Death Spiral

In a windowless conference room at the Pentagon on a Thursday afternoon in February 1987, Norman Augustine handed Caspar Weinberger a 103-page document and waited. The defense secretary had been briefed on its outlines for weeks. Augustine kept his pitch short. The American semiconductor industry was failing. The military depended on it. Without intervention, within a few years the Pentagon would be buying its critical chips from Japanese suppliers who had already proven willing to sell sensitive milling machines to the Soviets. The task force he had chaired recommended two billion dollars of public money over the next five years, half of it for a new manufacturing consortium to be built from scratch, half of it for university research and a string of supplier programs. “If we spend a year talking about this and not getting under way,” Augustine told a reporter that same week, “we’ll reach a point where we don’t have a recoverable situation.”

Charles Fowler, the Defense Science Board’s chairman, had attached a one-paragraph cover memo. The report, he wrote, focused on a national problem that “at some time in the future may be looked upon in retrospect as a turning point in the history of our nation.” The line was unusual for a Pentagon document. The Defense Science Board did not normally speak in the cadence of historians. The people who had sat through the task force’s hearings believed the cadence was warranted. They had watched American merchant DRAM producers go from a clear majority of the global market to a rump of two firms in less than four years. They had heard Japanese trading-company executives in private briefings talk about chips the way OPEC talked about oil. They had read internal memos from Pentagon procurement offices warning that certain weapons systems already depended on Japanese components. They had begun to understand that in some quiet corner of the postwar order, the United States was about to lose, on a single industry, the capability to defend itself.

The fear in early 1987 was specific. The American semiconductor industry was no longer dying in slow motion. It was vanishing.

The dynamic random-access memory chip, or DRAM, was the most measurable case. DRAMs were the workhorse memory of every personal computer, mainframe, and minicomputer in the world. They were also a brutal commodity, sold by the bushel, with margins that vanished in a downturn. In the early 1980s the United States had had at least eight serious DRAM producers: Intel, Mostek, AMD, Motorola, Texas Instruments, National Semiconductor, Inmos, and Micron Technology, the small Idaho startup bootstrapped out of a dentist’s basement only a few years earlier. By the spring of 1987 the list had collapsed. Intel, under Moore and Grove, had walked away from memory entirely in 1985, closing plants and laying off a third of its workforce in what Grove later described as wandering in the valley of death. Mostek had been sold by United Technologies to the French firm Thomson for seventy-one million dollars and folded into a European consolidation. Inmos, the British DRAM hopeful, had watched three of its key engineers walk into a Boise office building wearing Micron badges. National Semiconductor had quietly stopped competing in volume. AMD had pulled back. By the time the Defense Science Board met, the only remaining merchant U.S. producers of commodity DRAMs were Texas Instruments, which had retreated to a niche, and Micron, which most analysts assumed would not survive the next downturn. IBM still produced enormous volumes of memory inside its East Fishkill fab, but only for IBM machines; not a chip of it reached the open market.

The numbers told the story. In 1975 the United States had held just under sixty percent of the world semiconductor market. By 1985 the share was below fifty. By 1986, on Defense Science Board figures, it was forty-three percent against Japan’s forty-six. For the first time since the transistor had been invented at Bell Labs, the country that had invented the industry no longer led it. That fact alone, repeated by Augustine in conference rooms across Washington, was enough to break a three-decade consensus that government should keep its hands off Silicon Valley.

The CEOs who had been quietly building a Washington presence since the early 1980s now had a draft of the document they had been waiting for. The most aggressive of them was Sporck, who had been one of the original founders of the Semiconductor Industry Association in 1977 and had spent so much time in Washington over the previous two years that his board had openly worried he was neglecting his own company. Later observers, including Bill Spencer, would say bluntly that Sporck had nearly let National go under in his determination to save the industry. His instinct was that no individual American firm could outspend MITI’s coordinated industrial policy, and the only response was a coordinated American one. The Defense Science Board report, which proposed exactly that, gave him the legitimating document he needed to take to Capitol Hill.

Beside him stood Galvin, whose company had been picked apart, product by product, in markets it had once dominated, and who had partnered with the former trade representative Robert Strauss to push the 1986 semiconductor agreement and personally lobbied Reagan’s cabinet for a coordinated response. Where Sporck was the operating man’s evangelist, blunt and unembarrassed, Galvin was the institutionalist, comfortable in committee rooms, willing to spend years building bipartisan support.

Behind them stood a thinner but more credentialed bench. Bloch, at the National Science Foundation, was the most influential voice inside the executive branch arguing that something larger was needed; the Semiconductor Research Corporation he had helped Noyce stand up earlier in the decade gave him a working bridge between the federal science establishment and the chipmakers. Ian Ross, the British-born president of AT&T Bell Laboratories, would within two years agree to chair the National Advisory Committee on Semiconductors, a body Congress would create in the 1988 Omnibus Trade and Competitiveness Act. Sporck, Galvin, Bloch, Noyce, Ross, and Augustine formed the personnel core of what would become the most ambitious peacetime industrial policy intervention in American technology since the Apollo program.

The popular discourse had been priming the public for several years. Ezra Vogel’s 1979 book Japan as Number One, published when the Japanese economy was growing at ten percent a year and the Tokyo land registry valued the Imperial Palace grounds at more than the entire state of California, had argued that Americans should learn from Japan’s industrial policies. The book had been a Harvard hit; by the mid-1980s its title was shorthand for a generation of business books and magazine covers. James Fallows in The Atlantic, Pat Choate in his lobbying-policy critiques, Chalmers Johnson at Berkeley, and Clyde Prestowitz in Trading Places (which Basic Books would publish in 1988 from the manuscript Prestowitz had been drafting through the 1986 negotiations) formed what came to be called the revisionist school of Japan analysis. They argued that Japan’s economic miracle was not a market accident but the product of an industrial state, that Americans negotiating with that state were systematically outmatched, and that the United States needed its own, more honest version of industrial policy. The arguments were contested at every step by orthodox economists. They were also inescapable in any room in Washington that mattered.

Inside the Reagan administration, where free-market ideology was an article of faith, the policy response remained politically delicate. The president himself was not the obstacle. Reagan had already, in March 1987, imposed hundred-percent tariffs on Japanese consumer electronics in retaliation for the failure of the semiconductor agreement, an act that surprised even his own advisers given his record. The obstacles were the ideological holdouts who had been blocking action for two years: OMB, Treasury, and Sprinkel’s Council of Economic Advisers. The argument in those rooms was not whether Japanese competition was painful. It was whether the right American answer was to subsidize or to wait for the market to discipline both Japanese overinvestment and American complacency.

The argument that broke the logjam was the one Augustine and the Defense Science Board had constructed. The chip industry was not just a sector. It was the foundation of the offset strategy the Pentagon had been pursuing since the late 1970s, the bet that American precision would outclass Soviet mass. Lose the chip industry and you lost the offset; lose the offset and you lost the conventional advantage in Europe; lose that and the Cold War balance shifted in ways nobody in the administration was prepared to underwrite. Reagan’s free-market instincts could be overridden by national-security logic in a way they could not be overridden by trade-policy logic alone. By the spring of 1987 the policy question inside the White House had narrowed from whether to act to how to package the action so it did not read as the abandonment of free trade.

The vehicle the industry had been preparing was already named. SEMATECH, short for Semiconductor Manufacturing Technology, had been incubating inside the Semiconductor Industry Association as a concept since the previous summer. The plan was simple. Fourteen American chipmakers would each contribute dues, calibrated to one percent of their semiconductor revenue with a floor of one million dollars and a ceiling of fifteen million. They would together fund and operate a shared research and development facility, a manufacturing pilot line capable of running advanced processes, and a coordinated program to fix the American supplier base on which they all depended. The federal government, through DARPA, would match the industry’s contributions dollar for dollar. The total commitment would come to roughly two hundred million dollars a year for five years, with one hundred million of that coming from the Pentagon. The five-year federal envelope worked out to five hundred million dollars, the figure that would headline almost every newspaper account of the program for the next decade.

Congress wrote the authorization into the National Defense Authorization Act for fiscal years 1988 and 1989, signed by Reagan in December 1987. The legislative carriage was the Senate Armed Services Committee under Sam Nunn of Georgia, with Jeff Bingaman of New Mexico as the most active advocate. Both senators understood that a defense bill could carry industrial-policy content a stand-alone trade bill could not. The committee report described SEMATECH not as an industrial subsidy but as a defense-technology-base measure, a description neither inaccurate nor accidental. A separate provision created the National Advisory Committee on Semiconductors, the body Reagan would later seat with Ross in the chair.

The fourteen founding members represented almost the entirety of the American merchant industry that still mattered: AT&T Microelectronics, Advanced Micro Devices, Digital Equipment, Harris Semiconductor, Hewlett-Packard, IBM, Intel, LSI Logic, Micron Technology, Motorola, NCR, National Semiconductor, Rockwell International, and Texas Instruments. The list was a snapshot of an industry that no longer included most of the firms that had defined it a decade earlier. Some were absent because the companies had been sold to foreign owners; others because they had stopped making chips altogether. Cypress Semiconductor’s T.J. Rodgers, who would later complain that SEMATECH was a country club for incumbents, declined to join. The defection scarcely registered. The board the founders elected included Sporck as chairman, Galvin as a member, IBM’s vice president for industry operations Sandy Kane in a senior governance role, and a search committee charged with finding the consortium’s first chief executive.

Their first choice was Noyce. He was sixty in early 1988, retired in everything but title, hosting visiting Soviet engineers at his Los Altos home and giving the occasional university speech. He was the obvious figure: cofounder of Fairchild and Intel, joint inventor of the integrated circuit, the ambassador between the entrepreneurial sensibility of Silicon Valley and the institutions of Washington. He was also already chairing the search committee. When the committee, after months of effort, told him it could not find anyone better than him, he agreed to leave Los Altos and move to Austin. The line he gave colleagues, repeated in the National Academy of Engineering memorial that would be written for him two years later, was that the job was just too important to leave to others.

Austin had been chosen the previous month from a national bidding war that had narrowed thirty-four states to twelve and then to one. The selection was announced on January 6, 1988, by Sporck on behalf of the SEMATECH board. Texas had outbid Arizona, North Carolina, Massachusetts, New Mexico, and Oregon by an unembarrassed margin. The University of Texas board of regents had bought a ninety-four-acre former Data General campus in southeast Austin and leased it to SEMATECH for a dollar a year. The state government and the city had piled in tax abatements, infrastructure, and direct cash. George Kozmetsky’s IC2 Institute at UT, which had been promoting Austin as a “technopolis” for a decade, had coordinated the proposal with the political precision of a campaign organization. The choice was, among other things, a recognition that Silicon Valley was not the only American place capable of hosting a major federal program. It also moved Noyce out of the city he had made.

The clean room opened to ribbon-cutting on November 15, 1988. Noyce, who hated formal events, gave a brief speech that was less triumphal than the moment seemed to demand. SEMATECH’s role, he told the assembled engineers and reporters, was not to invent the next chip. It was to fix the supplier base — the equipment makers, the photoresist vendors, the inspection-tool vendors — on which every American fab depended. Without those suppliers in working order, every American chipmaker would have to source its tools from Nikon and Canon and Tokyo Electron, and at that point the question of who actually owned the American chip industry would already have been answered.

The work of the next eighteen months was harder than the founding politics had been. Engineers seconded from the member companies arrived in Austin carrying suspicions of every other member’s engineers. Noyce held morning all-hands meetings in the cafeteria, sat at random tables for lunch, refused a corner office, and wore the same khakis and short-sleeved shirts he had worn at Fairchild thirty years earlier. The egalitarianism, well-documented in Larry Browning’s later organizational study of the consortium with Janice Beyer and Judy Shetler, was deliberate. Noyce understood, better than anyone, that fourteen rivals could not be made to cooperate by org-chart authority alone. Within a year the Austin pilot line was running. Within two, the consortium had begun coordinated programs with Applied Materials, Lam Research, and KLA, the Silicon Valley equipment makers whose health would determine the ceiling of every member’s manufacturing capability.

Noyce did not see most of it. On the morning of June 3, 1990, he swam laps at his Austin pool, came inside, suffered a heart attack, and was taken to Seton Medical Center, where he was pronounced dead. He was sixty-two. He had been due to fly that day to take delivery of a new aircraft. The succession committee, meeting under emergency conditions, named the Xerox executive William Spencer, formerly of Bell Labs and Sandia, as the second chief executive in October 1990. Spencer would run SEMATECH until 1997 and oversee, year after measurable year, the closing of the yield gap between American and Japanese fabs from fifty percent in 1985 to nine percent in 1991, the metric on which the consortium would later be judged a success.

Whether SEMATECH actually saved the American chip industry would remain an argument among economists for decades, and a careful one; the academic literature is divided on whether the consortium increased total American chip R&D or only made existing R&D more efficient. What is harder to dispute is what its founding meant. In December 1987, when Reagan signed the defense bill that authorized it, the United States had publicly accepted a proposition that would have been unthinkable five years earlier: that semiconductors were a national-security industry whose viability the federal government would underwrite, that the market alone could not be relied upon to keep America in a strategic technology, and that an entire architecture of postwar free-market orthodoxy had been suspended, in this one case, because the alternative was to lose. The men in the windowless Pentagon conference room had won their argument. The consortium they had argued for now existed. Whether it could actually produce the manufacturing capability they had promised was a question that depended, more than anyone in Washington was yet prepared to admit, on the confidence of a country sitting on the other side of the Pacific, where the next chapter of the story was already being written in the inflections of a Japanese politician who believed his nation no longer had to take instruction from the United States about anything.