Cramming More Components onto Integrated Circuits

The situation

In 1965 the integrated circuit was only a few years old. Placing several transistors and other components on a single piece of silicon was still a specialist technology, used mainly in military and space programmes such as Apollo, where small size and reliability justified high costs. Gordon Moore was director of research and development at Fairchild Semiconductor, one of the leading chip makers. Electronics magazine, for its thirty-fifth anniversary issue, asked him to predict what would happen in his field over the next ten years.

What it argues

The article opens with confidence: “The future of integrated electronics is the future of electronics itself.” It then makes predictions that read today as remarkably accurate: “Integrated circuits will lead to such wonders as home computers—or at least terminals connected to a central computer—automatic controls for automobiles, and personal portable communications equipment.”

The central argument is about cost. For simple circuits, Moore explains, the cost per component falls as more components are put on a chip. But as complexity rises, manufacturing yields fall, and cost per component rises again. So at any given time there is a level of complexity at which components are cheapest. In 1965, he notes, that minimum came at about fifty components per circuit, and it was rising quickly.

Then comes the observation that made the article famous: “The complexity for minimum component costs has increased at a rate of roughly a factor of two per year.” He projects it forward cautiously: “Certainly over the short term this rate can be expected to continue, if not to increase. Over the longer term, the rate of increase is a bit more uncertain, although there is no reason to believe it will not remain nearly constant for at least ten years. That means by 1975, the number of components per integrated circuit for minimum cost will be 65 000.”

The projection rested on just a few data points. Much of the rest of the article argues that no fundamental obstacle stood in the way: components could be packed closely, yields could be raised, heat could be managed.

The idea underneath

Moore’s article is an economic argument disguised as a technical one. Its subject is not how many components can be put on a chip, but how many can be put there most cheaply. That distinction mattered. It meant the trend would continue only as long as it remained profitable to pursue, and it explains why the industry invested so heavily to keep it going.

It is also a study in how a forecast becomes a fact. After Moore revised the rate in 1975 to a doubling roughly every two years, the trend became an industry target. Companies planned products, equipment makers planned tools and software developers planned features around it. A regularity that began as an observation became a coordination device, a shared expectation that made itself true. This is close to what the uncertainty idea describes as a convention: in a world nobody can predict, people act on the assumption that the observed trend will continue, and their actions sustain it.

The trend also illustrates the Jevons paradox. Each fall in the cost of computation produced more demand for computation, not less.

How it was received and what followed

In 1968 Moore co-founded Intel with Robert Noyce. The doubling he had described, under the name “Moore’s law” coined by the engineer Carver Mead, became the heartbeat of the semiconductor industry. For roughly five decades, transistor counts rose on something close to the revised schedule, and the cost of computation fell by many orders of magnitude.

The compute essay describes how it slowed. From the mid-2000s, chips could no longer simply run faster without overheating, and later the cost per transistor stopped falling as reliably. The industry turned to multiple cores, specialised processors and other approaches. Moore himself said repeatedly that no exponential trend continues forever.

Reading it today as a decision-maker

Two lessons stand out. First, look for the economics underneath a trend. Moore’s projection held because there was a clear economic reason, falling cost per component, to keep pursuing it. Trends without such a reason are more fragile.

Second, beware the trend that becomes a habit. Whole industries, including software, built their planning on the assumption that hardware would keep getting faster for free. When the free lunch ended, some were ready and some were not. The Lisp machine and Itanium roads show different ways of being on the wrong side of this trend.

Sources and limitations

Quotations are taken from the IEEE’s 1998 reprint of the article, checked against the extracted text. The 1975 revision and the naming of the law are drawn from the biography by Thackray, Brock and Jones. The interpretation of the trend as a self-fulfilling coordination device is widely shared but remains an interpretation.