"Shipping Junk"

On April 21, 1981, three men posed for a photograph at the dedication of a new GCA Corporation facility in Bedford, Massachusetts. Governor Edward J. King stood on one side. Senator Paul Tsongas on the other. Between them, beaming, was Milton Greenberg, a former wartime meteorologist who had cofounded Geophysics Corporation of America in 1958 and who, twenty-three years later, ran what looked like the most exciting machine-tool company in the United States. The cameras clicked. Within four years, almost everything that mattered about the photograph would be gone.

Greenberg’s company sat at a chokepoint nobody outside the chip industry had heard of. To shrink a transistor, you had to project an image of a circuit pattern onto a silicon wafer, and as features approached one micron the contact and proximity printers used through the 1970s could not hold the geometry. The answer was the wafer stepper, a tabletop-sized machine that exposed one die at a time, marched the wafer over by a precise distance, and exposed the next, repeating across the full surface. The stepper used a reduction lens and a laser-interferometer-controlled stage to register patterns to within tenths of a micron. It was the most demanding optical instrument anyone had ever attempted to build for volume manufacturing.

GCA had stumbled into the business almost by accident. In its early years it was a Bedford-based government contractor that did atmospheric physics and lunar mapping for the Air Force and NASA, the kind of company that built spy-balloon instruments and then quietly diversified. In 1959 it bought David W. Mann, a small precision-instrument shop in Lincoln, Massachusetts, that built ruling engines and astronomical measuring devices. Mann’s photorepeaters, originally used to make photomasks for chip lithography, gave GCA a franchise in semiconductor tooling that no one had really planned. Through the 1970s a small group of Mann engineers — Burt Wheeler on mechanics, Aubrey “Bill” Tobey, who first proposed turning the photorepeater into a wafer stepper, and Griffith “Grif” Resor, who led the development team — converted the photorepeater idea into something new. By 1977 they had a prototype. The first commercial unit shipped to IBM that year. In 1978 they unveiled the production tool: the DSW 4800, a g-line wafer stepper using a ten-power Zeiss reduction lens with a 0.28 numerical aperture, a ten-millimeter-square exposure field, and a list price of $450,000.

For about three years GCA owned the future of chipmaking. The DSW 4800 was the only production-worthy reduction stepper in the world. IBM bought it for the East Coast fabs that were laying down the early generations of dynamic random-access memory. Bell Labs bought it. Intel, AMD, National Semiconductor, Fairchild, and Texas Instruments — every American chipmaker that wanted to push past one micron sent purchase orders to Bedford. By 1980 GCA controlled more than 90 percent of the emerging market for 64-kilobit DRAM lithography, according to Christian Science Monitor reporting at the time. Revenue climbed from below $50 million in the mid-1970s to around $300 million in 1981. The stock more than doubled in a year, from 37 to 84. American firms collectively held about 85 percent of the global lithography-equipment market. GCA was the crown jewel of a sector the United States invented, dominated, and assumed it would keep.

The first crack came from Tokyo. Nikon had been making precision optics since 1917, originally for the Japanese navy and then for cameras and microscopes. In the mid-1970s, MITI, alarmed at how dependent its rising chipmakers were on American tools, funded the VLSI Project and tasked Nikon with developing a domestic stepper. Nikon’s engineers did what reverse engineers do: they bought a GCA DSW 4800 through normal commercial channels, took it apart in Tokyo, mapped every subassembly, and started over with their own optics and stages. By 1980 they had the NSR-1010G ready for shipment to NEC and Toshiba.

What Nikon built was not just a copy. Nikon made its own lenses. GCA bought lenses from Carl Zeiss in West Germany under a sole-source contract and never moved away from it. Nikon, with a vertically integrated optics arm, could iterate on glass and machine in lockstep. It also brought a manufacturing culture honed on cameras and microscopes, where tolerances were measured against thousands of identical units rather than dozens. When Nikon began shipping the NSR-1010G in volume in 1981 and 1982, customers in Japan reported a difference GCA’s executives initially refused to believe. Nikon’s machines ran longer between failures. By the mid-1980s, multiple fab managers were citing roughly an order of magnitude more continuous use hours from a Nikon stepper than from a comparable GCA tool before a service call. In a fab, where a stopped lithography step idles a hundred-million-dollar production line, that ratio is everything.

Inside the Bedford factory, the machines were getting harder to build than to design. The DSW 4800 was a hand-tuned instrument. Each one required weeks of alignment by senior technicians, and the calibration drifted in transit. Field-service engineers became a permanent feature of customer fabs, swapping Zeiss lens elements, retuning stages, and chasing thermal effects nobody had anticipated. GCA’s order book in 1981 was so far ahead of its production capacity that wafers of demand piled up while Zeiss struggled to deliver lenses on schedule. Nikon’s gain from the period, about 20 percent of the global stepper market in a single year, came as much from GCA’s missed deliveries as from any technical leap. By 1984, Nikon’s share roughly matched GCA’s at about 30 percent each.

Inside GCA’s headquarters, none of it looked like a crisis until it was one. Greenberg had built the company on government contracts and political relationships, and in the early 1980s he treated the stepper boom as a rising tide that did not need close management. Industry oral histories describe an executive culture in which the CEO spent more time with senators and governors than on the factory floor, and in which inventory and supplier discipline collapsed under growth. The company at one point discovered roughly a million dollars of Zeiss lenses simply lost in the inventory system, set aside for builds that had been cancelled or rerouted and forgotten. Returns piled up. Service tickets piled up. The factory shipped tools engineering knew were not yet aligned, on the theory that field service could finish the work at the customer’s site. The phrase that attached to the practice was that GCA was shipping junk.

The phrase was not metaphorical. By 1984 a fab manager in Silicon Valley who opened a crate from Bedford was as likely as not to find that the stepper inside would not hold focus across the field, that the laser interferometer needed recalibration before first exposure, or that the autofocus subsystem failed to repeat across wafers. Engineers learned to budget weeks of post-delivery integration into a GCA install. The same managers, opening crates from Nikon, saw machines that powered up, exposed test wafers within tolerance, and held that performance through normal production runs. Word moved through the small community of lithography engineers faster than any sales team could counter. The next purchase order went to Tokyo.

The 1985 chip-industry downturn finished what the quality problems had started. Worldwide lithography-equipment sales fell roughly 40 percent. GCA, financially overextended from its growth phase and caught with excess capacity, lost about $100 million across 1985 and 1986. Payroll was cut by roughly 70 percent, down to about a thousand employees. Greenberg stepped back from operating responsibility in 1984 and from the chairmanship in 1986. GCA’s first deep-ultraviolet stepper, developed for Bell Labs in 1985, was a respectable piece of engineering and went almost nowhere commercially because the customer base was disappearing. By 1987 GCA was looking for a buyer; in March 1988 the conglomerate General Signal Corporation bought it for about $76 million, a fraction of the market value the stock had implied at its peak.

While GCA was unraveling, two other things were happening that would define the next thirty years of lithography. In Japan, Canon, which had begun stepper development later than Nikon and from a weaker optics base, pushed in with the same patient quality discipline and by the late 1980s had the second-largest share of the global market. And in the Netherlands, in a building borrowed from Philips’s Strijp campus in Eindhoven, a fifty-fifty joint venture between Philips and Advanced Semiconductor Materials International was incorporated on April 1, 1984, under the name ASM Lithography. Its mandate was to commercialize a stepper, the PAS 2000, that Philips had developed for its own internal fabs and concluded was too far from its core business to keep. The new company moved into a purpose-built factory in nearby Veldhoven in 1985. It had about a hundred employees. It launched the PAS 2000/10 g-line stepper in 1985 and the more competitive PAS 2500/10 in 1986. ASM International, struggling financially, would sell its stake back to Philips in 1988. Almost no one in the American industry, watching Nikon eat GCA’s market, paid attention to the Dutch.

In Washington, the GCA collapse landed during a moral panic about Japan. The Pentagon’s Defense Science Board had begun warning that the United States could not run an offset strategy built on chips it no longer made the equipment to fabricate. SEMATECH, the consortium of US chipmakers and the Defense Department launched in 1987 to rebuild the manufacturing-tool ecosystem, took GCA on as a flagship project. Robert Noyce, by then SEMATECH’s first chief executive, framed the decision in stark terms. Testifying before a House subcommittee on November 8, 1989, Noyce told members that SEMATECH would in large part be judged by how successful it was in saving America’s optical stepper makers. From 1989 through 1993, SEMATECH and General Signal together poured between $60 million and $75 million into Bedford to develop a new generation of deep-ultraviolet steppers, the XLS line.

The technology mostly worked. The XLS 7800 shipped in 1992 at a list price of $3.5 million, and on its merits it was a credible high-end stepper. The customers were not there. Intel, IBM, and the surviving US fabs had been buying Nikon and Canon for half a decade and had built process recipes, training materials, and supply contracts around them. Switching cost was real, and switching back was harder than not switching in the first place. A chokepoint, once relinquished, does not refill itself just because the country that lost it would prefer it back. Noyce did not live to see the outcome; he died of a heart attack at his home in Austin on June 3, 1990, at age sixty-two. The bailout he had sponsored survived another two and a half years.

In January 1993, General Signal announced that it intended to divest its semiconductor-equipment subsidiaries. By May, when no buyer materialized for GCA, General Signal simply closed the Bedford operation. A management buyout in June recovered the service business as Integrated Solutions, Inc., to keep installed AutoStep tools running. The lens-making subsidiary, Tropel, was bought out separately in 1994 and would be acquired by Corning in 2001 for $190 million, more than twice what General Signal had paid for the entire GCA. The stepper business itself, the original American chokepoint, ceased to exist.

The American share of the lithography-tool market did not stop falling. By 1988, US firms held about 50 percent of the global market by revenue, down from roughly 85 percent a decade earlier. Through the 1990s, with GCA gone and Perkin-Elmer’s lithography division surviving only as a struggling Silicon Valley Group, the share dropped into the teens. Nikon and Canon split most of the high-volume market. ASML, building patiently in Veldhoven and shipping the PAS 5500 series through the 1990s, climbed from a footnote to a peer. In 2002, ASML passed Nikon to become the world’s largest lithography-tool supplier. By 2025 it would be the only company on Earth capable of building the most advanced lithography machines used in the most advanced fabs. None of them were built in the United States.

What was lost in Bedford was not just a company. It was a category of capability: a workforce of optical engineers, mechanical designers, software developers, field-service technicians, and the supplier base that fed them, built up through twenty years of contracts with IBM and Bell Labs and Intel, and then dispersed in a few years of mismanagement and overconfidence. The engineers found new jobs. Some went to Ultratech, some to other equipment makers, a few eventually to ASML’s American offices. The institutional memory broke. The relationships with Zeiss, the most demanding optics partnership in the industry, never returned to American hands and would later become one of ASML’s deepest moats. The supply chain that grew up around Nikon and Canon in the 1980s, and around ASML in the 1990s, never had an American counterpart to compete with again.

Greenberg lived another fifteen years. He died in 2001, having spent his last decade running an unrelated company. He was, by every account, a charming man and a capable founder, and he had built one of the most important manufacturing companies in postwar American technology. He had also presided over its destruction in plain sight, while better-disciplined competitors learned from his machines and out-built him.

The story is sometimes told as a parable about Japan’s industrial culture, sometimes about American executive complacency. It is both. It is also something more specific. GCA’s collapse demonstrated that in a chokepoint industry, where one machine sits between a wafer of silicon and the next generation of computing, the loser does not get a second chance. The customers are too few, the switching costs too high, the institutional knowledge too tacit, the supplier relationships too embedded to recreate from scratch. By the time SEMATECH and Noyce decided to save GCA, GCA had already lost the only thing that mattered, which was the order book. After Bedford, the question was no longer whether the United States would lead the world in lithography. It was who, in Japan or in a small town outside Eindhoven, would replace it.

The Dutch did not yet know they were the answer. Neither did the Americans who had stopped paying attention.