More on Moore's Law

Without entering into the different points of view concerning Moore's Law expressed by Joel Birnbaum and R. Stanley Williams in their article (Physics Today, January 2000, page 38) and Igor Fodor's letter (Physics Today, October 2000, page 106), I must comment more generally on the observation called Moore's Law.

In Figure 3 of the Birnbaum and Williams article, the standard growth curve is shown for Intel processor chips from the 4004 of the early 1970s to the Pentium models of the late 1990s. The slope of the curve as drawn shows a doubling time constant of about 26 months; the caption, in contrast, describes the growth as "a factor of four every three years," or a doubling time constant of 18 months! This inconsistency is a frequently promulgated misinterpretation of the observation of Intel Corp's founder Gordon Moore.

In the article, Birnbaum and Williams use a 3.4-year time constant for the factor-of-four increase in the number of bits stored on a memory chip. This corresponds to a 20-month doubling time constant. Indeed the slope for memory chip growth has been and continues to be steeper than for processor chips.

Moore, commenting on the growth of the microelectronics industry in 1964, noted a doubling of the number of elements on a produced chip once every 12 months.¹ For a decade, that meant a growth factor of approximately 1000. Today, when Moore's Law is quoted, the time constant typically quoted is 18 months. Actually, it was 18 months starting in the mid-1970s, approximately 10 years after the original observation. For a decade, then, the growth factor was approximately 100.

The 18-month time constant was no longer valid by the end of the 1980s. For example, from 1980 to 1990 the number of transistors in the Intel-80x processor chips grew from about 29 000 to approximately 1.2 million—substantially less than a factor of 100. In the 1990s, the doubling time constant has been closer to two years. This gives a decade growth factor of approximately 32.

As the industry approaches the physical limits of the complementary metal oxide semiconductor (CMOS) technology curve we have been riding so effectively, let alone the economic limits that are also at work, the rate of growth of the number of transistors on a chip will further decrease. The Semiconductor Industry Association's road map² shows a growth of about a factor of 24 for microprocessors in the decade between 1999 and 2009. That implies a Moore's Law time constant for doubling of about 2.5 years.

References

1. 1. R. N. Noyce, Sci. Am., Sept 1977, p. 63. Information Technology and R&D: Critical Trends and Issues, rep. no. OTA-CIT-268, US Congress, Office of Technology Assessment, Washington, DC, (February 1985), p. 324.
2. 2. International Technology Roadmap for Semiconductors: 1999 Edition, Semiconductor Industry Association, San Jose, Calif. (1999), p. 3. Also available at [LINK].