Gordon Moore drew a line through five data points. That was the whole method. The points showed how many components fit on an integrated circuit at minimum cost per component between 1959 and 1964, and Moore, then director of research and development at Fairchild Semiconductor, pushed the line forward to 1975.
His paper, “The Future of Integrated Electronics,” began as an internal Fairchild document. It was edited for print and appeared in Electronics magazine on April 19, 1965, under the title “Cramming more components onto integrated circuits.” Moore projected 65,000 components per chip by 1975, a doubling every 12 months, and he described the exercise as a look at roughly the next ten years.
From forecast to yardstick
The forecast landed. At the 1975 IEEE International Electron Devices Meeting, Moore, by then at Intel, credited better photolithography, bigger wafers, process advances and what he called “circuit and device cleverness,” especially in memory arrays. Then he adjusted the pace. With more microprocessors in the mix, and those being less dense than memory, he now saw “a doubling every two years, rather than every year.”
Carver Mead gave the idea its name, Moore’s Law, and by 1970 people were already using it. It stopped being a prediction and turned into a target. The Computer History Museum calls it a self-fulfilling prophecy: engineers were expected to deliver yearly breakthroughs that kept the industry on the curve.
Moore checked in again in 1995, when an Intel Pentium held nearly 5 million transistors. His verdict then was that the trend was “not going to stop soon.” Devices with more than one billion transistors now exist.
What the law means, and what it does not
Wired’s Dylan Tweney noted in 2010 that the law is often misunderstood. Moore’s paper concerned transistors that could be made cost-effectively on one chip. Transistor count is a rough stand-in for processing power, so people stretched the law to mean power doubling every two years. The flip side is that the cost of computing power falls by about half in that time.
Tweney offered a snapshot. The Core i7 in top-end laptops and desktops carried 731 million transistors, more than double the 291 million in the Core 2 Duo of two years earlier. Faster hardware does not always feel faster, though, because software grows to use whatever speed it gets.
Where it runs into limits
Tweney pointed to the physical problems. Circuit features had shrunk from 90 nanometers to 45 nm to 32 nm, with 22 nm coming. Keep going and components are only a few atoms across, where quantum uncertainty can swamp the accuracy of the devices. Heat is a second problem: a few hundred million transistors packed into the space of a pinkie fingernail make plenty of it.
Money is the third. The cost of chip factories climbs exponentially, a pattern sometimes called Moore’s Second Law, and a new plant then cost at least $3 billion. Each one is a huge bet on the business. Some designers answered by getting more out of less, with low-power netbook chips and the iPad, whose interface feels quick despite a modest processor.
Even Moore, Tweney wrote, knows his law has a limit. The open question was always where it sits.
Sources
- 1965: "Moore's Law" Predicts the Future of Integrated Circuits | The Silicon Engine | Computer History Museum
- April 19, 1965: How Do You Like It? Moore, Moore, Moore | WIRED
