"All People Must Make Semiconductors"

In the spring of 1969, sixty women in the Xicheng district of Beijing reported for the morning shift at a workshop that had, until very recently, made household chimneys. The chimneys had been replaced, by ministerial order, with diffusion furnaces and a few hand-built clean benches. The workshop now had a new name, the Xicheng District Semiconductor Device Factory, and a new mission, which was to manufacture transistors. Most of the women on the line were housewives. Few had finished middle school. Before each day’s production, they were bused across the city to Tsinghua University, where physics lecturers, themselves recently rehabilitated from the cowsheds in which Red Guards had imprisoned them, gave the workers a forty-minute lesson in solid-state theory. The women then returned to Xicheng to bond gold leads onto germanium dies under microscopes that had been confiscated from a hospital. Engineers who walked past the workshop on their way to better-equipped institutes elsewhere in the city muttered that the project was, in the phrase that became attached to it, like watching ducks try to fly.

The Xicheng workshop was not an aberration. It was one of perhaps several thousand such operations that proliferated across China during the most intense years of the Cultural Revolution, when the slogan “the entire people must run semiconductors” turned chip manufacturing into a mass campaign of the kind that Mao Zedong had previously directed at steel and grain. Provincial commune cadres opened transistor lines in disused grain mills. Middle schools opened diode lines in their machine shops. The army opened lines in barracks. By the early 1970s, by one official count, more than six hundred semiconductor manufacturing facilities existed somewhere in China. Almost none of them produced anything that could compete, on yield or speed or reliability, with the products coming out of a single American or Japanese fab that operated three shifts in a single building in Sunnyvale or Kawasaki.

That was the verdict, delivered in 1977 in language so unsentimental that it has been quoted ever since, by Wang Shouwu, the man Beijing had appointed as its lead expert on semiconductors. Wang had been one of a small group of foreign-trained physicists who returned to the People’s Republic in the early 1950s to seed a domestic electronics industry. He had spent the Cultural Revolution working under reduced circumstances. By the time he was asked to assess what mass mobilization had produced, he was direct. The country, he said, had more than six hundred plants making semiconductors, and the total annual output of integrated circuits from all of them combined was equivalent to one-tenth of the monthly output of one large Japanese factory. The figure was not flattering. It was not meant to be.

The aspiration had not started as fantasy. In 1956, when Mao’s government drafted the Twelve-Year Plan for the Development of Science and Technology, it placed semiconductors on the short list of priority technologies, alongside computing, jet propulsion, and the peaceful uses of atomic energy. The plan was put together with Soviet advisors in the room and reflected the Soviet model of science: priority sectors, dedicated institutes, concentrated planning, talent recruited from abroad. That summer, two physicists who had taken doctorates in Britain and the United States, Huang Kun and Xie Xide, were brought to Peking University to set up China’s first specialty in semiconductor physics. Their first cohort numbered a few dozen students. Within two years they had trained more than three hundred. Tsinghua, Fudan, and Jilin opened parallel programs. The Chinese Academy of Sciences founded an Institute of Semiconductors. The Beijing Electron Tube Factory, designated as Factory 774, was built in the same year on the model of a Soviet vacuum-tube combine, with brick workshops, dormitories, and a recreation hall arranged around a central courtyard. Wuxi Factory 742, in Jiangsu, came online in 1960. Hebei Semiconductor Research Institute, founded in Beijing and relocated to Shijiazhuang in 1963, became the country’s principal military supplier. In 1965 a team at the Institute of Semiconductors etched a small array of seven transistors onto a square centimeter of silicon and announced China’s first integrated circuit, six years after Kilby and Noyce. By the standards of the developing world that was a respectable showing.

What followed was self-inflicted. In 1966 Mao launched the Cultural Revolution. The campaign that he intended to purge the Communist Party of capitalist roaders also, in passing, decapitated Chinese science. Universities stopped admitting students. The national entrance examination was suspended. Scientific journals ceased publication. The Chinese Academy of Sciences would later document that 229 of its scientists were killed or driven to suicide during the decade. Senior figures at the Institute of Semiconductors were beaten or sent to manual labor in the countryside. Huang Kun, by then the most distinguished solid-state physicist in the country, spent years cleaning latrines. The professors who had taught the first generation of semiconductor engineers were declared bourgeois reactionary academic authorities and removed from their posts. Foreign technical literature, the stuff that researchers in Tokyo and Mountain View read at breakfast, was a class enemy. Anyone caught reading it without political cover was suspect.

The slogan that justified the Xicheng workshop had its own internal logic. If the experts were unreliable, then the masses must replace them. If formal training produced bourgeois scientists, then untrained workers must be enlisted to make the products that the bourgeois scientists had been entrusted with. The peasants and housewives who showed up to bond transistors in renovated chimney shops were, in the language of the campaign, embodying the dictatorship of the proletariat in technological form. The output of these shops was, by every meaningful measure of the silicon trade, terrible. Yields were low. Defect rates were high. The diodes worked when they worked. Transistors carried no specifications because no one in the workshop was qualified to write them. A few mass-mobilized lines produced parts that were quietly used in toys and cheap radios. A larger number produced parts that ended up scrapped before leaving the building.

The damage that compounded over those ten years was less the loss of any particular set of devices than the loss of the cohort that should have been designing the next set. Universities trained no new undergraduates from 1966 to 1969 and no new graduate students until 1977. The pipeline ran dry for an entire technological generation. The young Chinese physicists who would otherwise have been sent abroad to study in those years were instead working in fields. Some never returned to research. The fortunate emigrated. Morris Chang’s parents, ethnically Chinese, had moved to Hong Kong before the founding of the People’s Republic; Chang himself was educated at MIT and Stanford and would, by 1987, found Taiwan’s first foundry. He was emblematic of the diaspora that the People’s Republic had handed away. Tens of thousands of educated Chinese with engineering instincts left for Taiwan, Hong Kong, Singapore, and the United States, and brought with them, when they founded chip companies and joined American fabs, the technical talent that the mainland had spent the Cultural Revolution suppressing.

When Mao died in 1976 and Deng Xiaoping consolidated his return to power two years later, the cadre that took stock of Chinese microelectronics knew what it was looking at. Wang Shouwu’s six-hundred-to-ten ratio was the headline number, but the texture was worse. Most of those plants used equipment that had been improvised from machine-shop parts. The ones with Soviet-era equipment had not been resupplied since the 1960 split with Moscow, after which Soviet technicians were withdrawn and spare parts stopped arriving. The country had no domestic production of photolithography masks at any modern resolution. It had no commercial market for chips, because it had almost no consumer electronics industry to consume them, and because the planned economy did not generate the kind of price signals that would have driven a fab toward the next process node. It had, in the strict sense, factories. It did not have an industry.

Deng’s answer was to buy what could not be invented in time. After Beijing normalized relations with Washington in December 1978 and signed a science and technology cooperation agreement a month later, China’s new electronics planners began importing semiconductor production lines from Japan, the United States, and Western Europe. The Sixth Five-Year Plan, covering 1981 to 1985, set up a Computer and Large Scale IC Lead Group inside the State Council to coordinate the buying. By 1985 the country had imported twenty-four secondhand semiconductor lines at a total cost of roughly 1.3 billion renminbi. Most of them, by the planners’ own subsequent admission, did not work. The lines had been pulled out of foreign fabs at the moment when those fabs were upgrading to a new node, and arrived in Chinese workshops with idiosyncratic process recipes that had been tuned to other facilities and were poorly documented. Chinese engineers, lacking the institutional knowledge that had grown around the original installations, struggled to bring yields up to anything resembling the original specifications. Of the twenty-four imported lines, contemporaneous Chinese reports concluded, only Wuxi 742 met its production targets. The others, in varying degrees, became expensive monuments to the difficulty of buying a competence.

In 1986 the Ministry of Electronics rolled out a new framework, called the 531 Strategy, which proposed to popularize five-micron processes, develop three-micron processes, and prepare the path to one-micron. The numbers themselves told the story. While the 531 Strategy was being announced in Beijing, the leading American and Japanese fabs were already producing reliably at one micron and were running engineering lots at half that. China was, by its own ambition, planning to reach in 1995 the geometries that the United States had been delivering in 1984. The plan, in other words, conceded the gap as a permanent feature of the architecture.

The next attempt was bigger. Project 908, announced in 1990 as part of the Eighth Five-Year Plan, was the country’s first concentrated bet on a single fab that could compete on something close to current geometries. The State Council allocated 2 billion renminbi to upgrade Wuxi Huajing, the operating company that controlled Factory 742, into a foundry capable of producing 12,000 six-inch wafers per month at process nodes between 0.8 and 1.2 microns. A foreign technology partner was to be selected; the contract eventually went, in 1994, to AT&T’s recently spun-off Lucent Technologies, which agreed to transfer process recipes, train Chinese engineers, and supply an intellectual-property design library. On paper the plan made sense. In practice it fell into the maw of the Chinese state.

What killed Project 908 was, by all accounts of those who lived through it, the approval system. The 2-billion-renminbi appropriation had to be unwound through more than a dozen ministries, each of which had veto authority over some component, and each of which insisted on its own review at each phase of the project. The choice of foreign partner alone consumed, by some industry tallies, three years. The choice of equipment vendors consumed another two. By the time the upgraded fab in Wuxi was finally ready to take its first wafers in 1997, the world had moved. The 0.8-micron process that had been written into the planning documents in 1990 was no longer a competitive node. Eight-inch wafer fabs were now standard in Taiwan and Japan; six-inch was a generation back. The Wuxi line, which had been intended as a flagship, opened as a product museum. The fab was still profitable in narrow lines of business, but as a vehicle for closing the gap with the leading edge it was, on the day it opened, already obsolete.

The lesson, if anyone in the State Council was willing to learn it, was that the gap moved while you were trying to close it. The chip industry’s defining feature, the one that the Soviets had failed to grasp at Zelenograd a generation earlier, was that the leading edge advanced at a cadence that no five-year plan could match. A fab planned in 1990 to be competitive in 1995 would, even on schedule, miss the target by one node. A fab planned in 1990 and delivered late in 1997 would miss by three. The reason American and Japanese chipmakers stayed at the front of the technology curve was not that they were brilliantly led, although some of them were. It was that they iterated on a clock that the planning state could not keep. They built test fabs that broke processes that informed pilot lines that drove decisions about the next test fab, and the loop closed in months, not years. Project 908’s loop closed in seven years, at which point the loop did not matter.

The bureaucracy that had strangled 908 was understood by its participants while it was happening, but understanding was not the same as fixing. In the autumn of 1995 the man at the top of the Chinese state, Jiang Zemin, took a trip to South Korea and toured a Samsung semiconductor fab in Suwon. Jiang was an electrical engineer by training, a Shanghai Jiao Tong graduate who had spent most of his career in machine-building ministries and had served, in 1983, as Minister of Electronics. He had ridden the post-Tiananmen reshuffle into the General Secretaryship in 1989 and was, by 1995, the rare Chinese paramount leader who could read a process flow and recognize what he was seeing. What he saw at Suwon was a country with no semiconductor industry to speak of in 1980 producing, fifteen years later, dynamic random-access memory chips at a scale and quality that defined the world market. Jiang flew home and, by every account that has surfaced since, told his Politburo colleagues that China had to develop a serious memory industry, in his phrase, at all costs.

Project 909 was the result. The Ninth Five-Year Plan, covering 1996 to 2000, dedicated more than 10 billion renminbi to a new joint venture in Shanghai’s Pudong district that would, this time, build an eight-inch fab targeting 0.5-micron logic and DRAM, with NEC of Japan as the technology partner. Shanghai Huahong Microelectronics, the operating company, was incorporated on April 9, 1996. The joint venture with NEC, called Hua Hong NEC, followed in 1997. The combined registered capital reached roughly $700 million in cash, against total construction costs that would push past $1.2 billion before the fab took its first wafer.

To run the project, Premier Li Peng made an arrangement that breached the usual rule against officials holding enterprise positions. He appointed Hu Qili, then minister of the Ministry of Machine-Building and Electronics Industry, to also serve as president of Shanghai Huahong. Hu’s biography was a particular kind of warning. He had been a member of the Politburo Standing Committee in the late 1980s, marked as a possible future general secretary, and had aligned himself in May 1989 with Zhao Ziyang against the imposition of martial law on the students gathered in Tiananmen Square. He had been one of two Standing Committee members to vote no. In the Fourth Plenum that followed the massacre, he was stripped of his party positions and removed to political exile. Two years later, when the leadership decided that he was still a reformer the country could use, he was rehabilitated into the electronics ministry as a vice minister, then as minister. The leader of China’s most strategically important industrial program of the 1990s was a man whose career had survived, at the cost of his political ambitions, by his willingness to be controlled.

Hu had a more sophisticated grasp of how chip industries worked than most of his colleagues, and he tried to insist on conditions, on the equipment chosen and the joint-venture structure, that gave Hua Hong NEC a real shot at competing on DRAM. The fab was built. Pilot DRAM production began in 1999. The first full wafers came off the line in early 2001. Then the world stopped buying DRAM. The 2001 collapse of the dot-com bubble, compounded by overcapacity in Korea and Taiwan, drove DRAM prices off a cliff and held them there. Hua Hong NEC posted losses of more than $200 million in its first two years of full operation. By 2003 the venture quietly abandoned its DRAM ambitions and reinvented itself as a generic foundry, taking design-for-manufacture business from anyone who could pay. The state company that had been built to make China a memory power had become, within five years of its first wafer, a contract manufacturer of analog and power chips at trailing-edge nodes. As an industrial outcome it was not catastrophic. As the centerpiece of a national strategy it was a confession.

What the planners began to understand, slowly, after 909 is what the Xicheng housewives’ supervisors had failed to understand a quarter century earlier and what the planners of 908 had refused to learn from. Chips did not yield to mass mobilization, because the device on the workbench was not a thing the masses could improve by trying harder. Chips did not yield to imported lines, because the imported lines stopped being current the day they shipped. Chips did not yield to ministry-led joint ventures, because the ministries operated on five-year clocks while the industry operated on eighteen-month clocks. Each Chinese attempt had assumed that the missing ingredient was money or scale or political will. Each attempt had produced a factory. None had produced what the Americans and Japanese and now the Taiwanese had built, which was a system in which engineers, customers, and capital argued in real time over what to make next.

Outside the planning offices, by the late 1990s, a different conversation was beginning to happen. Chinese-American executives in California’s Santa Clara Valley, men and women who had emigrated as graduate students in the 1970s and risen through American chip companies, were beginning to wonder whether the right way to build a Chinese chip industry was not through ministries and joint ventures at all but through a Taiwan-style foundry, founded by veterans of the foreign industry, financed in part by foreign capital, and run on commercial terms. One of those executives was a soft-spoken, devout man named Richard Chang, who had spent twenty years at Texas Instruments learning how a fab was actually run. He was beginning to think, in the late 1990s, about going home.