Moore's Law at 50: Defining the Speed of Change

That concept has been a major driver in our technological world, as we moved from integrated circuits that contained less than 50 transistors and resistors 50 years ago up to today's chips, where the new Intel "Broadwell" dual-core Core chips for laptops have 1.9 billion transistors and the high-end Xeon chip has 4.3 billion transistors. We've seen pretty amazing progress, and it has led to us having mobile phones that have the equivalent power of supercomputers from not too long ago.

Titled "Cramming More Components onto Integrated Circuits," Moore's original article appeared in the 35th anniversary issue of Electronics Magazine, dated April 19, 1965. (A reprint is online here.) In the paper, Moore noted that "the complexity for minimum component costs has increased at a rate of roughly two per year," meaning that the number of transistors per chip was doubling every year. There was even a graph showing how this would extend for the next 10 years.

To figure this out, Moore says he went back to the development of the initial planar integrated circuits in 1959, and plotted the number of components on a chip in the intervening four years on semi-log paper. He noticed that "Aha, it's been doubling every year." (Moore has told the story many times, including in a 1997 interview with me for PC Magazine and a recent interview with Intel.)

An upcoming biography of Moore suggests that he was actually thinking in similar lines two years earlier when he wrote an earlier paper, but it was the Electronics paper that introduced the concept of a regular doubling in components.

In the paper, Moore predicted that by 1975, "the number of components per integrated circuit for minimum cost (would) be 65,000"—a huge increase, but one that turned out to be very close to what engineers actually achieved.

At the time of the original article, Moore was running R&D at Fairchild Semiconductor, where he was one of the co-founders. He and Robert Noyce left Fairchild to form Intel in 1968, and the company has been pretty much defined by its commitment to continuing the doubling of transistor density on a regular basis.

The phrase "Moore's Law" was coined by Caltech professor Carver Mead about 10 years after the article appeared, and it stuck, even though Moore himself resisted the term for years.

In 1975, Moore updated his projection to a doubling every two years, and for most of the intervening years, we've seen chip makers trying to hit that projection. For years, Intel was introducing new processor nodes on a regular two-year schedule with its "tick-tock" cadence, and though the more recent 14nm and 16nm nodes have been a bit behind, the concept continues to drive the chip industry. Among those companies are Intel, the semiconductor foundries that make chips for other companies (such as Globalfoundries, Samsung, and TSMC), and the various memory makers (although NAND flash makers have recently moved from trying to get more dense planar chips to 3D NAND chips).

It's important to note that Moore's Law isn't a physical law – instead, it's more of a prediction of how fast the industry will move; and a goal that the industry tries to meet, spending billions of dollars to research, design, and manufacture new and increasingly complex chips.

How long will Moore's Law continue? No one really knows. Intel's current CEO Brian Krzanich has said that "it's our job to keep it going as long as possible." Along the way, chip makers have developed new materials and structures (such as high-k/metal gate and strained silicon) and new structures such as FinFETs or, as Intel calls it, Tri-Gate technology. At this point, all of the 14nm and 16nm logic manufacturing uses these tools along with multi-patterning optical lithography – in short, it's gotten more difficult and more expensive, but Moore's Law continues.

Recently, Intel and companies like Samsung and TSMC have started investing in 10nm manufacturing and we'll likely start seeing the first 10nm products in 2017 or so. Intel has said that it believes 7nm manufacturing would not only happen, but would continue to show a drop in the cost per transistor, and most of the chip folks I've talked to are convinced that 5nm manufacturing will follow, though it's unclear how much these new nodes would cost or whether a two-year cadence is still possible or efficient. To move forward, over the next few years, we'll probably need to use new materials such as Silicon Germanium or what are called III-V compounds; new structures, such as gate-all-around or nanowire technology; and new lithography tools such as extreme ultraviolet (EUV) tools.

As Moore said in the more recent interview, "In 1965, and when I updated my observation in 1975, I didn't predict when this trend was going to end. It's a good thing because I'm sure I would have been surprised. The industry has been phenomenally creative in continuing to increase the complexity of chips. It's hard to believe – at least it's hard for me to believe – that now we talk in terms of billions of transistors on a chip rather than 10s, hundreds or thousands.

"It's a technology that's been much more open ended than I would have thought in 1965 or 1975. And it's not obvious yet when it will come to the end."

Moore's Law has driven the technology industry forward for the past 50 years, enabling the amazing changes in electronics and related technologies we've seen over that period, from PCs to smartphones to communications and digital TVs. It's hard to predict what new things it will bring in the future.