Presentation on theme: "Benchmarking Microelectronics Innovation: Understanding Moores Law and Semiconductor Price Trends Kenneth Flamm Technology and Public Policy Program Lyndon."— Presentation transcript:
Benchmarking Microelectronics Innovation: Understanding Moores Law and Semiconductor Price Trends Kenneth Flamm Technology and Public Policy Program Lyndon B. Johnson School of Public Affairs University of Texas at Austin email@example.com
Outline n Why Do We Care? n Moores Law n The Economics of Moores Law n The Economic Impact of Moores Law n Benchmarking Moores Law n Tinkering with Moores Law n Point of Inflection? n Official Data on Semiconductor Prices n Better Benchmarking
Key Economic Features of the Semiconductor Industry n Extremely rapid technical progress n Large R&D Investments n Learning Economies n Capital Intensity n Capacity Constraints, Long Gestation Lags n One Complicated Industry
Why Do We Care? n Now largest U.S. manufacturing industry u Measured by value added n One 4-digit manufacturing industry now almost 1% U.S. GDP n Most important input to other industries we care a lot about u Computers, communications u Big impact on GDP, productivity growth u See Jorgenson AEA 2001 Presidential Address.
Changing Size: U.S. semiconductor mfg val added vs. GDP n 1958 n 1965 n 1975 n 1985 n 1995 n 1997 n.04% n.09% n.13% n.26% n.70% n.77%
Moores Law n In the beginning: the original law u 2x devices/chip every 12 months u ca. 1965 n Moore rev.2 u 2x devices/chip every 18 months u ca. 1975 n Self-fulfilling prophecy? u it happened because everyone believed it was going to happen u The receding brick wall
Economics of Translating Moore into $ and ¢ $/device = $ processing cost area silicon Area/chip _____________________ Devices/chip New technology node every 3 years Lithography advance means.5X area per chip feature Moores law4x devices/chip every 3 years Would predict Area/chip 2X every 3 years $ processing cost/wafer area roughly constant CADR = -21%
An Economists Default Corollary to Moores Law: Moores Law + constant wafer processing cost + new technology node every 3 years = -21 % CADR
The Ingenuity (DRAM) Corollary: n Instead of doubling chip size, use ingenuity to increase it only Z (Z < 2) times u real recent example (DRAMs), Z=1.37 u 3-D device structures n Implications of ingenuity u for DRAMs recently, CADR = -30% u for DRAMs, in 70s and 80s, wafer processing cost also fell, CADR more like -37% F Japan/VLSI project, competition impact? n Another example is ASICs, more rapid leading edge technology adoption u transitory impact on CADR
Benchmarking Moores law: Differences in Semiconductor Price Movements Are HUGE Source: Aizcorbe, Flamm, and Khurshid (2001).
Implications for Input Prices in Different User Industries Also Great Source: Aizcorbe, Flamm, and Khurshid (2001).
Accounting for the economic impact of Moores Law n The standard model n Estimated cost decline n Estimated price elasticity n Calculations of benefits
The Numbers: Summary of Consumer Welfare Calculations Benefit in 1995 of Billion $ Percent of GDP Percent of 1995 GDP Growth 1 years price decline 20 years price declines 30 years price declines 1.8 378 1503 0.16 5.2 21 8 260 1039
Magnitudes n 1 years tech improvement yields.16% GDP …forever n 20 years tech improvement would cost you about 5 percent of GDP if rolled back n If youre feeling really brave, roll the clock back 30 years and you shave off up to 20 percent of GDP!
More Comparisons n Other well-studied cases-- the railroads in the 19th century n The old guys vs. the new guys: a historical parable
Tinkering with Moores Law: The Technological Acceleration (Sematech Roadmap) Corollary n Suppose new technology node every 2 years instead of 3 u Industry coordinated push through Sematech in late 1990s u Competitive pressures also pushed n New default (2X chip size) CADR = - 29% n New DRAM (1.37X chip size) CADR = -41% n Constant chip size (1X chip size) CADR = -50%
Decline Rates in Price-Performance Percent/Year Microprocessors, 1975-85-37.5 Hedonic Index1985-94-26.7 DRAM Memory, 1975-85-40.4 Fisher Matched Model1985-94-19.9 DRAMs, Fisher Matched Model, Quarterly Data 91:2-95:4-11.9 95:4-98:4-64.0 Intel Microprocessors, Fisher Matched Model, Quarterly Data 93:1-95:4-47.0 95:4-99:4-61.6 Sources: Flamm (1997); Aizcorbe, Corrado, and Doms (2000) Point of Inflection?
Implications of This Interpretation of Moores Law n Ultra-high rate of innovation in late 1990s temporary n Transitory factors increased innovation above long-term sustainable rates u Shortened product lives u Intensified competition u More rapid adoption of leading edge processes in other products n Future CADR will look more like –40% than –60%+ n Economic impacts may decline to lower but more sustainable rates
Benchmarking Moores Law in the U.S.: Official Statistics on Chip Prices n BEA got ball rolling, taken over by others n BLS-- Much improved for DRAMs and Microprocessors, not so hot for other products u Data sources a concern u Documentation a concern n Fed Reserve has stealth program, currently best numbers in town u Data sources a concern u Weights a concern u Possible application in estimating capacity a big concern u Access/availability outside Fed a concern
Comparison of BLS with Other Price Indexes for Microprocessors
Better Benchmarks for Semiconductor Innovation n Tracking it better in a time of change u Focus more scarce stat resources on price indexes for IT sectors, reflecting growing relative importance to economy u A real collection program for underlying price data, perhaps coordinated with industry trade organizations F Under the hood at Dataquest (& others) not a pretty story u Decent coverage of products besides memory and microprocessors u New initiatives in communications n Better understanding of R&D trends n Better coordination of public/private R&D investments