Zum Inhalt springen

Moore's Law: What It Actually Said

Zusammenfassung

Gordon Moore’s 1965 observation — now called “Moore’s Law” — stated that the number of components per integrated circuit was doubling approximately every year. It said nothing about processing speed. The popular version of Moore’s Law, widely quoted as “computing power doubles every 18 months,” is a misquotation that blends Moore’s transistor count observation with a separate 1974 observation by Intel executive David House. Moore himself later revised the doubling period to approximately two years. The law has been simultaneously a prediction, a roadmap, and a self-fulfilling prophecy — and it began to break down around 2015.

The 1965 Paper

Gordon Moore published “Cramming More Components onto Integrated Circuits” in Electronics magazine on April 19, 1965. At the time, Moore was the director of research and development at Fairchild Semiconductor. He had been asked to predict how semiconductor technology would evolve over the next decade.

Moore plotted the number of components (transistors, resistors, capacitors, and diodes) on the most cost-effective integrated circuits from 1959 to 1965. He observed a doubling each year and extrapolated that this rate would continue for at least another ten years. His 1965 paper predicted circuits containing 65,000 components by 1975 — a prediction that proved roughly correct.

In 1975, Moore revised his prediction: the doubling would occur every two years rather than every year. This 1975 revision is the version usually cited as “Moore’s Law,” though the name was coined by Caltech professor Carver Mead in the early 1970s.

What the Law Is and Isn’t

Moore’s Law makes a claim about transistor density — how many transistors fit in a given area of silicon. More transistors per unit area generally produces either more capable circuits (more logic) or faster circuits (shorter interconnect lengths), but the relationship between transistor count and “computing power” (however that is measured) is not direct.

The “18 months” version conflates three separate observations:

  1. Moore’s (revised) ~24-month transistor doubling period
  2. Dennard Scaling (1974) — that transistor shrinkage allowed power consumption per transistor to decrease proportionally, allowing more transistors at the same power
  3. David House’s estimate that combined transistor improvements and clock speed improvements produced a ~18-month doubling of overall performance

Dennard Scaling broke down around 2005-2007. As transistors became too small to reduce voltage further without leakage, clock speeds stopped increasing. Processor performance improvements shifted from single-core clock speed to multi-core parallelism — a fundamentally different kind of improvement.

The Roadmap Function

What made Moore’s Law particularly powerful was that it functioned as an industry roadmap. Semiconductor manufacturers, software developers, and hardware designers all planned on the assumption that transistor density would double on a predictable schedule. This coordinated investment: foundries built new facilities, compiler writers optimized for next-generation architectures, software companies wrote applications that “the next generation of hardware” would make fast enough.

Gordon Moore later described the law’s self-fulfilling nature: by publishing the observation, he gave the industry a target, and the industry hit the target partly because everyone was aiming at it. The Integrated Circuit Revolution documents the technological progress this coordinated effort produced.

📚 Sources