efforts on the devices that contribute more strongly to the system cost and, as a result, are reticent to ask their engineers either to design-out a solesourced FIFO or to qualify an alternate vendor.

Thus, the business in a particular design tends to stay -with the company that achieved the design win. For this reason, companies with highly technical direct sales forces (for example,

IDT, AMD, Cypress) tend to do better in the

FIFO business than companies with sales forces consisting mainly of sales representatives and distribution, even though companies in the latter category may be able to offer strong price, delivery, and quality incentives.

A surprisingly large portion of today's FIFO market exists in the military. One vendor estimates that the military market accounts for about one-fourth of the entire FIFO market (in dollars), perhaps because of the large number of military applications for digital signal processing (DSP), such as sonar and radar.

FIFOs are used in very diverse applications.

The function is found useful in interprocessor communication, often between a DSP chip and a controlling CPU, a combination used in end applications such as sonar and radar. In a similar vein, FIFOs are often used within customdesigned DSP processors as line buffers or similar delays, once again in radar and sonar processing; in local area networks (LANs) and other networks that are widely used in telecommunication; and in extremely high-performance graphics engines.

A more esoteric application is the FIFO's use in array processors such as those made by Floating Point Systems or Intel Scientific Computers.

FIFOs fit into any high-speed application where a steady stream of data must be adapted in flow to the fits and jerks of a differendy aligned system, or, conversely, if the fits and jerks must themselves be turned back into a steady flow. Such problems abound in LANs, bus interfaces, and laser printers. Except for the array processor application, most of these problems can be solved through the use of a single

FIFO; so most systems use very few devices.

As a result, the FIFO has a very broad but shallow market. A typical sales order for FIFOs would amount to between $10,000 and $50,000, with orders rarely reaching the $100,000 level

Everybody uses them, but most designs use just one or two. Unlike in the microprocessors industry, there is tough competition and the barriers to entry are low; therefore the average selling prices (ASPs) are modest.

Figure 1 shows estimated FIFO dollar sales for the years 1987 through 1992. Dollar growth is slow because the number of manufacturers has increased as the overall market has softened, despite a significant increase in unit sales. ASPs have dropped significandy; a device that sold for $30 in 1988 now sells for $7. This difference implies that unit volume is increasing

©1991 Dataquest Incorporated / 1290 Ridder Paric Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0012433

§ ®

K ^

^ l - l

D

f

\

I

M

1^

s

0 0 o

f

1

R-

O p.

;$

1

P

»

i

Table 1

Standard FIFOs Cross Reference

Organization

Output

Enable

Low-De osity

16x4

16x5 l6x5

Yes

Yes

Yes

16x5

64x8

64x9

Yes

Yes

256x8

256x9

64x4

64x4

64x5

64x5

64x5

64x8

64x9

Yes

Yes

No

No

Yes

Yes

Yes

Yes

Yes

Yes

VUgji

Fixed

Fixed

No

Fixed

Fixed

Fixed

Fixed

Fixed

No

No

No

No

Fixed

Fixed

Fixed

AMD

7200

67401

67Oi013

67402

67C4023

67C4033

Cypress

7C401

7C403

7C402

7C404

7C408A

7C409A

Dallas

Logic

Hitachi IDT Devices

7200

72401

72403

72402

72404

72413

L8C200

L8C401

L8C403

L8C402

L8C404

L8C413

L8C408

L8C409

Medium-Density

512x9

512x9

No

Yes

Fixed

Fixed

512x9

No

Prog

512x9

Yes

Fixed lKx9 lKx9

No

Yes

Fixed

Fixed

1KX9 lKx9

1KX18

Yes

Yes

Prog

Fixed

2Kx9

2KX9

2Kx9

2KX9

Yes

No

Yes

Yes

Yes

Prog

Fixed

Fixed

Prog

Proa

7201

4601

7202

7203

7G420

7C424

7C428

2009

2010

2011

7201

Micron

L8C201

L8C2011

52C9005

52C9007

Mosel Quality Samsung

7200

7201

7202

72021

L8C202

L8C2021

52C9010

52C9012

7202

63921 7203 L8C203

72031 L8C2031

52C9020

52C9022

7203

7201 75C01

75C101

7211

7202 75C02

7212

75C102

7203 75C03

75C103

SGS

548

549

548

549

4501 549

4502 549

4503 549

<s to

VO tv>

^

•b.

UJ

8

"V

g

1

«

8 ®

S^

"•g

I

•o

I o

T a b l e 1 ( C o n t i n u e d )

S t a n d a r d FIFOs C r o s s R e f e r e n c e

Organl- Output zatlon Enable Flags AMD Cypress Dallas Hitachi IDT

Logic

Devices

Micron Mosel Quality Samsung SGS

High-Density

4KX9 NO

Fixed 7204

4Kx9

4KX9

Yes

Yes

Fixed

Prog

8Kx9

8Kx9

8Kx9

16KX9

16KX9

32KX9

,32Kx9

No

Yes

No

No

No

No

No

Fixed 7205

Prog

Prog

Fixed

Prog

Fixed

Prog

7C432 2012 63941 7204 L8C204 52C9040 7204 7204 75C04

4504 549

72041 L8C2041 549

7C460 2013 7205

7C470

7C462 720^

7C472

7C464

7C474

52C9042

52C9082

4508 540

540

Source: Dataquest (December 1991)

Memories Worldwide

T H

0 0

R

S i n

O

SP

B t ^

m

O

99

B i ^ o

•<r in

^

in o

O) (S

M VO

o ffS

in lA o i n

•f

w

e

<M

(N

I ^ r-^

&

O l-H

<M N

-g< - g .

<N N

r^ r^

8 S

r j M

(M (M

i^ r^ o

rH

CM

(M

r^

<

1 1 i n

T-< <-l

( N ( S

(N (M

r- r^

<! o i H i n

M (M N

( S N (M

(VI CM (M

1^ r-- r^ o

fTi

CJ

CM

I ^ t H r f i

( N

( N

1 ^ i n o » i t o -"T •<»<

CM CM M

1^ i^ r^

I / '

•^

<S

r^

i n ' ^

SS

1^ t^ to to in -«> ss

1 ^ 1 ^

I III III IIIII ml III III il f

n

M

%

e

4

I a f

00 00

0 0 0 0 O S ' - i 0 0 O \ O \ » - i

00 VO 00 00 i n o O O ^ i - H f r j O O O s O s O s O s r H O O O v O V r - i s l l i l ^ - - - ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^

\ C l \ 0 \ £ i r s J C M C M t n i n i n i n < - l i - l i - l i - l r H ( V | C M ( M C M ( N C M ^ - 4 < ^ - < 3 *

©1991 Dataqpiest Incorporated / 1290 Ridder Park Drive, San Jose, <::A 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0012433

i i i

Memories Worldwide

> o

M

•O" iri

• * CO

P v OS

•fl" - ^ l A » A

1 -

SS « i i n i n

o o

I-H \ 0 r^ 1^

r^ i-H i n vo

N C^ (N (N r^ r^ r^ 1^ i n i n

M M tM

^i

1 ^ i n i-H o t o

M M OJ ( N O CO

rH m in in r-( 1^

N tN r j M N C^ r>- i-v r^ r ^ r ^ r^

M i-i r i -g-

CO < 0 ^ O

1-H rH r-t rH

M CJ <N ( S r^ 1^ 1 ^ t ^

( S

1 ^

0\

CO

I ^

>

>

B

. « .1) ^ O tt) 1) V 4 ) (U 1 ) u 1) (U

l i S i i S i S i S z ^ J ^ ^ i z ^ J i S J z z i z z z z ^ J ? :

h

IIIII!IlllfjlIIIlllllllllllII,

i S ^ ^ S S S S S S S S S & i S S S S ^ S S & S ^ ^ S S ^ S

> - Z > - i > - Z Z Z Z Z Z Z Z Z > - > - Z Z Z Z > - Z Z > ' Z > H > - i Z Z > < Z

•§•8

•6-6 ^-S-6-S-S

^ q -o p p p p

^a

•6 ^S "S "6 •§ 43 XI

C u u

>» C P

:s ?• ?J

< 1/3 tfl i tn a fft

1^

I

• 3

2 a<

^ u "53 TJ

'S

"a ^

n «

S

«

CO t>Q

^ S

S J S S I S ) S S 2 2 S ! ^

5 (L>

11

II

s

• 3 -oJ w ' S

3 O

CQ BQ

a

On

Rh P i f^ X ! t ?

(u o a

CO t a P i

OS OS

X X is

5 ^

SD (SI " „

•O X -^^^ - ^

CO X S C J « J S C 0 0 ( ] C O 0

OS X

M i n rO t-J

I ^ ^ S S H S H S c ^ i n X T - l T - H i - i r H T H r - l ^ X

M r J i A i A i n i n i n i A m N

NO 00 OO OO SD

OS <ss osi rt

^ s s s g ^ ^ ^ s

so •«" •a'

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-6000 / F a x (408) 437-0292

0012433

8 Memories Worldwide significantly but more slowly than the growth of the supply base. One mamifiacturer actually claims to have seen its dollar sales decrease from 1990 to 1991 ^i^iile unit volume almost doubled.

Because the FIFO is needed to solve speed problems that cannot be easily addressed via software buffers or less complex hardware, semiconductor manufacturers place a strong emphasis on the speed of their FIFOs. Currently, the fastest available asynchronous devices run at 66 MHz. Synchronous devices that run at

70 MHz have been recendy announced.

At these speeds, system designers have a very difficult time producing dean waveforms on the read and write input pins. To help solve this problem, vendors have designed synchronous

FIFOs that can internally synchronize relatively dirty read and write waveforms against two externally generated dock signals. FIFO manufacturers are bullish about future acceptance of synchronous FIFOs, despite the fact that synchronous static RAMs have been available for several years and have met with extremely limited acceptance.

T h e Players

The following sections profile the players in the industry.

Advanced Micro Devices Inc. (AMD)

AMD inherited a number of FIFO products through its acquisition of Monolithic Memories

Inc. in 1989. Because it has not been introducing new designs aggressively, its prospects in the market may be hurt. Still, AMD is the second-laigest supplier of FIFOs worldwide and is capable of maintaining this business through sheer size. AMD's sales force is large and good.

Surprisingly, AMD does not support the military

FIFO business, despite the company's overall commitment to the military electronics marketplace. AMD uses a Japanese foundry service to fabricate its FIFOs.

Cypress Semiconductor

Clypress has been aggressively pursuing the

FIFO market for almost five years. The company's offerings are all high-speed versions of industry-standard oiganizations. Cypress' penchant for high visibility has helped to assist its efforts at winning a growing share of this market.

Recendy, Cypress introduced 70-MHz synchronous FIFOs—the fastest in the market.

Cypress is a strong player in the military market, a market that is not as well-supported as one would expect given the number of American FIFO manufacturers. Dataquest estimates that Cypress is the fourth largest FIFO manufacturer and could well become number two by

1993, displacing AMD.

Dallas Semicondtictor Corporation

Although Dallas has a number of good FIFO products, it is not a big player in the FIFO market The company's main focus is batterysupported memory and time-keeping devices, both high-ASP lines, so FIFOs are nearly a commodity by comparison. Dallas sells most of its FIFOs through distributors.

Hitachi

Although Hitachi is a significant player in the standard static RAM business, it does not compete aggressively in FIFOs. The two devices it offers are not highly promoted in the United

States, and many of its competitors are unaware that Hitachi even partidpates in the market.

Integrated Device Technology Inc.

(IDT)

The long-term leader in the FIFO marketplace,

IDT has the broadest product offering and the dominant share of sales, accounting for about one-third of the total market. Evidence of the company's continued focus on irmovation is easily observed in the sheer number of products in Tables 1, 2, and 3. Some of the company's competitors view this focus as a strength because it ensnares new designs using high-ASP proprietary products.

IDT has done pioneering work in bidirectional

FIFOs, synchronous FIFOs, and applicationspecific products.

As with all of its other product lines, IDT pays dose attention to the military market. IDT recently lost some market share, apparendy because of delivery problems. If this trend continues, IDT could lose its first-place market standing to an astute rival. Still, the company has a strong commitment to the FIFO market, where it reaps about 20 percent of its gross sales.

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0012433

Memories Worldwide

Logic Devices Inc.

Logic Devices' current woes have affected all phases of its business. As a result, it has not really penetrated the FIFO market and is one of the least significant players in the market.

Micron Technology Inc.

w h e n Micron enters a market, its competitors suffer the consequences. The company's basic operating strategy is to enter only markets where it has a significant cost advantage and drive the price low enough to discourage competition. Other FIFO manufacturers, then, are relieved that Micron's FIFO offerings have not been put into production as early as anticipated.

Micron is about to sample the most widely sourced products, the 72Qx series of 9-bit wide

FIFOs. Dataquest expects these devices to suffer serious price erosion after Micron's entry into production.

Mosel

Although Mosel is a small fabless memory company, it has recently grown significantly with proprietary design wins in the FIFO market.

Mosel does not suffer from the lack of a fab, but rather is capable of taking advantage of some of the best semiconductor processing capability available by spreading its business among competing leading-edge fab foundry services.

Mosel's merger with Vitelic has added Vitelic's

FIFO products to Mosel's product line, but the

Vitelic versions w^ere altemate-sourced and are being discontinued.

Mosel recendy aimounced two specialty FIFOs oiganized as 64x16 and 256xl6. Mosel taigeted these devices at Intel Corporation's EISA bus controllers, indicating its desire to get into a proprietary product market and increase ASPs and account control.

Qtuilily Semiconductor Inc.

Quality is a small innovative company that has entered the FIFO market using standard devices with an eye to rapidly introducing newer valueadded designs.

Samsung Semiconductor

Samsung is poising itself to become a major force in the FIFO market as it has done in the

DRAM market. The company currently offers a limited range of devices, all of which will be altemate-sourced by Micron, and all of which, at ^ MHz, are the fastest asynchronous FIFOs in the market.

Samsung's potential success in this effort revolves around three Victors. First, Samsung is a large company that can afford to weather some losses in establishing itself as a major

FIFO supplier. Second, the company is a manufacturing powerhouse that has the ability to squeeze costs out of a design. Third, Samsung's new products have all been extremely technologically competitive.

The only potential obstacle to Samsung's eventual dominance of the FIFO market would be a lack of focus. Such a failing would be understood, as FIFO is probably the smallest market

Samsung has chosen to enter. Samsung traditionally has not been a semiconductor house that pursues the design-in of new proprietary devices; therefore it will probably try to continue to take existing business from the market innovators rather than design-in proprietary products at a higher ASP.

Samsung's strategy to employ sales representatives rather than direct salespeople will tend to hamper any design-in efforts. The company's goal is to find the big markets and win those first. However, as stated earlier, the FIFO market is broad and shallow and such a feat might be more difficult to accomplish in FIFOs than in DRAMs.

SGS

Although SGS was the inventor of the memorybased FIFO, it lost its lead in this market several years ago, allowing Monolithic Memories

(now AMD) and IDT to take over the memorybased FIFO market. Although the company does not have a broad produa offering, it is increasing its focus on the FIFO market and expects to grow sales significantly this year over last.

Sharp Microelectronics

Technology Inc.

Sharp is taking an interesting stance with its memory business. Whereas the company's standard static RAMs are being sold in the United

States mainly by third parties (National Semiconductor Inc., Mosel, and Electronic Designs

Incorporated), Sharp has chosen to sell its

FIFOs by itself.

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-SOOO / Fax (408) 437-0292

0012433

10

Memories Worldwide

Shaqj, like Samsung, has phenomenal resources at its disposal, which allows it to withstand narrow margins in gaining market share and probably allows it to boast of a very low die cost. These factors could help push Sharp rapidly to a higher ranking in market share than its current position of number five. Still, this would require a design-in effort, which would require some restructuring of Sharp's field sales force.

Texas Instruments Inc. (TI)

TI apppears to have put itself into the enviable position of identifying business that eludes its competitors. Most FIFO manufacturers believe the company to have about one-third the sales

Dataquest has uncovered. The higher figure is possible because TI's graphics processor and

DSP strengths offer a total system solution for a customer's DSP and graphics needs. TI is also strong in LAN components, another major FIFO market. Its product offerings consist mainly of shallow FIFOs; the evidence is that TI's part numbers start with the "74ACT" prefix used for its SSI and MSI logic families, and that the bulk of its offering is in low-density parts.

Competitors believe that TI has lost focus on the business, but this looks to be the case only because of T['s long-term emphasis on older designs. Until 1990, TI shipped only bipolar

FIFOs; it only recendy implemented FIFOs in

MOS. The company does not currently offer any members of the 720x series of devices, the most widely sourced FIFOs today. TI is developing a broad new range of devices, many of which are synchronous, and will be using its advanced SSOP packaging technology to its advantage. TI also has innovated nonmetastable flags, a vexing problem to other

FIFO suppliers, and will use this to offer more reliable operation to system designers.

United Microelectronics

Corporation (UMC)

UMC is a Taiwan-based company that specializes in slow ROMs, slow SRAMs, and chip sets.

It is currently moving into the more lucrative fast memory marketplace. Although UMC offers

FIFOs, it is not an important competitive factor in the market.

Vitettc Corporation

Since the merger with Mosel, Vitelic has deemphasized its FIFO offering in deference to

Mosel. Vitelic is no longer manufacturing its

FIFOs and is selling off its inventory. Even combined with the sales of Mosel, Vitelic's sales do not put it into the top five FIFO suppliers.

However, Mosel's share of the market is growing quickly and the company could rapidly gain si^iificant market share if it can keep up its current pace.

Dataquest Perspective

Tomorrow's FIFO Market

Several factors could affect the FIFO market over the next few years. First, unit volume is ejqjected to grow about 30 percent per year.

Second, the number of competitors is too large for the size of the market. Subsequendy, ASPs have come under considerable pressure. Should the 14 vendors shown split the market equally, each would get less than $10 million in sales.

Third, the product offering is large, requiring vendors to produce and inventory a remarkable breadth of produa types. (Each device listed in

Tables 1, 2, and 3 is offered in numerous speed grades, and most are offered in more than two package types.)

Because most FIFO manufacturers also manufacture fast SRAMs, Dataquest expects to see continued rapid price erosion over the long term as manufacturers more aggressively pursue their

FIFO business to make up for disappointing performance in the fast SRAM business. Specialty FIFOs are not expected to become a mainstay but will be neglected in favor of multiple-sourced devices at more competitive prices. Given this scenario, we expect a significant share of the commodity synchronous and asynchronous FIFO market to be taken over by companies with strong low-cost manufacturing prowess, unless they determine that the market is too small and fragmented to pursue, while companies with direct sales forces try to compensate for ASP erosion by focusing their eflForts on achieving design wins for specialty

FIFOs.

Not only are FIFO marketers bullish about the future of synchronous parts, as mentioned earlier, but they also believe that tomorrow's offerings will include significandy faster parts (100-

MHz and faster), wider parts (in widths of Xl6,

Xl8, and x36), and new packaging technology that will allow the wider parts to compete favorably with traditional parts in board space

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0012433

Memories Worldwide

11 consumption. Depths, too, will continue to grow to match the available SRAM technology.

Memory cards perform a function similar to that of a floppy disk. They store binary data.

Still, the market is fiill of innovators trying to increase their ASPs through the rich feature sets of specialty FIFOs. It is Dataquest's opinion that these innovators could lose significant market share to the production houses (that is, Micron,

Sharp, and Samsung) should their focus become too diverted from the top-selling standard products. •

As program or data storage media, memory cards are not new. They have been used in computer games, point-of-sale (POS) systems, photocopiers, and laser printers. More recentiy, electronic organizers such as the Casio BOSS and the Sharp Wizard along with palmtop PCs such as the Poqet and the HP 95LX have begun using memory cards for data storage.

Figure 2 shows their application in portable

PCs.

By Jim Handy

Memory Cards: An

Emerging and Potentially Explosive Market

The memory card form factor has not changed much over time, but the type of edge connector and the electrical/mechanical interface have.

The edge connector of a memory card is the conduit that allows data to move to and from the card's memory IC^. It defines the card's capabilities. To date, we have seen cards with a variety of connectors including 38-, 40-, 50-, and 60-pin.

"What Are Memory Cards?

A memory card is a portable semiconductor storage device that contains memory ICs. It resembles a thick credit card (3.3mm) with an edge connector at one end (see Figure 1).

Figure 1

Example of Memory Card

Memory Card Varieties

Memory cards contain mostly semiconductor memory ICs that belong to one of the following families: mask ROM, EPROM, OTP,

SRAM, DRAM, EEPROM, and flash. DRAM memory cards are relative newcomers and are meant to be used as "extended/expanded" memory with no need for battery backup. SRAM cards with battery backup have been used as solidstate "floppies" in the current generation of electronic organizers. Until recendy, SRAM cards

(with battery backup) were the only nonvolatile memory cards. Flash memory cards today provide a promising alternative. Items such as language translating software and dictionaries typically come in mask ROM cards, as they are the most dense and least expensive. Functionally, they are huge look-up data tables that need no chajige. Table 1 lists the various memory card alternatives.

Source: Panasonic Industrial Company

Memory Card Applications

Memory card applications include the following:

• Personal computers

• Factory automation

• Instrumentation and testing

• Avionics

• POS terminals

• Musical equipment

• Medical instrumentation

©1991 Dataquest Incorporated / 1290 Ridder Paik Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0012433

12

Figure 2

Memoiy Card Usage in a Portable PC

Memories Worldwide

Source: Intel Coipoialiaa

Table 1

Memory Card Alternatives

Type

ROM

EPROWOTP

DRAM

SRAM

EEPROM

Flash

Source: Dataquest (December 1991)

Density

128KB—16MB

128KB—8MB

64KB—12MB

32KB—4MB

8KB—512KB

128KB—4MB

On Standards

what inhibited memory card growth in the past was the lack of standards. In June 1989, the

Personal Computer Memory Card Industry

Association (PCMCIA) was formed in the United

States, with a broad-based membership that included semiconductor companies along with software and hardware vendors. The P(]M(3A's originally stated goal was to establish a standard for memory cards used with DOSbased PCs. It succeeded rather quickly as standards go. The first revision of a memory card standard was published in August 1990.

Revision 1.0 of the POICIA/Japan Electronic

Industry Development Association (PEIDA) standard defined the following:

• Tlie form factor—a device the size of a credit card, 3.3inm thick with a 68-pin socket coimector

• The interface—^parallel type bus, 8-bit/l6-bit

• The address space—64Mb

The PCMCIA worked closely with the JETDA and JEDEC. This dose cooperation enabled the prompt international acceptance of the standard.

Revision 2.0, as announced in September, addresses XIP (eXecute-In-Place) and I/O functions such as modems and LANs for PCMCIA bus cards. Intel Corporation also announced the

Exchangeable Card Architecture (ExCA), a hardware and software implementation of the

PCMCIA Revision 2.0 system interface. It is

Intel's stated intention to make ExCA an industry standard so that different types of cards

(memory, LAN, modem, and wireless communications) from different manxifactuiers will be interoperable.

Do Memory Cards Replace Hard

Disks?

Stiicdy speaking, memory cards are not hard disk replacements. Rotating media have not been terribly successful with removable hard disks. A number of companies have tried that approach, but technology and costs kept it out of the mainstream. Thus, after a decade of using PCs, we are conditioned to think of hard disks as storage devices that belong inside the

PC enclosure. This idea is a technologydependent perception, and there is no reason why it should be so. On the other hand, memory cards, being a solid-state storage medium, are removable and portable. At a density of 20Mb, is a memory card acting like a "removable hard disk"? We believe that it is.

©1991 Daaquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0012433

Memories Worldwide 13

Memory cards have the following advantages over floppy/hard disks:

• Faster access and transfer rates

• Space, power, and weight reduction

• More ruggedness

However, they do have the following disadvantages:

• Expensive

• Lower capacity

The Cost IssueHow Important

Is It?

In 1991, the average selliag price (ASP) of a

2.5-inch 40MB hard disk drive was $250.00, which translates to $6.25 per megabyte. The

3.5-inch floppy cost is close to $1.00 per megabyte. By comparison, a 1MB flash card costs approximately $300.00 or $300.00 per megabyte—a substantial disparity! Semiconductor memory certainly costs more.

The question is, "Can you put a floppy disk drive in a palmtop PC to take advantage of that cost disparity?" The answer is, "No." There

Figure 3

Pen-Based and Hand-Help PC Forecast

Millions of Units

12-

10

K V ] Pen-Based

Hand-Held is not enough power (or space). The issue,

then, is not cost. Here the removable storage medium dictates the product's capabilities and its success or failure in the marke^lace. Without a memory card, a palmtop is nothing more than an electronic organizer. It is the memory card that transforms a palmtop into a fullfledged personal computer.

The Memory Card Market

As with any emerging technology, market size projections are difficult at best. The following assumptions may be used to gauge a portion of the total available market:

• The majority of hand-held PCs will use memory cards (80 to 95 percent).

• A portion of pen-based PCs will use memory cards (50 to 80 percent).

• Notebook PCs are forecast to grow from

686,000 units in 1991 to 7 million by 1995.

A portion will use memory cards (10 to

20 percent).

Figure 3 provides some useful boundary conditions. Dataquest expects worldwide shipments of pen-based VCs to grow at a compound

4 -

2-

1.559

0-1-

,503

.096

1391

Source: Dataquest (December 1991)

1992 1993

3.087

3.498

1994

5.588

5.422

• \ x

I

f

i *

h'A^

1995

©1991 Dataquest Incorporated / 1290 Bidder Park Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0012433

14

Memories Worldwide annual growth rate (CAGR) of 174 percent, from 96,000 units in 1991 to nearly 5.5 million units in 1995. At the same time, hand-held PC shipments will grow at a 108 percent CAGR from 503,000 units in 1991 to approximately

9.4 million in 1995. Together they amount to approximately 600,000 units in 1991, growing to almost 15 million by 1995. Some simple assumptions on memory card average selling prices indicate that this could easily become a billion-dollar-plus market by 1995.

Memory cards used in non-PC applications

(which may account for as high as 90 percent of total memory card shipments in 1991 and

40 to 60 percent by 1995) are not included in this discussion. Electronic still photography alone may provide an explosive market for memory cards.

What Are the Key Developments

Needed for Memory Cards to

Succeed?

Three developments are necessary for the success of memory cards. These developments and the applications where they are needed are as follows:

• Cost reduction—all applications

• Development of data-compression ICs— electronics "filmless" still photography and PCs

• XIP—^palmtop PCs

Cost Reduction

Flash memory cards hold the promise for becoming the least expensive form of solid-state storage. From a cell standpoint, flash rivals that of DRAM. Unlike DRAM or SRAM, it is nonvolatile, which means there is no need for battery backup. The need for bulk erasing of current-generation flash ICs creates a problem that requires clever solutions. With SRAM or

DRAM cards, a single byte can be erased;

EPROM-derived flash most often can be erased at the chip level Ci.e., the whole chip). Recently, some vendors have armounced products that allow erasure of particular memory segments. A prime example is the Intel 28F001BX 1Mb flash memory, which is segmented into areas of one

8KB, two 4KB, and one 112KB—all of which can be independendy erased and programmed.

EEPROM-derived flash is far more flexible at a cost premium (larger die). Flash EEPROM cells are laiger than flash EPROM. Mask ROM memory cards will be the least expensive for the foreseeable future.

Data Compression ICs

Data compression ICs represent a key development for the electronic photography market and, to a lesser extent, for palmtop and penbased PCs. Data compression ICs will be the subenabling technology devices. Without them, the future of electronic photography is in doubt. Thirty-six exposures (pictures) can be stored in a 2MB flash memory card in compressed form. If no compression were used,

40MB would be needed!

XIP

Simply stated, XIP allows a memory card to

"plug-and-play." That is, once the card is plugged into the PC, program execution begins much in the way a program runs after one types in the program name and hits carriage return. That procedure is in contrast with current-generation PC architectures that need to copy the program code from secondary storage

(hard disk or floppy) to main memory (DRAM) before execution. A palmtop PC with XIP capability needs just a single copy of a program, usually stored in the memory card, thus freeing up main memory.

The PlayersSolid-State Disks

A number of companies are working on solidstate disk (SSD) replacement—a challenging task, to say the least. SunDisk Incorporated, located in Santa Clara, California, chose to focus primarily on hard disk replacement (solidstate disk) with a proprietary flash memory technology and architecture. The venturecapital-funded start-up launched three SSD products recentiy, all aimed at pen-based and pakntop PCs. The 2.5/5/lOMB SSD plug-andplay subsystems come with an IDE industrystandard interface. The company is producing a

20MB solid-state disk subsystem on two

PCMC2A form factor cards and expects to offer

40MB capacity shordy.

Toshiba announced a 4MB 5V EEPROM IC

(TC58400) that is aimed at the SSD market. This device is by far the most dense EEPROM introduced to date. Architecturally, it is organized in a way that should facilitate SSD implementations. Toshiba uses a NAND cell structure that is 70 percent of its 4Mb DRAM cell; it is manufactured using a 0.7-micron double-poly CMOS process. The die size is

58.55mm'.

Hitachi announced a 5.25-inch form factor SSD based on 4Mb DRAM technology. This p r o d u a is taigeted at CAD/CAM., imaging, and graphics

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0012433

Memories Worldwide

15

systems that demand a higher I/O throughput than what hard disk drives provide. The Hitachi

SSD has access time of 0.35nis, incorporates a

SCSI interface, and comes in 32MB or 64MB PC boards. The SSDs may be expanded to a capacity of 320MB. The data can be protected from power failures by using an optional batterypowered backup hard disk drive.

The PlayersMemory Cards

Table 2 lists some of the companies active in the memory card market and their products.

Other companies include Datakey and riT-

Cannon.

AUemate TechnologiesFRAM,

novRAM

At least two different technologies may be used in future SSD and memory card implementations, assuming that they become cost competitive. Both of those technologies are nonvolatile

(that is, need no battery to retain data) and are easily reprogrammable. FRAM (Ferroelectric

RAM) devices are now becoming available from

Ramtron International Corporation of Colorado

Springs, Colorado. At this point, the 4Kb and

16Kb production offerings may find only limited use in memory cards and SSDs. However Ramtron is working on 64Kb and 256Kb devices and hopes to offer 4Mb densities by 1995.

From a technology standpoint, ferroelectric devices have the potential of reaching densities similar to those of DRAM, The other alternative—^novRAM—was, until recentiy, available in low densities (256 bits to 8Kb). However,

Simtek Corporation, also of Colorado Springs, has demonstrated that it is possible to substantially increase novRAM densities. The company offers 64Kb devices now and plans to introduce

256Kb and 1Mb products in the future. A novRAM is essentially a combination of SRAM and EEPROM. Every SRAM bit has a corresponding EEPROM bit that is used to store the information when power is removed.

Because the SRAM section of the device is used during normal operation, high-speed (30ns) read/write is available. However, the resulting die is larger than either an SRAM or an

EEPROM device of the same density.

Some Thoughts o n the Future of

Memory Cards and PCs

in the past, the computer was the expensive component and the storage medium (floppy disk) the inexpensive one. We've become accustomed to that oddity and do not seem to question it. However, the computer is just a

Table 2

Memory Card Offerings

Toshiba

Intel

Mitsubishi

Fujitsu

Oki

Rohm

Epson

Maxell

Fujisoku

Panasonic

DuPoot

8KB to 64KB

128KB to 1MB

128KB to 4MB

64KB to 512KB

64KB to 256KB

256JCB to 1MB

128JCB to 1MB

1Mb to 8MB

64KB to 1MB

256KB to 1MB

256KB to 2MB

1MB to 4MB to 4MB

512KB to 4MB to 512KB to 4MB to 8MB

256KB to 2MB

128KB to 1MB

256KB to 2MB

256KB to 8MB

128KB to 4MB

1MB to 4MB

256KB to 2MB

64KB to 512KB

128KB to 192KB

512KB to 16MB

256KB to 4MB

64KB to 512KB

16KB to 128KB

256KB to 1MB

256KB to 2MB

512KB to 16MB

256KB to 2MB

64KB to 2MB

512KB to 4MB

1MB to 8MB

128KB to 4MB

32KB to 1MB

512KB to 3MB

512KB to 6MB

128KB to 2MB

32KB to 1MB

* by 16 oiganizatioii

Source: Dataquest (December 1991)

OTP mask ROM flash

SRAM

OTP mask ROM flash

SRAM

OTP mask ROM flash*

SRAM*

EEPROM* flash

SRAM

DTP mask ROM* flash flash

SRAM

EEPROM mask ROM flash

SRAM

EEPROM

EPROM

OTP* mask ROM*

SRAM*

EEPROM*

EPROM*

OTP* mask ROM*

SRAM

EPROM

OTP mask ROM flash

SRAM

EEPROM

OTP mask ROM

SRAM

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0012433

16 Memories Worldwide

machine that manipulates infonnation. It is the infonnation that is important and valuable, not the machine that manipulates it. So perhaps it is fitting that the infonnation carrier, a memory card, may cost more than the computer it is attached to. In the future, we will be using platforms (palmtop PCs) that cost much less than the storage media (memory cards) they use. Imagine a $50 PC attached to a $100 memory card. At least losing the PC will not be a problem anymore!

Dataquest Perspective

Dataquest believes that memory cards represent an important enabling technology. They have the potential to transform still photography and to make the 35mm film and cameras that use it obsolete. In the process, they will change that industry and provide tremendous opportunities for growth in the consumer electronics market.

Memory cards will not eliminate rotating magnetic media any time soon. Instead, they will selectively replace them only when and where it makes sense. The bulk of the memory card growth will not come at the expense of rotating media. Growth will come from the creation of new markets. This should be good news for the semiconductor memory industry.

Ultimately, we believe, memory cards may revolutionize portable PCs by enabling them to become smaller, more rugged, lighter, faster, and perhaps user-friendly in a way that appeals to the vast majority of people who at present have no use for them. In doing so, memory cards may be the enabling technology that will make the PC of the future a true consumer item. •

By Nicolas Samaras

In Future Issues

Look for articles on the following topics in future issues of Memories Worldwide Dataquest

Perspective:

• 16MB Update

• Processor-specific SRAMs

• EEPROMs

i i

For More Information . . .

On the topics in this issue Memories Worldwide (408) 437-8228

About other Dataquest publications Sales (408) 437-8250

About upcoming Dataquest conferences Conferences (408) 437-8245

About your subscription Customer Service (408) 437-8250

Via fax request Fax (408) 437-0292

The content of this r^>ort represents our inter[»etation and analysis of information genosdly available to the public or released by reqx)nsible individuals In the subject companies, but is not guaranteed as to accuracy or completeness. It does not contain material provided to us in confid«ioe by our clients. Individual con^xuies reported on and analyzed by

Dataquest may be dients of this and/or other Dataquest services. This information is not furnished in coimection with a sale or ofSsr to sell securities or in connection -with the solicitation of an offer to buy securities. This firm and itspe(centand/oTtheirofficers,stocldioIders, or merhbers of their families may, from time totinw, have a long or short position in the securities mentioned and may sell or buy sucb securities.

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0012433

Dataqyest

w n 3,5om^nx>^

The Dun & Bradstrcct Corporation

Dataquest

Perspective

co^^

bo Not Remove

Memories

Worldwide

Vol. 1, No. 1 November 25, 1991

Market Analysts

Major Changes to Occur in the Memory Market

Major changes in the memory market will create opportunities for those who read the trends correcdy and adjust accordingly. This article analyzes DRAM changes and opportunities in particular

By Sam Young Page 2

Technoiouy Analysis

Revolutionary Pinouts: Bane or Bounty?

The static RAM market is on the verge of embarking on a new pinout standard—or is i6

JEDEC's "revolutionary" pinouts, which place the power and ground pins at the center, rather than the comers of the package, appear to have a rough batde ahead before becoming the industry standard for fast SRAM designs.

By Jim Handy Page 6

©1991 Dataquest Incorporated / 1290 Bidder Park Drive,

San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0011614

Memories Worldwide

Market Analysis

Maj€)r Changes to

Occur in the Memory

Market

Changes in the Memory Business

As we move forward into the 1990s, the memory business will see several significant changes.

In static random-access memory (SRAM), we will see a continuing increase in the use of high-speed SRAMs—primarily in cache memory applications. Processor-specific features in 16-bitwide and even 32-bit-wide parts will allow processors with cache systems to mn at ever higher frequencies. Very low power SRAMs will come into use in portable computers, and a reduction of the power supply voltage to 3V will extend battery life in portable systems.

In nonvolatile memory, we will see Flash memories emerge as the fastest growth segment.

Revenue will grow to over $1.5 billion by 1995.

Flash memory sold in the form of memory cards compatible with the Personal Computer

Memory Card Industry Association (PCMCIA) standards will replace magnetic storage in some portable applications. These memory cards,

Figure 1

MOS Memory Product Forecast—Revenue

Bllltons of Dollars

15-

5 DRAM Qslow

SRAM

U Fast

SRAM about the size of a credit card, will allow companies to share portable computers among several people without losing any security or privacy because the removable cards will contain the data. Size, weight, ruggedness, and battery life are other obvious advantages for this technology.

Dynamic random-access memory (DRAM), which is by far the largest dollar component of memory (see Figure 1), will undergo the greatest changes. For example, 40 percent of 1995

DRAM revenue will come from produas that are just now sampling from most vendors.

These changes will create opportunities for those who read the trends correaly and adjust accordingly. The focus of this article is on several of the major changes occurring in

DRAMS in the 1990s.

DRAM Trends

The first and most significant trend change is a shift in the word width for DRAMs. In the

1970s, DRAMs were offered in a 1-bit-wide configuration. During the 1980s, a 4-bit-wide configuration became popular. During the 1990s,

8-bit, 9-bit, l6-bit, and 18-bit parts will become major factors for the DRAM production total available market (TAM). Dataquest will provide a detailed analysis of wideword DRAMS in the first half of 1992. For the scope of this

EPROM Q ROM G EEPROM • Flash

10-

{

\

1968 1989

Source: Dataquest (November 1991)

1990 1991 1992 1933 1994 1995

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0011614

I

Memories Worldwide discussion, ho^vever, we will use a general analysis to show the magnitude of the change.

Dataquest forecasts that, on a worldwide basis,

40 percent of the DRAM revenue, 35 percent of the DRAM units, and 40 percent of the DRAM bits will be shipped in configurations of 8 bits or greater by 1995. We expect regional differences in the product mix. The calculations and assumptions broken down by DRAM density are show^n in Table 1.

The following questions are often asked regarding wideword memories:

• Why wideword?

• What are the price premiums for each configuration?

• Which configuration will be most popular?

Table 1

Wideword General Analysis

Wideword DRAMs Reduce

Power Dissipation

Power dissipation savings is the most significant reason for using wideword DRAMs. A typical

4-megabit (Mb) DRAM has an active power dissipation specification of approximately 550 milliwatts (Mw) and a standby specification of

5.5Mw, representing a factor of over 100 in difference between the two specifications. The first obvious question is: So what? WeU, if we look at a basic 32-bit-wide system and do some math, we see that if we built this system out of

4-bit-wide memories, the power dissipation would be 550Mw x 8 (which equals 4.4 watts) because each chip must be selected for every active cycle. Using an 8-bit part, 4 chips are active and 4 chips are in standby. This scenario results in a power dissipation of 550Mw x

5.5Mw X 4, which equals 2.22 watts. If a l6-bit

Wideword Analysis—Bits

I>ensity

256K

1Mb

4Mb

16Mb

64Mb

Total Bits

% Total Bits

Value

High

0

47,186

1,491,075

2,558,526

4,027

4,100,813

47.22

Density

256K

1Mb

4Mb

16Mb

64Mb

Total Units

% Toul Units

Density

256K

1Mb

4Mb

16Mb

64Mb

Total $

% Total $

Source: Dataquest (November 1991)

10,486

235,930

3,313,500

5,117,051

6,711

8,683,678

100

0

10

35

40

50

0

20

45

50

60

0

23,593

1,159.725

2,046,820

3,356

3,233,494

37.24

Wideword Analysis—Units

Value

Units (M) Low High

Low

40.0 0 0 0

225.0

790.0

305.0

0.1

1,360.1

100

10

35

40

50

20

45

50

60

23

277

122

0

421

30.96

Wideword Analj^sls—Dollars

\ ^ u e

Total $ Low High

Low

84 0 0 0

878

6,241

6,558

30

10

35

40

50

20

45

50

60

88

2,184

2,623

0

13,791

100

4,895

35.50

Value

High

0

176

2,808

3,279

0

6,263

45.41

Value

High

0

45

356

153

0

553

40.66

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0011614

Memories Worldwide part is used, then the calculation becomes 2 x

550 + 5 5 X 6, which equals 1.13 watts. Dataquest recognizes that this analysis is a gross oversimplification neglecting refresh currents and the fact that all possible memory cycle time slots are not used; however, the point being made is still quite valid. In summary, widew^ord DRAMs save po-wer. Even if a wideword DRAM's power specification should have to go up (a point ignored above for simplification), the results would not change sigriifi candy.

Wideword DRAMs Increase

Memory Modularity

Wideword DRAMs also allow more modularity in the DRAM size. What this means is that the next-generation memory density can be used even if the memory size does not want to increase. In the case of the lMbx4 DRAM, if we need 32 bits, the smallest memory size is

4 megabytes. This memory size is about right for 386 systems using Windows software. Until recently the average memory size shipped was only 1 megabyte. If 16Mb DRAMs are used in a x4 configuration, the smallest memory size is l6 megabytes—definitely too large for most of the personal computers available today or for the next two or three years. Most PC suppliers also prefer to keep the entry system cost down and tiierefore rarely load the box with large amounts of memory. The current single in-line memory module (SIMM) technology allows for very easy upgrade by the user. For portable computers, the same philosophy should occur with possibly memory cards being the add-on memory vehicle.

DRAM Price Premiums

A frequent question is: What will be the premium for wideword memory? The evasive answer is: Whatever the market will bear. As you might guess, this answer does not go over well. In truth, the market is currentiy "feeling out" the correct price. Several manufacturers surveyed by Dataquest do not have a clear answer.

In deriving a "rough" estimate, several factors must be considered. A wideword DRAM die costs more to build. Input/output pins require, on the die, an input and output buffer as well as bonding pads. Both of these take up significant die area. As we all know—the laiger the die, the higher the cost. The package used is also larger, requiring more bond operations.

Highly significant is the fact that the volume is lo-wer, greatiy impacting cost. Testing is also an issue, particularly in l6-bit-wide parts.

Dataquest's Price Estimates

Relative to a standard x4 DRAM, the x8 will initially cost 1.15 to 1.30 percent more. Within one year after volume production, the premium will drop to between 1.05 and 1.15 percent, with a price nearer to the lower end being the more likely scenario. The x l 6 initially will cost

1.25 to 1.35 percent more. Within one year after volume production, the premium will drop to between 1.15 and 1.25 percent, with a price nearer to the lower end being the more likely scenario.

Which Configuration Will Be the

Most Popular?

For the next several years, Dataquest forecasts that the 8-bit-wide will win. TTie reasons are as follows:

• Price—^The xB configuration wins.

• Packing density—^The x8's smaller package will make it more attractive.

• Availability—^The x8 will cause far less manufacturing and design problems and will be more readily available.

• Convenience—^The x8 solves most issues creating the need for wideword memory with the least amount of pain for both users and manufacturers.

Life Cycles Are Increasing

Historically, each new generation of DRAM occurs every three to four years. During the

1990s, this trend will not change. What will change is the slope of the edges and the peak value. The 256K DRAM—which peaked in 1988 at 956 million units—is the highest peak volume part the industry will see. In the 1990s, the next generation will take longer to reach maximum unit volume and then continue in production for a longer period of time. Figure 2 graphically displays this point.

The 4Mb DRAM is the first generation where the time to ramp into volume production has increased. This DRAM has increased volume slower than most vendors would have preferred. The factors that previously drove acceptance of the next-generation part were the following:

• Price

• Density

• Power dissipation

• Reliability improvement

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0011614

Memories Worldwide

Figure 2

Product Life Cycle—Units

Millions of Units

900-

800-

J^ ^ • • ' ' ^ ^ ^ ^ ^ • ' ^ ' ^ X

700-

600-

500-

400-

300-

^

#\A\ \ fir.

V V \ ^

Mr/ ' \ t

^ ^ N

\

r ^ X X \

•W ' '' ^ ^ ^ f t b t

M^ j^ ^, -'*.'^'- v ^ ^. ^L ^ ^ jff^ ij. 200-

' ' • ^ • i i ^ ^ ^

100-

x * ^

•'fj ^ ^ n T i ^ n n i i i i i ?

Q

...^,'>i!^r\'.<^-^-"

1 1 1

1 3 5 7 9

11

Source: Dataquest (November 1991)

^ S 16K

^ B &*K

H 2 S 6 K t i i j 1Mb

• 1 4Mb

^ S 16Mb

E S 64Mb fc^^^^ ZSGMb

13 15

In previous generations, volume ramp of a new generation would occur when the unit price reached approximately five times the price of the previous unit price. Today that requirement appears to be four times or less. We can explain the change from two directions. At this time, the personal computer accounts for approximately 47 percent of all DRAM sales.

The PC manufacturer is under incredible cost pressure and therefore will not increase cost unless absolutely necessary. The desktop PC has adequate room for all the DRAM required using

1Mb technology mounted on SIMMs. Power dissipation is also not a major issue when compared with cost. Because the part count is low, the reliability issue is also not a factor. For the desktop PC, therefore, the main factor is cost.

The workstation, mainframe, and minicomputer segments do have other motivations than cost, but they too are under far more price pressure than in previous days, and their demand represents a much smaller part of the TAM. In the first half of 1991, most of the 4Mb production shipments were going to this segment, but the volume was inadequate to meet expectations of the suppliers. In the future, portable computers will emerge that definitely care about power and density. However, concern still exists about shipping minimum-configuration systems.

Here come the memory card solutions,

The tail end of the life cycle is increased by two primary factors. First, not all equipment will require the memory size dictated by the next-generation memory device. Low-end PCs are an example. In Europe, the telecom industry absorbed large numbers of lower-density

DRAMs long after computer manufacturers phased down. Also, because of the huge investments required to stay in the DRAM business, an extension of life cycles is necessary to recoup investment.

Dataquest Perspective

The technical challenge for each new generation is increasing. It is becoming more and more difficult to cost-effectively bring the next generation to market. The 16Mb DRAM suppliers are motivated to bring prototypes and qualification units to market as early as possible, in response to their customers' desire to cut back on the number of suppliers. It is therefore very advantageous to deliver early.

The early samples do not necessarily pull in the volume production capability, however. In

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0011614

Memories Worldwide

April 1991, vendors were bullish about increasing production on 16Mb devices; most today have pushed ramp plans out by three to six months. One question often asked is: Will the

4Mb DRAM have a short life cycle because of the 16Mb? Dataquest believes that the 4Mb will have a full life cycle.

By Sam Young

Technology Analysis

Revolutionary SRAM

Pinouts: Bane or

Bounty?

The Problem

Speed and Edge Rates

High-speed SRAM manufacturers are locked in a never-ending battie among themselves to produce the fastest produas in the industry. Today, some vendors are offering 64Kb SRAMs in the

8ns range, 256Kb parts that operate as fast as

10ns, and 1Mb devices in the 15-to-20ns range.

As a general rule, the outputs of this speed of transistor-transistor logic Crn.)-compatible device, in order to be useful at the fastestpossible access time, would have to exhibit rise and fall times of about 10 percent of the access time or 0.8ns for an 8ns SRAM. The bandwidth required to support such an output signal works out to a frequency of 625 MHz (a fseriod of one rise time plus one fall time or 1.6ns).

DIP Package Inductance

Figure 1 illustrates an "evolutionary" pinout for the

256KX4

SRAM. The power is supplied on pins 14 and 28 in the corners of the package.

In order to reduce the amount of the die consumed by wide ground traces, the output pins, where most of the ground current is produced, are located close to the ground pin. This configuration was developed early in the history of memory devices and has continually been modified in only the slightest manner in order to support increases in memory sizes brought about by semiconductor technology advances; thus, it is called the evolutionary pinout.

In time, the package size used for corner power-ground devices has grown from a 14-pin,

300-niil wide dual in-line package (DIP) through a 52-pin, 600-mil DIP. Although some microprocessors have even been supplied in 900-mil

DIPs, standard SRAMs have only grown as lai^e as 32 pins with a 600-mil width. With this package growth come two problems. First, because the package continues to increase in size, the power and ground pins can get farther away from the die, increasing any parasitic elements inherent to the package's lead frame.

Second, at higher speeds, no matter how long or short the lead frame is, a single bonding wire to ground has parasitic elements, which hinder the high-frequency switching capability of the chip within the DIP.

The most important parasitic element in this path is inductance. At the 625-MHz frequency previously mentioned, every nanohenry of induaance presents an impedance nearly equivalent to that of a 4-Ohm resistor.

Ground Impedance and Ground

Bounce

A typical DIP package can exhibit between

5 and 10 nanohenrys of inductance on the ground lead. As just mentioned, this inductance comes from two sources—the package lead frame and the bonding wire from the lead frame to the die (as shown in Figure 2). The inductance of these elements has been neglected in the past because of its minor importance at frequencies below 10 MHz. At high frequencies (i.e., very fast edge rates) a

5nH inductance can cause sizable problems.

Looking at Figure 3, we can visualize the scenario for ground bounce. Several I/O pins on the device can move from a logic high level to a logic low level at the same time. When this occurs, the node capacitance (which includes the capacitance of the printed circuit board trace, all of the parasitic capacitances of the driving outputs, all the pins attached to the node, and the gates of any MOS inputs on the node) discharges through the device's ground pin. Because there is inductance between the chip ground bus and the PC board's ground plane, the di/dt (current suige of the discharging capacitance) of this sudden change causes the on-chip ground voltage to raise significantly above the ground reference on the circuit card.

The sudden rise in on-chip ground voltage causes all of the chip's input thresholds to move up correspondingly, possibly causing certain input levels to become redefined to be

"zeros" where they were previously read as

"ones."

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0011614

i i

Memories Worldwide

Figure 1

2 5 6 K X 4 Evolutionary Pinout

A

A

A

A

A

A

A

A

A

A

A

i,s

G

vss

1

2

3

4

5

6

7

a s

c

' c c c c c

L

C

C

2 8

PIN

DIP

10

11

12

13

14

c c c c c

TOP VIEW

D

5

D

5

3

J

1

3

J

J

3

J

3

3

28

VCC

27

A

26

A

25 A

24

A

23

22

21

20

19

18

DQ

DQ

A

A

A

NC

17 DQ

16 DQ

15

W

Source: JEDEC

Impact of Ground Bounce on

Overall SRAM Speed

With the inputs appearing unsettled, as in the scenario just p>osed, the outputs cannot settle down themselves. Any of the 20 or so stages within an SRAM design can misinterpret the input value during ground bounce. A single misinterpretation could cause a delay lasting until the ground reference bottomed out, but if the change on that input causes the output to change state, then the ground current will again change, and in the worst case, the component will break into oscillation. Real life lies somewhere between the worst case and a single threshold-crossing, and multiple ground bounces often occur, severely impacting a stage's settling time. That settling adds itself to the RAM's access time, because the RAM's outputs are not considered valid until they have completely stabilized. The fastest SRAM will exhibit a much slower access time in an environment with ground bounce than would be possible in a completely bounce-free environment.

History

The Comer Power-Ground Tradition

Ever since the days of small-scale integration, a convention has been followed to put the power and ground pins of logic devices on diagonally opposite sides of DIP packages. This makes sense because the distance between these pins reduces the possibility that a power to ground short will develop, and because the signal lines can be routed in such a way that they are never required to cross either a power or ground bus. Corner power-ground arrangements can save considerable layout effort, space, and cost in the design of single-sided or two-sided

PC boards. Corner power-ground pins on TTL logic devices were probably suggested by PC board layout personnel.

When the first memory devices were introduced, they did not adopt the corner powerground standard. A problem arose when 4Kb

DRAM memories started being shipped in l6-,

18-, and 22-pin packages. In 1973, this chaos brought about an effort by Sam Young of

Burroughs Computer (now a DRAM analyst at

Dataquest) to work with existing semiconductor memory suppliers to define comer powerground memory pinouts so that the advantages of the corner power-ground standard could be used in memory systems. Memory suppliers also worked to put in place a new convention in which sockets would be configured to be upgraded, allowing laiger memories to be plugged into the same sockets where smaller memories once resided. The Electronic Industries Association's (EIA's) Joint Electronics

Devices Engineering Council (JEDEC) group has since worked to establish pinout standards before a density of memory would be designed.

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0011614

8

Figure 2

2 5 6 K X 4

with Revolutionary Pinout

Memories Worldwide

A

A

A

A

E

DQ

6

7

VCC

- - - - -

VSS

8

• — — -

9

DQ 10

11

W

A 12

4

5

1

2

3

A

A

A

13

14

15

NC 16

L

c i: i: i: i:

L

L

L

C

C

C

C c c i:

32

PIN

DIP

&

SOJ

TOP VIEW

3

1

3

3

3

3

3

3

32

31

30

29

28

27

26

A

A

A

A

A

• Q

DQ

25

•VSS

- —"' " - - —-

3

J

3

3

3

3

23

22

21

20

19

DQ

A

A

A

A

3

3

18

A

1

/CCA (VSS OR VSSA) as a manufacturer option

Source: JEDEC

JEDEC Rromotion of New Pinouts

For a while, the JEDEC pinouts were anything but controversial. There were rare breaks from a relatively predictable course until speed worries caused one member company to propose a sweeping change that would move the power and ground pins to the center of the package and add multiple power and ground pins, rather than one of each.

There were several reasons why this should be done. First and foremost, although moving the power and ground pins to the center of the package would not offer any reductions in bonding-wire inductance, a substantial improvement could be made in the parasitic elements of both the lead frame and the chip metallization paths. Both of these would allow memory designers to circumvent a significant amount of ground bounce and to produce faster devices using a given technology. Second, high-speed

PC boards are now almost exclusively made using multilayer PC boards, removing any advantage or disadvantage to board layout that might have once resulted from the placement of the power and ground pins. Any pin is as easy to route to the ground plane as the next. Finally, DIP packages are now offered in higher pin counts than were possible at the dawn of the comer power-ground era. Twenty years ago, it would have been difficult to justify consuming more than 2 of the 14 or l 6 available pins for more power and ground support. Now 32-pin,

3(X)-mil DIPS and 44-pin, 400-mil small-oudine

J-heads (SOJs) are in mass production, and the impact of adding extra power and ground pins is no longer as great as before.

After much discussion, JEDEC members decided to settle on both "evolutionary" (corner powerground) and "revolutionary" (center powerground) versions of current and future pinouts, with the notion that all members would be able to move from the production and consumption of evolutionary to revolutionary devices as they saw fit. Tlie same body also expects to see all proposed pinouts moving to a revolutionary style in the long term.

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0011614

Memories W o r l d w i d e

Figure 3

Parasitic V,, Inductance Ground-Bounce Mechanism

PCB Vcc Plane

Vcc Lead

Output

Buffer

Switching

Currents from

Other

Circuits on Chip

CHIP Vcc

K/Hlf-f

Output Lead

C L

<T\ Load

Ground Lead

PC Board Ground Plane

Source: Integrated Device Technology Inc.

TI and Center Power-Ground Logic

In the niid-1980s, Texas Instniments Inc. (TI) tried to strike out on its own and use the center power-ground and multiple power-ground pin ideas to reduce ground bounce in their line of high-speed MSI logic products. Ground bounce is an even bigger problem with logic than it is with RAMs because the outputs of logic devices are expected to cross a threshold only once (RAM outputs are allowed to be dirty before they settle). Sometimes the outputs of a logic gate are used as clock inputs on another device, so any jitter due to ground bounce might cause false triggers in a downstream circuit.

TI's solution suffered from poor market acceptance because of four faaors. First, the parts were not drop-in replacements for existing devices. Second, they offered no speed advantage over existing MSI produas available from

TI's competitors. Third, the added power and ground pins pushed devices out of the standard

20-pin package into a significandy larger 24-pin package, a distinct disadvantage. Fourth, there was no alternate source, and TI's competitors were not committing to supply pin-compatible devices until they saw market acceptance. The more traditional pinout was preferred by users.

You First

Once JEDEC's standards for center power and ground SRAM pinouts were in place, the next step was for the manufacturers to produce them. For a while it seemed that resource constraints were prohibiting most, if not all, companies from freeing a designer to work on a revolutionary device. More likely, the market for evolutionary products was a known, while the revolutionary concept was a gamble. Most static

RAM manufacturers probably remember TI's experience in logic pinouts and are now taking a "wait and see" approach, hoping not to get too far behind the market leaders should the revolutionary pinout take off.

Technical Trade-Offe

Package Size

The addition of power and ground pins is an impact to the size of DIP required for a given density SRAM. Although the impact on package cost is small and fades in comparison to savings that can be attained through improved manufacturing techniques, the impact to the size of a printed circuit card is more important.

When a 28-pin device is replaced by a 32-pin version of the same function (see Figures 1 and

2), it uses about 14 percent more PC board

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-SOOO / Fax (408) 437-0292

0011614

10

Memories Worldwide space. This space could come at an added cost, but is also quite likely to require desirable features to be omitted from a board simply due to a lack of available space.

As mentioned, higher pin-count packages are becoming more widely available, so package availability should not be expeaed to impaa the industry's migration toward the revolutionary pinout.

Package Trends vs. Pinout Changes

A strong trend exists among high-speed SRAM system designs to abandon the DIP package in favor of surface-mount packages, usually the

SOJ. This trend waters down the need to use the revolutionary pinout for two reasons. First, the lead frame of an SOJ is significantly smaller than that of a DIP, to the point that there is only a slight difference in parasitic inductance between the comer leads and the center leads.

Second, most larger RAM designs use end bonding rather than radial bonding. The corner of the lead frame will be closer to the bonding pads than will be the center pin, which means that the lead frame will actually be less inductive on a comer ground pin than it will be on a center ground pin. As die sizes increase to support increasingly larger memory arrays, the length of the lead frame to the power and ground pins diminishes considerably (see

Figure 4).

In the future we can expect lead frame inductance to become even less of an issue, once

TSOP, tape-automated bonding (TAB), and other extremely dense packaging technologies become commonplace.

Perhaps the biggest contribution the revolutionary pinout will bring to high-speed designs in light of these packaging trends will not be the position of the pins, but rather the increase in the number of piower and ground pins that will be supported. The lai^er the number of ground pins, the lower the ground inductance, because the bonding wires will be in parallel with each other.

Figure 4

Evolutionary and Revolutionary Pinouts

Corner

Vcc

_££b__i£l3_r£!a__E£b__E£b

Center

Vss Vc

,Ih-i£h-r£h-s£h-i£h-n!l.

i i

L j J L ^ L ^ L ^ 4 _ r J L ^ L T _ r J L ^ S _ r J 4 ^ K ^ 4 y S j ^

Small Side-Bonded Die In DIP Package

Note: The need for center power-ground arrangements

IS much more Innportant with a small die or In a DIP package than with a large end-bonded die or a product in an SOJ package (note length of corner-pin lead frame).

Center er

Vss Voc n n

1

^ ^ = ^

• D O ! & • •

a a a a

ts •

L a D o o

O O O • • •

li U

u u

Corner

Large End-Bonded Die in SOJ Package

Vss

Source: Dataquest (November 1991)

Corner

V s .

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0011614

i

Memories Worldwide

11

Effects of Future SRAM

Organization and Operati€m

Certain side benefits will result from the adoption of JEDEC's revolutionary^ pinouts. For example, SRAM manufacturers now feel free to use multiple power and ground pins to support the manufacture of wider high-speed devices.

Toshiba Corporation is now sampling a 64Kxl6

SRAM, which uses multiple ground pins to achieve a speed that rivals diat of narrower

64Kx4 RAMs despite the device's word width.

IDT has moved to supply DIP package versions of its highest-speed, dual-port SRAMs only in center power-ground DIPs, as opposed to the corner power-ground package offered for lowerspeed devices.

Another important trend that came into being along with the revolutionary pinouts was the introduction of JEDEC's first standardized synchronous SRAMs. Synchronous SRAMs are another means of tackling some of the trickier problems associated with write cycles in highspeed systems. C^ertain JEDEC revolutionary pinouts specify standards for synchronous clock inputs and control circuits.

T h e Players

JEDEC

The Joint Electron Devices Engineering Council

(JEDEC) of the Elearonic Industries Association

(EIA) is the means by which members of the

EIA attempt to ensure standardization across the electronics industry. Their accomplishments are commendable considering the extreme pace of the industry and the secrecy that veils the majority of most member companies' future efforts.

JEDEC's JC-42 committee has put a sizable effort into causing the center pOTver-ground revolutionary pinout to become an accepted standard. It is now up to JEDEC's member companies to either produce or consume devices that meet the standard, depending on the nature of their business.

Philips Semiconductor

The main driver for the revolutionary pinout was Philips Semiconduaor. Ironically, this company was an insignificant player in the highspeed SRAM market and never shipped an

SRAM using the revolutionary pinout. The intent appeared to be that Philips would introduce revolutionary products in 1990 or 1991; however, in 1991, the company decided to abandon its SRAM efforts.

Motorola Incorporated

In the absence of Philips, Motorola appears to be the current champion of the revolutionary pinout. At the moment. Motorola is shipping revolutionary pinout 64Kx4 and 32Kx8 SRAMs in moderate volume. Motorola claims to be able to reach access times as fast as 10ns with this device.

Mitsubishi Electronics Corporation

Mitsubishi has recendy announced plans to sample a 32Kx8 revolutionary pinout device late this year, which is expected to boast an access time of 8ns.

Toshiba Corporation

Toshiba has not yet shipped any revolutionary pinout devices, but by nature of its agreements with Motorola, we can expect Toshiba to produce revolutionary 32Kx8 and 64Kx4 SRAMs sometime soon. So far, it appears that Toshiba's only multiple-ground SRAM is its 15ns 64Kxl6, a produa that has been introduced without fanfare and appears to be easing its way rather slowly into design-ins, despite its bam-burning

15ns speed. This device will replace four 15ns

64KX4S,

which until only recentiy were considered to be the state of the art. At these speeds, the advantage of cutting capadtive loading on a processor's address outputs to one quarter of its previous value should be of major interest to many designers.

Hitachi Ltd

Hitachi Ltd. has made public its plans to produce revolutionary pinout lMbx4 SRAMs late this year, revolutionary pinout 128KX8/9 SRAMs in the first quarter of 1991 and the 64Kxl6

SRAM, which will be pin compatible with the

Toshiba device.

Others

Fujitsu Ltd. and others have strongly voiced support for the revolutionary pinout standards, but there is no word of what will be introduced when. Other companies have kept silent on their endeavors. Orily time will tell how well the revolutionary pinout will be accepted from a supplier's level. The only

SRAM user we have noticed that has publicly shown its support by purchasing revolutionary pinout devices has been HP/Apollo, whose machines have recently graced Motorola's advertisements.

Dataquest Perspective—The

Future

Manufacturers' Plans

At the moment, it appears that none but the bolder manufacturers are starting to produce revolutionary devices and then in a single density—^25DK. Volume requirements for these

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0011614

12 Memories Worldwide

products are extremely uncertain and are therefore not forecast in this report. These 256K devices are follow-ons to already successful devices; therefore, the stakes to the supplier of introducing a revolutionary product are low.

Also, there is still much room for growth in the high-speed 256K RAM market, so it makes more sense to try revolutionary devices at the 256K density than at lower densities.

Customers' Needs

Historically, considerable attention has been devoted to supporting simple system upgrades via plug-in replacement of slower parts with faster ones as the faster ones become available.

Despite all this attention, most SRAM users do not take advantage of this practice. Whatever the reason, the issue of having to redo the board layout will probably not impact the decision of whether to use revolutionary or evolutionary pinout devices. These layouts will happen anyway, and the question then simply boils down to which part to use.

The more important question of availability will continue. Are the revolutionary parts secondsourced.^ Because JEDEC passed both revolutionary and evolutionary pinouts for several densities, there is no imminent switch-over point after which the user will be forced to use revolutionary pinout devices. The evolutionary/ revolutionary decision will be deeply affected by the personal judgement of various persormel within the memory users' organizations. Expect to see a gradual decline in the use of evolutionary pinouts to revolutionary devices as the masses convert.

The Question of Inertia

Dataquest has often seen instances of incredible resistance to moving in a new direction, despite the fact that the new technology offers significant improvements. Many high-speed system designers get into trouble designing systems using T i l levels where ECL would make more sense. Although synchronous SRAMs have existed for at least four years, it is difficult to find applications that take advantage of them.

It appears that even those designers on the forefront of technological advancement occasionally hold on to comfortable tools of the past in spite of the availability of superior solutions. In this light, it would not be at all surprising if revolutionary pinout SRAMs were to get off to a slow start, with system designers breathing a sigh of relief every time an evolutionary device was coaxed into running at speeds previously only attained by revolutionary devices. The real turnaround will be indicated when a manufacturer introduces the revolutionary pinout version of a device first and follows it with the introduction of an evolutionary device (on a device for which both revolutionary and evolutionary standards exist), rather than vise versa.

Chicken and Egg Issues

As with any other advancements, two opposing forces are working against each other to the detriment of progress. On one hand, the customers' buyers and component engineers want to avoid allowing the design-in of a device that is sole-sourced; yet, on the other hand, the potential manufacturers of that device want to know that there will be a ready market for their product once it is introduced. A disadvantage from the supplier's viewpoint is that these devices will only be purchased for the absolute highest-speed applications, and devices that do not match such high speeds will not be salable as speed-downgraded products. It is fominate for everyone involved that Mitsubishi and Motorola have embraced the 32Kx8 pinout. These devices will certainly fly with these strong suppliers. The fate of JEDEC's other revolutionary pinouts, however, is still in limbo.

By Jim Handy

For More Information . . .

On the topics in this issue Memories Worldwide (408) 437-8228

About on-line access On-Line Service (408) 437-8576

About other Dataquest publications Sales (408) 437-8246

About upcoming Dataquest conferences Conferences (408) 437-8245

About your subscription Customer Service (408) 437-8402

Via fax request Fax (408) 437-0292

The content of this report represents our inteipretation and analysis of information generally available to the public or released by responsible individuals In the subject companies, but is not guaranteed as to accuracy or completeness. It does not contain material provided to us in confidence by our clients. Individual conqianies reported on and analyzed by Dataquest may be clients of this and/or other Dataquest services. This information is not lumished in connection with a sale or offer to sell securities or in connection with the solicitation of an offer to buy securities. This firm and its parent and/or their officers, stockholdeis, or membera of their families may, irom time to time, have a long or short position in the securities mentioned and nuy sell or buy such securities.

©1991 Dataquest Incorporated / 1290 Ridder Park Drive, San Jose, CA 95131-2398 / (408) 437-8000 / Fax (408) 437-0292

0011614

Was this manual useful for you? yes no
Thank you for your participation!

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

Download PDF

advertisement