Well-VersedLibraryCoverageAbout
2026-09-23 · Technology
Technology · Silicon

Moore’s Law

The most quoted forecast in technology is a four-page article about the price of components, and it says nothing about speed.

Portrait photograph of Gordon Moore in a dark suit and red tie, 2004.
Gordon Moore (1929–2023), photographed in 2004. A physical chemist by training, he was running research at Fairchild Semiconductor when Electronics asked him for a ten-year forecast, and he co-founded Intel three years later. Source: Wikimedia Commons, where the author and licence terms for this file are recorded.

The thirty-fifth anniversary issue of Electronics wanted a forecast, and the man it asked was a thirty-six-year-old physical chemist who ran the research laboratories at Fairchild Semiconductor. Gordon Moore had one good chart and very little else. He plotted the largest number of components anyone at Fairchild had managed to fit onto a single silicon chip in each year since the planar transistor arrived in 1959, found five points, laid a straight line through them on semi-logarithmic paper, and carried the line a decade forward. The result ran four pages in the issue of 19 April 1965 under a title almost nobody remembers, “Cramming more components onto integrated circuits,” and at the far end of the line sat 65,000 components on one chip by 1975.

The sentence that made him famous is stranger than its reputation. “The complexity for minimum component costs has increased at a rate of roughly a factor of two per year.” Complexity means components per chip, and the clause that decides the meaning is for minimum component costs, which points at a curve Moore drew on the facing page. Put a handful of components on a chip and each one is dear, since the silicon and the package are shared among too few. Add more and the cost of each falls. Add too many and the yield collapses, because a single flaw ruins the whole chip. Between those two failures sits a cheapest chip, and in 1965 that chip held about fifty components. Moore’s claim was that the bottom of the valley slides to the right, doubling every year, so a manufacturer who wanted the cheapest transistors would have to build ever more crowded chips to get them.

1 10 100 1,000 10,000 100,000 COMPONENTS PER INTEGRATED CIRCUIT RELATIVE COST PER COMPONENT 1962 1965 1970 50 components the cheapest chip of 1965
Moore’s cost curves, redrawn from Figure 1 of the 1965 article. Both axes are logarithmic. Each curve is one year’s manufacturing technology; the dot marks the minimum Moore reported for 1965, at roughly fifty components. The prediction is the horizontal march of those minima, and it concerns price rather than performance.
ReadMoore, 1965 · tap the dotted words for a gloss

For simple circuits, the cost per is nearly inversely proportional to the number of components, the result of the equivalent piece of semiconductor in the equivalent package containing more components. But as components are added, decreased more than compensate for the increased , tending to raise the cost per component. Thus there is a minimum cost at any given time in the evolution of the technology. At present, it is reached when 50 components are used per circuit.

The complexity for minimum component costs has increased at a rate of roughly a factor of two per year. 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.

Gordon E. Moore, “Cramming more components onto integrated circuits,” Electronics, 19 April 1965, pp. 114–117.

He got nine doublings in the ten years rather than ten. Carver Mead, a friend at Caltech, gave the observation the name it carries, and the name did it no favours, since a law implies physics and this was a trend drawn through five measurements. Moore revised it once, at the International Electron Devices Meeting in December 1975. Three things had been driving the doubling: larger dies as defect rates fell, finer minimum dimensions, and what he called circuit and device cleverness, the knack of wasting less silicon on each function. The third had been worth roughly half the gain and was now spent, so from about 1980 he expected a doubling every two years. The eighteen-month version belongs to someone else. Dave House, an Intel executive, put the two-year doubling together with the fact that smaller transistors also switch faster, and forecast that computer performance would double every eighteen months. “But I never said 18 months,” Moore said later.

What happened next is the part worth having a view about. A forecast an entire industry believes stops being a forecast. Chip companies began writing their plans against the curve, then published shared roadmaps fixing the delivery date of each generation, and the equipment makers, the materials suppliers and the design-tool houses all aimed at the same dates. Moore’s line became the industry’s timetable, and companies met it because missing a date cost them a generation. Intel’s 4004, the first microprocessor, shipped in 1971 carrying 2,300 transistors. Nvidia’s GB202, the densest processor on sale in 2025, carries more than 92 billion.

An Intel C4004 microprocessor in a white ceramic package with a gold lid and sixteen gold pins, photographed against white.
An Intel C4004, 1971. Sixteen pins, a white ceramic package, and 2,300 transistors on a die of about twelve square millimetres. Six years after the article, Moore’s curve was still comfortably ahead of what anyone was shipping. Source: Wikimedia Commons, where the author and licence terms for this file are recorded.
· · ·

Now the misunderstanding. Chips stopped getting dramatically faster in the middle of the 2000s, and the law that failed then carries a different name. Robert Dennard, with five colleagues at IBM, showed in October 1974 that shrinking a transistor’s dimensions leaves its power density unchanged, so smaller arrived with faster and cooler at no extra charge. That is why clock speeds climbed for three decades. The bargain depended on operating voltages shrinking along with everything else, and below a certain size they will not: current leaks through a transistor that is nominally switched off, and the heat turns up regardless. Intel cancelled two processors in 2004 when the power wall stopped them, and clock speeds have sat in the low single-digit gigahertz ever since. Designers went sideways, putting several cores on a chip and asking software to find the work. Density kept doubling. Speed did not.

Scaling itself is now slower and far more expensive than the phrase suggests. In 2016 the International Technology Roadmap for Semiconductors issued its final edition and stopped organising its research around shrinking. The node names survived as marketing: nothing on a chip sold as two nanometres measures two nanometres, a point Intel’s Mark Bohr conceded of the older numbers, saying he could not point to the one dimension that was 32 nm, or 22 nm, or 14 nm. Imec’s 2026 roadmap keeps a key transistor spacing flat from the A10 generation through A5 and buys its density by stacking transistors vertically instead. Moore saw the end coming and was unbothered by it. “I see Moore’s Law dying here in the next decade or so, but that’s not surprising,” he said in 2015, at the fiftieth anniversary. He died on 24 March 2023, aged ninety-four.

He left a habit as well as a number, and it has worn better. Moore looked at a manufacturing process and asked what a single component cost on it, because that answer told him what the industry would go on to build and sell. Headlines converted the cost curve into a promise about speed, and the promise is the part that broke.

Up next
The Waltz
A London dancing master engraved nine numbered positions of the hold, because Englishmen needed to be shown where the arm went.
Read →

Images on this page are reused from Wikimedia Commons. Each is linked to its file page, where the author and the licence terms for that file are recorded. Thumbnails in “Up next” are credited on the lesson they link to.