Skip to content
mentua.org

What Is Moore's Law?

Line drawing of three chips of equal size whose grids grow finer from one cell to four to sixteen, with one cell filled, beside a fourth chip drawn only as a dotted outlineLine drawing of three chips of equal size whose grids grow finer from one cell to four to sixteen, with one cell filled, beside a fourth chip drawn only as a dotted outline

5 min read≈ 1,200 words3 questions · up to 3 points

What this is about

Intel's first microprocessor, built in 1971, had 2,300 transistors, and fifty years later a single laptop chip had tens of billions. You learn the rule that predicted this growth, how to check it yourself with real numbers, and why chips stopped getting much faster around 2005. You also learn what people mean when they say the rule is dead.

In a nutshell· skip ahead

Moore's law is the observation that the number of transistors that fit on a computer chip doubles about every two years. Gordon Moore first described the trend in 1965 and gave it this pace in 1975. It is not a law of nature but a trend the chip industry turned into a target, and it has slowed since the mid-2000s.

In 1971 Intel introduced its first microprocessor, the 4004, which put a whole computer processor on one small chip with 2,300 transistors. In 2021 the largest chip Apple put into its laptops held 57 billion. That growth by a factor of about 25 million followed a steady pace, which a chemist had written down six years before the 4004 existed. Who set that pace, and can it continue?

What does Moore’s law say?

Moore’s law is the observation that the number of transistors on a computer chip doubles about every two years. A transistor is a tiny electrical switch, and a processor does all its work by switching billions of them on and off. Gordon Moore first noticed the trend in 1965, when he was head of research at the chip maker Fairchild Semiconductor. He wrote:

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

Gordon E. Moore, “Cramming More Components onto Integrated Circuits”, Electronics, 19 April 1965

In plain words, the number of parts on the cheapest chip to make had doubled every year. So Moore’s law was about cost from the start, because smaller parts meant more of them per chip and a lower price for each. In 1975 Moore looked at newer data and slowed the pace to a doubling every two years, and that is the version people mean today.

We can test the rule with real numbers. Doubling every two years means five doublings in ten years, which multiplies the count by 2 × 2 × 2 × 2 × 2 = 32. We start with the 4004 and multiply by 32 for each decade:

Year Predicted transistors per chip
1971 2,300 (the real Intel 4004)
1981 about 74,000
1991 about 2.4 million
2001 about 75 million
2011 about 2.4 billion
2021 about 77 billion

Apple’s M1 Max from October 2021 had 57 billion transistors, about three quarters of the prediction. After 25 doublings, a trend that Moore had drawn through only five data points from 1959 to 1964 was off by less than one doubling. In 2022 Apple joined two of these chips into the M1 Ultra, with 114 billion transistors.

Why did the pace stay so steady? For about thirty years, shrinking a transistor brought three gains at once. More transistors fit on the same area, each one became cheaper, and each one switched faster. The third gain is called Dennard scaling, after a 1974 paper by Robert Dennard and his colleagues at IBM. They showed that if you shrink a transistor and lower its voltage in the same proportion, it becomes faster while the heat produced per area stays the same.

The steady pace also had a human cause. Chip makers, their suppliers and software developers all planned around the two-year rhythm, and each company worked to meet it because its competitors would. The engineer Chris Mack calls the law a self-fulfilling prophecy, a forecast that comes true partly because people believe it.

What Moore’s law teaches you

The transferable insight is a question: when something doubles at a fixed pace, what drives the doubling, and what will stop it? Steady doubling feels slow at first and then overtakes every forecast drawn as a straight line.

  • Claims in the news. You will read that something doubles every year or two, such as the computing power used to train AI systems. Moore’s law shows that such a trend can hold for decades, but only while someone pays for it and no physical limit stops it.
  • Money and debt. A doubling every two years is the same as growth of about 41 percent per year. What Is Compound Interest? explains this arithmetic with savings, where a fixed percentage also produces a fixed doubling time.
  • Security. Faster computers also break codes faster, so the keys that protect your messages must keep getting longer. What Is Public-Key Cryptography? tells the story of a code locked in 1977 and broken in 1994.
  • Shared targets. A public target that everyone believes can coordinate people who never meet. In a team project, a timetable that all members trust can help to make itself true.
  • Forecasts. In 1950 Alan Turing tied talking machines to memory and expected about 10⁹ bits by 2000. What Is the Turing Test? shows that the memory arrived on time, while convincing conversation took until the 2020s.

Is Moore’s law dead?

The answer depends on which part of the law you mean. Many people think Moore’s law says that computers become twice as fast every two years. It counts transistors, and speed rose with the count only as long as Dennard scaling worked. In the early 2000s it stopped working, because transistors became so small that they leaked current even when switched off. The voltage could hardly be lowered any further, and faster chips became too hot to cool, so in May 2004 Intel cancelled its next faster Pentium 4 and turned to chips with two processor cores.

According to a 2011 report by the US National Research Council, computer performance grew by more than 50 percent per year from 1985 to 2004, and much more slowly after that. Chips still gain transistors, but the extra ones go into more cores and special circuits, which help only with work that can be split into parts.

The second limit is physical. Transistors cannot shrink forever, because at some point their parts would have to be smaller than the atoms they are made of. Moore himself told engineers in 2003 that no physical quantity can grow exponentially forever.

The third limit is cost, which matters because the law was about cost from the start. Each new generation needs more expensive factories, so the price per transistor now falls much more slowly. In 2022 Nvidia’s chief executive, Jensen Huang, said that Moore’s law was dead and that chips would no longer become cheaper. Others disagree, because the transistor count still grows, partly by joining several chips in one package as Apple did with the M1 Ultra.

So the old combination of more, cheaper and faster transistors with every generation ended around 2005, while the doubling of transistors continues, more slowly and at a much higher price.

Learn more

Moore’s law has a clear father. Gordon Moore, born in 1929, earned a doctorate in chemistry at Caltech in 1954. He published the forecast in the magazine Electronics on 19 April 1965, predicting 65,000 components per chip by 1975, and in the same article he imagined home computers and personal portable communications equipment. He changed the pace to two years in a talk in 1975, and his friend Carver Mead, a professor at Caltech, gave the trend its name. Moore co-founded Intel with Robert Noyce in 1968 and died in March 2023 at the age of 94.

For one view of where ever more powerful computers might lead, read “Homo Deus” by Yuval Noah Harari: Summary and Review. For the history of chips and why governments fight over them, read Chris Miller’s Chip War (Affiliate Link).

Quiz

3 questions · up to 3 points

Open a question and pick one answer. Only your first answer counts: +1 if it is right, −1 if it is wrong.

  1. In 1975 Gordon Moore revised his forecast. What does the revised version of Moore's law say?
    Show answer

    The number of transistors on a chip doubles about every two years. In 1975 Moore slowed his 1965 forecast of a doubling every year to a doubling every two years, and this is the version people usually mean.

  2. A chip made in 2020 has 1 billion transistors. If the transistor count doubles every two years, about how many would a chip have in 2030?
    Show answer

    About 32 billion. Ten years hold five doublings: 2, 4, 8, 16 and 32. Each doubling builds on the one before, so the count grows by a factor of 32.

  3. Since about 2005, a single processor core has not become twice as fast every two years. What does that show?
    Show answer

    Transistor counts kept rising, but smaller transistors stopped also getting faster without producing more heat. Moore's law is about the number of transistors. What ended around 2005 was Dennard scaling, which had made each smaller transistor faster at no extra cost in heat.

The takeaway

Moore's law is Gordon Moore's observation that the number of transistors on a chip doubles about every two years. Starting from the 2,300 transistors of Intel's first microprocessor in 1971, that pace predicts about 77 billion in 2021, close to the 57 billion on Apple's largest chip that year. It is an industry target, not a law of nature: around 2005 chips could no longer get faster without also getting hotter, and the doubling has since slowed. When you read that something doubles every few years, ask what drives it and what will stop it.

Key quotes

“But no physical quantity can continue to change exponentially forever.”
Gordon E. Moore, Keynote at the International Solid-State Circuits Conference, San Francisco, 10 February 2003 (Intel transcript)

References

  1. Cramming More Components onto Integrated Circuits (reprint). Gordon E. Moore, Electronics 38 (8), 114–117; reprinted in The Future of Computing Performance, National Academies Press, Appendix C, 1965

    Moore's original four-page article, with the graph and the forecast of 65,000 components per chip by 1975.

  2. 1965: "Moore's Law" Predicts the Future of Integrated Circuits. Computer History Museum, The Silicon Engine

    A short history of the 1965 forecast, the five data points behind it and the 1975 revision to a doubling every two years.

  3. The Multiple Lives of Moore's Law. Chris A. Mack, IEEE Spectrum, 2015

    Why the law has always been about cost, how it became a self-fulfilling prophecy and how its meaning changed over time.

  4. The Future of Computing Performance: Game Over or Next Level? (Summary). National Research Council, National Academies Press, 2011

    Why single-processor performance stopped growing fast in the mid-2000s, and why power limits pushed the industry to multicore chips.

  5. Keynote at the International Solid-State Circuits Conference (transcript). Gordon E. Moore, Intel, 2003

    Moore looks back on his forecast, explains the 1975 change and says why no exponential trend lasts forever.

moores-lawgordon-mooretransistorssemiconductorsintegrated-circuitsmicroprocessorsexponential-growthdoubling-timedennard-scalingintelcomputer-historychip-industryself-fulfilling-prophecytechnology-forecasting