clock menu more-arrow no yes

Moore's Law Hits 50, but It May Not See 60

Intel's CFO says it can keep turning out chips with less expensive transistors for a few more generations. Science says it can't go on forever.

Intel

Moore’s Law turns 50 years old this Sunday. It may not make it to 60.

Since Intel co-founder Gordon Moore first made the observation in a 1965 paper published in Electronics Magazine that the number-crunching capacity of our computers and the number of transistors on silicon chips double roughly every two years, that prediction has underpinned the unrelenting tide of technological progress.

The computing industry has benefitted directly from the chip industry’s steady march: Every new chip is smaller than the one that came before it, and at least as powerful but less expensive to make. In 2015 transistor counts on Intel chips number in the billions; in his paper, Moore predicted 65,000 by 1975.

The physics involved in making chips has its limits, however, and companies like Intel are coming close to reaching them. Even Moore himself said so in a 2010 interview. Once those transistors shrink so small that they’re the size of individual atoms, you can’t go any smaller, he said. From there the only obvious answer to increase computing power is to reverse course and start making chips that are bigger, and thus more expensive.

Few people are more aware of this than Stacy Smith, the CFO of chip giant Intel and the person whose job it is to ensure that the engineers at Intel keep Moore’s Law alive. In an interview with Re/code following the company’s earnings report Tuesday, he said that as of now, Intel can see ahead only a couple more generations in chip-making technology. Beyond that, he would say no more.

“Moore’s Law is as much an economic observation as a physical one,” Smith said. He observed that “the cost per transistor comes down over time at such a pace that you can continue to improve performance.”

The end of the line is now a real concern as Intel undergoes the next phase of what once appeared to be a never-ending progression of chip miniaturization. As of the end of its fiscal first quarter, more than half of the chips it shipped were built on its new 14-nanometer manufacturing process, supplanting its older 22-nanometer technology, which is slowly being ramped down.

For some perspective, a nanometer is a billionth of a meter. Printed on paper, the period at the end of this sentence would be about one million nanometers. An object of 14 nanometers in size is smaller than a typical virus cell particle, and about the equivalent to the thickness of the outer cell wall of a typical germ.

Looking ahead, Smith says, Intel can realistically see building chips on 10 nanometer technology, though he was careful not to say when it might start. If it holds to its typical schedule, Intel would probably start building 10-nanometer chips for production sometime late next year.

Intel has also taken what Smith called “an early look” at seven nanometers. “We can see through to seven nanometers and our early look tells us that we can continue to bring down the cost of transistors with historical rates of improvement,” he said. Smith didn’t say it, but if history is any indicator, seven-nanometer technology should arrive sometime in 2018, give or take.

Beyond that, things start to get fuzzy. The next milestone is the five-nanometer node, which history suggests should arrive sometime in late 2020. (Five nanometers is about twice the size of a strand of DNA.) Assuming the typical two-year product cycle, that gets us to 2022. A lot of people who know about these things have predicted it is at about this time, some three years short of its 60th anniversary, that Moore’s Law will reach its logical end.

Here Smith deferred to his boss, CEO Brian Krzanich. “Brian would say that the difference between seven and five is that it’s only going to get harder, and there will be fewer and fewer people able to do it.”

What will Intel do? He didn’t say, but that doesn’t necessarily mean that Intel doesn’t have a long-term plan. After all, it spent $11.5 billion on research and development last year, up from $10.1 billion in 2012 and another $10 billion and change on capital expenditures. Few can keep up with that. A decade ago the number of companies with “leading edge” chip manufacturing capacity was 18. Today there are only four: Intel, Samsung, GlobalFoundries and Taiwan Semiconductor Manufacturing Corp.

And hope springs eternal among scientists: Researchers around the world have demonstrated transistors built at the three-nanometer and one-nanometer scale. And in 2012 a team in Australia built a transistor from a single atom of phosphorus. But these demonstrations under strict laboratory conditions remain far away from mass scale production.

There are also other materials that might be used to replace or enhance silicon in making chips; silicon germanium and just plain germanium are occasionally mentioned, as is graphene.

Intel hasn’t yet said how it intends to get to five nanometers and beyond. It has at times hinted that a technology called Extreme UltraViolet Lithography, which has been on the drawing board for years, might help it get there. But EUV technology is expensive, complicated and among other things, it only works in a vacuum.

Intel has proven me and other Moore’s Law doomsayers wrong on more than one occasion. But it won’t be easy.

For now, listen to Moore, now 86 years old, talk about the paper he wrote and its long-term impact.

https://www.youtube.com/watch?v=ylgk3HEyZ_g

This article originally appeared on Recode.net.