New Life for Moore's Law

After four decades of remarkably steady progress, advances in computer-chip technology seemed in danger of slowing. Not anymore

Interstate 580 connects Oakland, on San Francisco Bay, with Stockton and the farming towns of California's Central Valley. Heading east from Oakland, I-580 crosses the modest Coastal Range at the Altamont Pass. Altamont was famous a generation ago as the site of a Rolling Stones concert that turned into a riot. Now it features an eerie conjunction of two different kinds of "farming." Herds of cattle graze among thousands of pylons and derricks supporting giant propeller-shaped wind turbines—components of the world's largest "wind farm." When wind roars through the pass, the blades all turn, generating nearly enough electricity to power a city the size of San Francisco.

A series of low white buildings nearby adds to the sense that the area has been taken straight from a Back to the Future movie. These buildings, in the town of Livermore, house two national laboratories—Lawrence Livermore and the California branch of Sandia—where classified work on nuclear weapons and nuclear energy has been under way for decades. Usually the labs are anything but welcoming. Visitors cannot bring laptops or cell phones onto the premises; airplanes are restricted from flying overhead. But one day last spring the fences surrounding the laboratories were decorated with balloons, like a house proclaiming a child's birthday party, and hand-lettered signs directed visitors to a celebratory reception. The occasion was the announcement of a technical achievement that may have surprising economic and political consequences.

The celebration was for the room-sized Engineering Test Stand—the first working prototype of a new system for producing semiconductor chips. What makes the Engineering Test Stand special is its use of "extreme ultraviolet" light, or EUV. In the thirty-plus years since the first mass production of computer chips, progress has been defined as the ability to fit more and more circuits into increasingly smaller spaces. This has been accomplished mainly by devising finer and finer tools for etching circuits onto the surface of a chip. Just as a fine-point pen can put more words into a given space than a Magic Marker can, so the steady refinement of etching machines, or "steppers," has allowed the number of circuits per chip to go from a few thousand in the 1970s to tens of millions today.

However, throughout the past decade each move toward a better, finer stepper increased worries about how many more moves were possible. The fundamental limit to the size of circuits on a chip is tied to the wavelength of light used to etch the circuits. Existing chips were approaching that limit. X-ray beams have much shorter wavelengths than visible or ultraviolet light, but they are too destructive to be practical in etching. This is where EUV comes in.

EUV, which exists in the realm between visible light and x-rays, is a sort of domesticated x-ray. Its wavelength is short enough to allow the creation of extremely small circuits, and it is easier to work with than x-rays. But not that easy: EUV beams can't be transmitted through normal air, which absorbs them, and they can't be magnified or focused with normal lenses, which weaken and distort them.

The machine celebrated at Livermore works around these limitations. It transmits EUV beams through a vacuum, and it focuses them with extraordinarily smooth mirrors—ones whose surfaces have no irregularity greater in width than an atom. The wavelength of EUV is about a twentieth that of the light used in current steppers. Over time this should allow EUV lithography to create chips that have more than twenty times as many transistors as today's models and that run thirty times as fast.

From the archives:

"The New Old Economy: Oil, Computers, and the Reinvention of the Earth" (January 2001)
"Knowledge, not petroleum, is becoming the critical resource in the oil business" the author writes in this firsthand account of how technology is stretching the bounds of finitude. By Jonathan Rauch

The development of the EUV machine is the tech industry's equivalent of the discovery of a vast new oil reserve. To put it in the industry's own terms, it amounts to an extension of the principle known as Moore's Law. The modern computer and electronics industries have differed from most others in continually offering entirely new products—digital cameras, cellular phones—and continually driving down the price of what's already for sale. Behind these achievements lies Moore's Law. This is the assertion, by Gordon Moore, a co-founder of Intel, that the computing power available on a given chip (or, in a variant, the power available for a given price) will double every eighteen months. To imagine this principle in any other context is to understand how remarkable it is: suppose that cars got twice as much mileage, or dropped by 50 percent in price, every year and a half.

Moore made his assertion at the dawn of the semiconductor age, in the 1960s. Since the early 1990s he and everyone else in the technology business have wondered when some problem such as the wavelength limit would "revoke" his law. The EUV etching machine does not mean that computer chips can keep doubling in capacity forever. But it probably gives Moore's Law at least another ten years.

The EUV prototype proves a scientific concept, rather than solving all the practical problems that stand in the way of commercial production. And it is not the only hope for the continuation of Moore's Law. For example, last summer researchers at Intel announced that they had found a way to keep making circuits smaller through adjustments in "conventional" technology. But whatever the case, the organization of the EUV project may prove to be as important as its discoveries: it suggests how the tech economy may tackle expensive research projects in the future.

The EUV effort was launched, in 1997, as a kind of supercollaboration among entities that usually keep a clear distance from one another. Three national labs—Sandia, Lawrence Livermore, and Lawrence Berkeley—pooled scientists, who worked in a "virtual national lab." All of the money for research and development—some $250 million—came from private companies, ones that are ordinarily fierce rivals. Intel, the world's leading maker of logic chips, contributed—and so did AMD (originally Advanced Micro Devices), which is constantly launching technical, marketing, and even legal challenges to Intel's dominance. And so did Motorola and IBM and nearly every other important American semiconductor company. The cooperation was "wonderful, heartwarming and just phenomenal," Craig Barrett, the CEO of Intel, said in a speech at the lab. Now, he added, members of the consortium would use the results of their shared research to "beat each other over the head in the marketplace—which is as it should be."

The only notable absentees from the lineup were the Japanese firms Nikon and Canon, whose steppers are now used by Intel, AMD, and everyone else. They were left out because part of the impulse behind the EUV project was to stimulate the revival of a U.S.-based stepper industry. The politicians, mainly Democrats, who endorse such collaborations between government and business through "industrial policy" could therefore claim EUV as their success. They could, in fact, claim it as their second important victory of this sort. Ten years earlier the major American computer-chip makers had joined in a research consortium known as Sematech. The companies put up tens of millions of dollars for the development of new chip-making systems; the government matched their funds; and the results greatly improved the U.S. companies' position relative to Japanese and Korean competitors. Although it was mainly a Democratic effort, Sematech began at the end of the Reagan Administration. So Republicans can share the credit for it if they want to—and they can claim EUV as an even clearer vindication of their policies. After all, crucial parts of the EUV technology emerged from research on Ronald Reagan's Strategic Defense Initiative.

In practical terms the difference between the Democratic and Republican versions of the EUV story may not really matter. Gordon Moore has warned for years that the challenge in maintaining Moore's Law will be not just technical but financial. The products of the computer industry get cheaper and cheaper, but the installations necessary to manufacture them steadily mount in price. That companies as strong as IBM and Intel needed to pool resources for this project suggests that collaboration will be the rule—among companies and with the government. This will be so whether the results are credited to "industrial policy" or to "strategic defense."