Transcript How the Semiconductor Industry Interacts with the Rest of

Benchmarking
Microelectronics Innovation:
Understanding Moore’s 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 protected]
Outline









Why Do We Care?
Moore’s Law
The Economics of Moore’s Law
The Economic Impact of Moore’s Law
Benchmarking Moore’s Law
Tinkering with Moore’s Law
Point of Inflection?
Official Data on Semiconductor Prices
Better Benchmarking
Key Economic
Features of the
Semiconductor
Industry






Extremely rapid technical progress
Large R&D Investments
Learning Economies
Capital Intensity
Capacity Constraints, Long
Gestation Lags
One Complicated Industry
Why Do We Care?

Now largest U.S. manufacturing industry



Measured by value added
One 4-digit manufacturing industry now
almost 1% U.S. GDP
Most important input to other industries
we care a lot about
Computers, communications
 Big impact on GDP, productivity growth
 See Jorgenson AEA 2001 Presidential
Address.

Changing Size: U.S.
semiconductor mfg
val added vs. GDP






1958
1965
1975
1985
1995
1997






.04%
.09%
.13%
.26%
.70%
.77%
Moore’s Law

In the beginning: the original law
2x devices/chip every 12 months
 ca. 1965


Moore rev.2
2x devices/chip every 18 months
 ca. 1975


Self-fulfilling prophecy?
“it happened because everyone believed
it was going to happen”
 The receding brick wall

0
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
Log2 of the Number of
Components Per Integrated Function
Figure 1
The Original "Moore's Law" Plot
From Electronics April 1965
16
15
14
13
12
11
10
9
8
7
6
5
4
,
,
,
3
2
,
1
,
Economics of
Translating Moore
into $ and ¢
$/device =
$ processing cost Area/chip
area silicon
_____________________
Devices/chip
New “technology node” every 3 years
Lithography advance means .5X area per
chip feature
Moore’s law
4x devices/chip every 3 years
Would predict Area/chip 2X every 3 years
$ processing cost/wafer area roughly constant
CADR = -21%
An Economist’s
Default Corollary to
Moore’s Law:
Moore’s Law + constant wafer processing
cost + new technology node every 3 years
=
-21 % CADR
The Ingenuity
(DRAM) Corollary:


Instead of doubling chip size, use ingenuity to
increase it only Z (Z < 2) times
 real recent example (DRAMs), Z=1.37
 3-D device structures
Implications of ingenuity


for DRAMs recently, CADR = -30%
for DRAMs, in 70s and 80s, wafer processing
cost also fell, CADR more like -37%


Japan/VLSI project, competition impact?
Another example is ASICs, more rapid leading
edge technology adoption
 transitory impact on CADR
Benchmarking Moore’s 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
Moore’s Law




The standard model
Estimated cost decline
Estimated price elasticity
Calculations of benefits
The Standard
Model: Consumer’s
Surplus
Figure 5
Social Benefit From a Decline in Semiconductor Price
Dollars
a
d
b
Po
c
e
PN
D
Qo
QN
Q
Q = Quant it y; D = Dem and.
The Numbers:
Summary of Consumer
Welfare Calculations
Percent of
1995 GDP
Growth
Benefit in 1995 of
Billion $
Percent
of
GDP
1 year’s price decline
1.8
0.16
20 years’ price declines
378
5.2
260
30 years’ price declines
1503
21
1039
8
Magnitudes



1 year’s tech improvement yields
.16% GDP
…forever
20 years’ tech improvement would
cost you about 5 percent of GDP if
rolled back
If you’re feeling really brave, roll
the clock back 30 years and you
shave off up to 20 percent of GDP!
More Comparisons


Other well-studied cases-- the
railroads in the 19th century
The old guys vs. the new guys: a
historical parable
Tinkering with Moore’s
Law: The Technological
Acceleration (Sematech
Roadmap) Corollary

Suppose new technology node every 2
years instead of 3
Industry coordinated push through
Sematech in late 1990s
 Competitive pressures also pushed




New default (2X chip size) CADR = 29%
New DRAM (1.37X chip size)
CADR = -41%
Constant chip size (1X chip size)
CADR = -50%
Point of Inflection?
Decline Rates in Price-Performance
Percent/Year
Microprocessors,
Hedonic Index
1975-85
1985-94
-37.5
-26.7
DRAM Memory,
Fisher Matched Model
1975-85
1985-94
-40.4
-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)
99
:4
98
:2
98
:4
99
:2
96
:4
97
:2
97
:4
95
:2
95
:4
96
:2
93
:4
94
:2
94
:4
92
:2
92
:4
93
:2
91
:2
91
:4
Change of Pace?
1000
100
10
DRAM
Intel MicroProc
1
0.1
Implications of This
Interpretation of
Moore’s Law


Ultra-high rate of innovation in late
1990’s temporary
Transitory factors increased innovation
above long-term sustainable rates
Shortened product lives
 Intensified competition
 More rapid adoption of leading edge
processes in other products



Future CADR will look more like –40%
than –60%+
Economic impacts may decline to lower
but more sustainable rates
Benchmarking Moore’s
Law in the U.S.: Official
Statistics on Chip
Prices



BEA got ball rolling, taken over by others
BLS-- Much improved for DRAMs and
Microprocessors, not so hot for other products
 Data sources a concern
 Documentation a concern
Fed Reserve has stealth program, currently best
numbers in town
 Data sources a concern
 Weights a concern
 Possible application in estimating capacity a
big concern
 Access/availability outside Fed a concern
Comparison of BLS with
Other Price Indexes for
Microprocessors
Microprocessor Price Indexes
10000
flamm hedonic
100
BLS PPI
FRB
10
Ja
n
Ma
y
S
19 ep
97
Ja
n
Ma
y
S
19 ep
98
Ja
n
Ma
y
S
19 ep
99
Ja
n
Ma
y
S
20 ep
00
Ja
n
Ma
y
1
19
96
12/98=100
1000
Better Benchmarks
for Semiconductor
Innovation

Tracking it better in a time of change
Focus more scarce stat resources on
price indexes for IT sectors, reflecting
growing relative importance to economy
 A real collection program for underlying
price data, perhaps coordinated with
industry trade organizations
 Under the hood at Dataquest (&
others) not a pretty story
 Decent coverage of products besides
memory and microprocessors
 New initiatives in communications



Better understanding of R&D trends
Better coordination of public/private R&D
investments