Science & Technology
ONE STEPPER FORWARD FOR SEMATECH
The revival has been five years in the making, but the U.S. may finally be about to retake some ground in the technological war with Japan. Back in 1987, 14 U.S. electronics companies formed the Sematech consortium and vowed to restore America's competitiveness in the esoteric equipment used to make integrated circuits. On June 15, one key product of Sematech's initiative will be unveiled by Silicon Valley Group Lithography Systems Inc. (SVGL) in Wilton, Conn. Its new tool for "printing" millions of transistors on microchips shows promise of surpassing everything the Japanese offer. "We have a technology that will bring superiority back to the U.S.," declares Robert Virgalla, an SVGL marketing manager.
Called Micrascan II and developed partly with Sematech funding, the SVGL machine is already winning plaudits from analysts and potential customers. "It's one of the few technologies that can take us to the 256-megabit chip," says Jerry D. Hutcheson, chief executive officer of VLSI Research Inc., a San Jose (Calif.) consultancy. Computer memory chips that hold 256 million bits of data--roughly the text in 200 issues of this magazine--should show up around 1997. Well before then, IBM, another SVGL backer, plans to buy dozens of Micrascans for making 64-megabit chips, which could start production trials by late next year. Industry sources expect Toshiba Corp., a leading supplier of memory chips, to place an order as well.
The Micrascan isn't the only sign of a U.S. rebound in this pivotal technology. GCA Corp. in Andover, Mass., has also developed advanced circuit-printing tools with Sematech's help, and it recently logged a major order from Digital Equipment Corp. And Ultratech Stepper Inc. in Santa Clara, Calif., is working on technology for 256-megabit chips. "We are starting to see American products that can compete with the Japanese," says Peggy Wood, who follows semiconductor equipment for market researcher Dataquest Inc.
STAYING AMERICAN. There's a lot of territory to recover before Sematech can be judged a success. But this experiment with industrial policy, which has been costing taxpayers $100 million a year in matching funds, is looking like an antidote to earlier efforts by Japan's Ministry of International Trade & Industry. Until the early 1980s, U.S. companies dominated the market for hardware used to etch ever-thinner circuit lines on silicon wafers. But by then, MITI had picked Nikon Corp. and Canon Inc. to develop better systems. Just as the Japanese were doing that, the U.S. pioneers dropped the ball. An executive at GCA, No. 2 in the world in 1980, admits that it failed to keep pace and alienated customers with poor reliability and service. GCA soon fell off the top-10 chart.
Meanwhile, Perkin-Elmer Corp. in Norwalk, Conn., No. 1 in 1980, missed the crucial move away from older circuit-printing machines. By 1988, PE had plummeted to No. 8 in the world, and management despaired of making a go of a high-risk innovation being developed with IBM's help. So in early 1989, PE put its lithography unit up for sale. When Nikon emerged as the only serious suitor, U.S. electronics companies rallied to the flag. The Bush Administration urged PE's chairman to find a U.S. buyer. As Nikon retreated, Silicon Valley Group Inc., then a $131 million supplier of other wafer-processing equipment, stepped in. With IBM's help, SVG got PE to settle for a fraction of its $100 million asking price--$28 million--plusa 20% share in a new company, SVGL.
At first glance, the fuss seemed out of proportion. The U.S. market for all microlithographic gear was only $314 million last year. But it's a critical link in the technological food chain of the $31 billion semiconductor industry, which in turn supports the $310 billion U.S. electronics industry. IBM and other heavyweights worried that if Japan cornered the lithography business, its chipmakers might get dibs on new steppers, as today's chip-printing machines are called. Then they could give their computer divisions chips that would hand Japanese computers a cost or performance edge.
This could still happen. SVGL and GCA are dwarfs compared with Nikon and Canon. Of the 679 steppers shipped last year, at $1 million to $4 million each, more than 80% were Japanese-made, according to Dataquest. Advanced Semiconductor Materials Inc. (ASM), a Dutch supplier partly owned by Philips, had a 6% share. By comparison, SVGL shipped just eight of its first-generation Micrascan units, which enjoy only a small advantage over the Japanese. Chipmakers such as Intel Corp. and Motorola Inc. are evaluating the latest model but wonder if it can match Japanese reliability. U.S. companies "are back in the running," says Howard I. High, an Intel spokesman. "But they aren't at a point where we're ready to buy."
SPLITTING HAIRS. A stepper fabricates tiny chips with millions of transistors, making it perhaps the most exacting factory tool ever. It shines light through a mask, or stencil, of the circuit pattern to imprint the image on a photosensitive layer covering a silicon wafer. Then the wafer is moved one step ahead--hence, the name "stepper"--to print the next group of chips. Current steppers can print circuit lines at least as thin as 0.8 micron, or 1/125 the diameter of a human hair. That's fine for 4-megabit memory chips. However, the 16-megabit chips now coming to market require shrinking the circuit lines to 0.6 micron or less, and 64-megabit memories will need lines slimmer than 0.4 micron.
To pass the Japanese, Perkin-Elmer planned to marry steppers with an earlier technology, the scanner. Instead of exposing circuits with gne flash, a light source scans a moving mask (diagram, page 110). This approach has a 50%-plus larger "field of view," so three chips typically can be printed with each exposure, vs. two with a stepper. The higher output saves money. And an innovative new light source, called "deep-deep ultraviolet," promises to shrink step-and-scan line widths to below 0.25 micron, the target for 256-megabit memories.
Building such a machine presented major problems. Especially tough was keeping the light beam in tight focus. When PE threw in the towel, IBM, which long ago stopped making its own chipmaking equipment, wasn't willing to tackle the job alone. Papken S. Der Torossian, chairman of SVG and a passionate advocate of keeping American innovations in the U.S., stepped forward. But he didn't agree to proceed until IBM promised to buy 19 first-generation machines for $76 million. "The major hurdle was convincing SVG that this could be a money-making proposition," says Michael J. Attardo, Big Blue's chief of semiconductor manufacturing.
The deal took five months to structure and nearly collapsed many times, recalls Der Torossian. Eventually, IBM also agreed to pay $20 million for research and development, and Sematech added $20 million more. In late 1990, the Defense Advanced Research Projects Agency tapped SVGL for a project whose aim is etching even finer lines. Says David C. Shaver, head of the effort at Massachusetts Institute of Technology's Lincoln Laboratory: "DARPA wants this industry healthy."
Soon after SVGL got up a head of steam, there was "a major scare," recalls SVG President Vahe A. Sarkissian. By March, 1991, engineers had produced lenses that could print lines just 0.35 micron thick. But the lenses also left ghost images--faint lines in the wrong place. "The sky was falling," says Sarkissian, until engineers solved the problem by switching to purer glass for the lenses.
All told, SVGL has delivered 13 first-generation step-and-scanners to IBM. Rated at a 0.5-micron line width, they make IBM's leading-edge 16-megabit memory chips and are being pushed to 0.4 micron to fashion prototypes of 64-megabit chips. Another machine is at Sematech, where it's being sized up by engineers from companies such as Texas Instruments Inc. and Motorola. Last year, Toshiba bought the only unit sold to a non-U.S. company. Now, SVGL is gearing up to produce the Micrascan II, which provides 0.35-micron lines. And if the DARPA-funded project is successful, the next SVGL machine will reach 0.18 micron.
BIG YEAR. Such performance catapults the U.S. back into the technological lead, though Nikon and Canon officials note that only IBM has placed many orders. The Japanese also say it may take forever for GCA, which now offers a 0.35-micron stepper, to live down its mistake-ridden past. And they say it will be five years before chipmakers need to go below 0.35 micron, vs. the 0.5 that the Japanese now achieve. By then, Nikon and Canon will be ready, too.
Talk like this reminds the Americans of their own mistakes. "If the Japanese don't take this competition seriously, they'll be on the path to losing market share," declares Gordon B. McMillan,Sematech's lithography chief. He sees the next 12 months as pivotal. That's when big producers will begin ordering equipment for their "sub-half-micron" chips. Sematech is trying to prove to Intel, Motorola, and other prospective customers that the new U.S. tools are as reliable as they are innovative. If it succeeds, then one key U.S. industry--with a little help from taxpayers--may stand up to the Japanese juggernaut.John Carey in Washington, with Otis Port in New York