Chips: Breaking the Light Barrier

To make tinier, faster circuits, researchers are working with the extreme edge of the spectrum

Since the first Sumerian clay tablets 5,500 years ago, "the progress of mankind is directly related to our ability to store, process, and retrieve information," says G. Dan Hutcheson, president of chip-watcher VLSI Research Inc. in San Jose, Calif. And in recent decades, that progress has been remarkable. Like clockwork, the power of chips has been doubling every 18 months, transforming yesterday's dreams into real-world desktop supercomputers, vast storehouses of digital data, and increasingly smarter cars and home appliances.

Behind this relentless march of silicon technology are the steady advances in microlithography--the key process used to "print" ever-smaller circuits on silicon wafers. "Most of the microelectronics revolution is really the lithography revolution," says Lloyd R. Harriott, an electrical engineering professor at the University of Virginia. But optical lithography may reach its physical limits before the end of the next decade, threatening to bring the expansion of the Information Age to a sputtering halt.

"SPECTACULAR SUCCESS." That's why so many hopes are riding on a major leap in lithography that will be unveiled April 11 by an unusual consortium of companies, national laboratories, and academics, led by microprocessor king Intel Corp. (INTC ) Until now, circuit patterns have been printed on silicon with light, and today's lines have shrunk to 0.13 micron, or 130 nanometers. That's about 1/800th as wide as a human hair.

The new system will instead use so-called soft X-rays--now better known as extreme ultraviolet (EUV) radiation to distinguish the project from a different approach using regular X-rays. If it works as advertised, it could keep the microelectronics party going for at least an additional decade, into the 2020s, with circuit lines shriveling to a minuscule 35 nanometers or even less. If it doesn't, other contenders still under development may seize the stage, such as electron-beam lithography or advanced versions of today's optical technology.

For now, all eyes are on EUV. The prototype EUV machine, currently at Sandia National Laboratories' facility in Livermore, Calif., is a remarkable achievement, the result of some 13 years of work and more than $250 million in research-and-development funding. Early on, EUV lithography was plagued by unknowns and engineering brick walls, says Jeffrey Bokor, an electrical engineer at the University of California at Berkeley. "But we've invented our way past essentially all the obstacles. It's been a spectacular success--and the finishing line is in sight."

Yet the tale of this impressive system isn't just about technology. It is also an example of how a controversial approach to R&D can pay big dividends, with industry leveraging strategic government investments. Ironically, similar feats may now be difficult, thanks to Bush Administration plans to withhold funding for one of the programs that jump-started the project. The soft X-ray saga also offers a lesson in Washington mind sets: Congress allowed the new technology to be shared with a Dutch company, ASM Lithography (ASML ), but not to Japanese producers. "I learned that there is foreign and there is foreign--and that European foreign is less offensive than Japanese foreign," says Richard R. Freeman, a former team leader for the EUV consortium who's now at the University of California at Davis.

There are huge implications for business as well. If EUV does beat out competing electron-beam systems, Intel will take a major lead over rivals like Advanced Micro Devices Inc. (AMD ) and IBM (IBM ) because Intel will get first dibs on the machines. In addition, Japan's big lithography suppliers, Nikon Corp. (NINOY ) and Canon Inc. (CAJ ), could lose market share in the $5.4 billion business of selling chip-printing equipment--taking the same sort of hit they inflicted on U.S. litho suppliers in the 1980s. Indeed, the Japanese are now looking to form their own EUV consortium, says Shinya Sasayama, head of Nikon's chip-equipment division.

At its simplest, a microlithography machine is like a complex stenciling tool. It projects a circuit pattern through reducing lenses onto a light-sensitive coating on a silicon wafer. Where the light strikes, the coating hardens. Then the coating's soft regions can be etched away, leaving a maze of lines that further processing turns into millions of transistors and connections. For this to work properly, the color, or wavelength, of the light is crucial. "The shorter the wavelength, the better the resolution and the narrower the lines," explains Papken S. Der Torossian, chief executive of Silicon Valley Group Inc. (SVGI ), the main U.S. lithography company, which is now being acquired by ASML (though the deal has been delayed by national security concerns). That's why the history of lithography has been a steady journey down the spectrum of light, from a wavelength of 436 nanometers, or billionths of a meter, in the early 1980s, to today's workhorse 248-nm machines and on to cutting-edge tools with 193-nm light sources. Coming soon are 157-nm systems, which promise lines just hundreds of atoms wide.

However, semiconductor experts long ago realized that lithography using light would eventually run out of spectrum. So in the 1980s, the chip industry began exploring alternatives, from X-rays to beams of ions and electrons. The problem with X-rays is that they can't be focused by lenses, so the masks must be as tiny as the chips--and almost impossible to fabricate. But Bell Labs scientists decided that X-rays in the EUV range--around 13 nm--could be focused with mirrors instead (diagram). These mirrors would have to be tens of thousands of times more precise than anything ever built before. Federally funded collaborations with the national labs helped resolve that and other riddles, and AT&T (T ) got $2 million from the Commerce Dept., which it used to boost U.S. optics technology. "That was the best money the U.S. ever spent," says Freeman.

Then came a crisis. After winning control of Congress in 1994, Republican lawmakers axed funding for joint research among national labs and companies. By 1996, the EUV program was months away from shutting down when Intel stepped into the breach. Intel put together a consortium involving Sandia, Livermore, and Lawrence Berkeley national labs, along with chipmakers AMD, Infineon Technologies (IFX ), Micron Technologies (MU ), Motorola (MOT ), and equipment suppliers. Intel coughed up the lion's share of the $250 million budget--but drove a hard bargain. It won a guarantee that it would get the new machines first, before fellow consortium members. In the fast-paced semiconductor market, analysts say that even a three-month head start could bring in enough revenue to recoup all the EUV development costs. But Intel is counting on still more. "We may have it a year ahead--or as much as three years ahead," says Charles W. Gwyn, Intel's program director for the consortium.

To build a production-ready version of Sandia's experimental prototype, Intel in 1998 wanted to license the design to Japanese suppliers as well as SVG. But Washington politicians and Silicon Valley execs alike howled about handing the crown jewels to the Japanese. So the consortium brought in ASML. Despite the controversies, the consortium has made astonishing headway. For example, new methods of making the reflective masks had to be developed, along with quality-control techniques to spot 90-nm defects. "That's like looking for a golf ball in the state of Rhode Island," says Intel's Gwyn.

Now, Intel is getting antsy about seeing some return on its investment. It wants ASML and SVG to begin delivering the litho systems, which are expected to cost up to $40 million each, by 2005. Then Intel could begin fabricating chips that make today's Pentiums seem like Model Ts. It could also deliver a knockout blow to a rival electron-beam approach being developed by Nikon and IBM, although IBM has now joined the EUV group as well. "Intel is pushing us like hell," says SVG's Torossian. "It's like running in front of a locomotive."

Torossian thinks a more realistic time-table would be production-ready machines in 2007. But that would leave the door ajar for rival systems, which could level the playing field and enable IBM and others to produce circuitry just as supersmall as the equipment that Intel has funded. However things turn out, though, it now appears certain that humanity's ability to manipulate information will stay on the fast-growth track for years to come.

By John Carey in Washington, D.C., with Irene M. Kunii in Tokyo

    Before it's here, it's on the Bloomberg Terminal.