Semiconductor Real Estate
Understanding Moore's Law: Four Decades of Innovation. Edited by David C. Brock. x + 122 pp. Chemical Heritage Press, 2006. $12.50.
The rule of thumb known as Moore's Law has many versions and variations, but they all amount to roughly the same thing: The performance of digital computers doubles every year or two. The steady doubling and redoubling has been going on for more than 40 years, long enough that it has come to seem a normal and unremarkable fact of life; yet such sustained exponential growth is unmatched and unprecedented anywhere else in economics or technology. And it's not over yet.
The law is named for Gordon E. Moore, one of a legendary group of eight who founded Fairchild Semiconductor in 1957; a decade later Moore was a cofounder of Intel Corporation. The earliest expression of Moore's Law was a graph published in 1965 in an article written at the invitation of Electronics magazine. The graph plots growth in "the number of components per integrated function"—essentially the number of transistors on a silicon integrated circuit, or what we would now call a chip. Five data points on the graph show that the component count increased from 1 to 60 over the interval from 1959 to 1965. On a semilogarithmic scale, the points form a reasonably straight line, with a slope that implies a doubling of circuit density every year. Moore extended this line out to 1975, where it predicted chips with 60,000 transistors. This projection, which many at the time considered preposterous, turned out to be right on the mark.
Moore got a chance to revisit his predictions in 1975, in a talk delivered at the International Electron Devices meeting. He argued that the annual doubling of chip complexity would continue for a few years and then gradually slow, settling into a two-year doubling time. Again he was largely correct, although the period of slower growth began sooner than he expected. There have been further changes in the slope of the trend line since then, but the essential observation—that circuit density grows exponentially—remains valid.
In 2005 a symposium was held in Philadelphia to mark the 40th anniversary of Moore's Law. The slender book Understanding Moore's Law grew out of that meeting, although it is not a conventional proceedings volume. It includes several interpretive essays as well as four works by Moore himself: a facsimile of the manuscript of his 1965 article for Electronics, the published version of the same article, the 1975 talk reassessing the predictions a decade later, and finally a further reflection written for the 2005 gathering. This last update shows that Moore still has a knack for capturing the essence of an economic trend. "The semiconductor industry is really selling real estate," he writes. "The price of that real estate has been nearly constant for as long as I have been in the business: on the order of a billion dollars an acre." Semiconductor manufacturers sell a few hundred acres a year, densely packed with more than 1018 transistors. While the cost per acre has remained constant at a gigabuck, the price per transistor "has dropped to about a hundred nano-dollars."
The other essays in this volume—one by Arnold Thackray and three by the editor, David C. Brock—provide useful historical and technological background. But readers should be warned that the book has an agenda. The 2005 symposium was organized by the Chemical Heritage Foundation, and part of the motive was evidently to assert the claims of chemistry as a key contributor to the semiconductor industry. The claims are doubtless valid—Moore was trained as a chemist, for example—but at times the rhetoric of interdisciplinary rivalry gets in the way of telling the story.