How China Built a Semiconductor Industry

Show video

In his later years, Qian Xuesen, the  founder of China's rocket industry,   who famously left America after  persecution, had reportedly said: > In the 60s, we devoted all our efforts to the  “Two Bombs, One Satellite” project, and we gained   a lot from it. In the 70s, we didn't focus on  semiconductors, and we lost a lot because of that. After the chaos of the Cultural Revolution,  the People's Republic of China made three major   pushes towards a domestic semiconductor industry. Two of them failed. The third  succeeded. In this video,   we look at how China built  a semiconductor industry.

## Beginnings The founder of China's solid state physics   academic community is a Chinese  returnee named Huang Kun (黄昆). Born in Beijing in 1919, he traveled abroad   to England and Bristol University  to study physics - earning a PhD. In the late 1940s, he collaborated with the Nobel  Prize winner Max Born to finish a textbook Born   had started called "Dynamical Theory of Lattice  Physics". It was eventually published in 1954. After the outbreak of the Korean War, Huang wanted  to return to China. Huang wasn't the only one. Another major academic was Xie Xide (谢希德), then   studying at MIT. Her speciality was  in surface and semiconductor physics.

Despite the efforts of America's Truman  administration, Huang and Xie made their way   home and founded China’s first solid-state  physics department at Peking University.   They also wrote a major textbook on the  subject titled "Semiconductor Physics" Other universities like Fudan University  in Shanghai followed suit. But today,   Peking University is generally seen as the  cradle of China's semiconductor industry.

## The 774 Factory China's first production semiconductor  industry comes from the Soviet Union. In 1953, China laid out its first Five Year Plan,   covering the years until 1957. The Soviets  helped make and carry out this Plan. One of the 156 industrial projects  completed with Soviet assistance   was the 774 Factory - also known as  the Beijing Electron Tube Factory. In the 1950s, the 774 Factory successfully  pulled its first Germanium crystal,   establishing itself as China's  most technologically advanced   semiconductor factory. A few years later,  a silicon crystal was pulled the same way. A small Chinese electronics  cluster in the Jiuxianqiao   (酒仙桥) sub-district of Beijing grew up around it.

Then in the early 1960s, the Factory  started producing germanium alloy   junction transistors. These transistors  are an improvement on Shockley's original   bipolar junction transistors. Over in the  United States, before the introduction   of the planar transistor, alloy junction  transistors were quite big in the market. You take a piece of pure germanium and fuse beads   of indium onto both sides of it. This  was done manually. Workers in the 774   factory used a homemade soldering iron to  melt the indium balls onto the germanium.

Many of these devices went into  China's first transistor radios,   which began hitting the world  market in late 1963 via Hong Kong. Much later, the 774 factory's assets were sold  off to a private equity group to found Beijing   Orient Electronics Technology Group.  Today, a world-leading OLED company. ## Integrated Circuits The integrated circuit trend that began in  the 1950s allowed a second semiconductor   cluster in China's south to  join Beijing's in the north.

In 1965, a Shanghai component factory cooperated  with the Shanghai Institute of Metallurgy of the   Chinese Academy of Sciences to produce China’s  first integrated circuits - a TTL-type circuit. A key early use case for  transistors and integrated   circuits were the military and  their space programs. Famously,   the Minuteman II missile program bought a peak  of 15,000 integrated circuits a week in 1965.  

And the Apollo Space Program is estimated  to have bought 200,000 ICs before 1964. Same as it was in China. Semiconductor  production in aeronautics was supported   by another Chinese returnee named  Huang Chang (黄敞). A Harvard PhD,   he worked at Sylvania Semiconductors, and  obtained permanent residency in the United States. In 1959 he decided to return to China - inspired   by Qian Xuesen's contributions to  the Chinese missile program. Many of   the computers for those missiles were  built with self-produced transistors.

Based on this progress as well as  others not mentioned in this video,   one might judge China's semiconductor industry in  the 1960s to be ahead of many others in the world. ## The Third Front China's semiconductor progress ran  into obstacles in 1964. That year,   Mao Zedong decided to relocate  many of China's heavy industry   and military installations into the country's  interior - the "Third Front" policy (三线建设). This massive relocation was done in light of  worsening relations with the Soviet Union.  

And thus it was carried out with great  haste - involving nearly 500 factories,   1.5 million workers, and 200  billion RMB of investment. In 1970, the military relocated its integrated  circuit research in Hebei - Institute 24 - to   the remote area of Yongchuan County  (永川县) within the mega-city of Chongqing. The Third Front's results are mixed. On  one hand, it stimulated the electronics   industries of many underdeveloped areas.  Some areas reported company growth of 700%. Yet on the other hand, the  Third Front slowed development   by deliberately stretching the supply  chain. Many of the relocated areas were   in mountainous regions and lacked proper  transportation networks - causing delays.

It also separated R&D and manufacturing,  making it difficult for the research produced   in the labs to ever make it to the fabs.  Moreover, Problems on the factory floors   had to be handled by unqualified factory staff,  perpetuating poor yields and low production. ## Cultural Revolution And then we have Mao Zedong's Cultural  Revolution, which lasted from 1965 to 1975. The Revolution was an immensely  damaging period - not only for China   itself but also its semiconductor  development. Many colleges closed  

down. Those that remained open  still suffered immense disorder. People with western education or connections -  including the aforementioned Huang Kun - were   persecuted, removed from positions of power,   or sent to the communes to work. The students  either worked in factories or studied politics. The universities did not reopen until December  1978, when Deng Xiaoping emerged as China's   paramount leader, refocusing the country's  direction. This left a 16-year gap between   graduating classes in semiconductor  technology education - 1966 to 1982. There were a few pockets of progress.  The Third Front relocation allowed the  

aforementioned Institute 24  to work in relative peace. But overall, China was isolated and steadily  fell behind the rest of the world. People of   the era express frustration with this long  gap in Chinese semiconductor progress.

## Catchup Thus, the late 1970s and early 1980s  were a period of recovery and catch-up. Early Chinese efforts on semiconductor  industry development were simple and   rather decentralized. On the talent  front, expand the base by producing   more semiconductor graduates and send  Chinese overseas to learn in the West.

Secondly, governments on various levels tried to   acquire new equipment from abroad  to upgrade their old state-owned   semiconductor factories or build new ones -  spending about 1.3 billion RMB to acquire it. Local governments, state-owned enterprises,  and universities funded new semiconductor   fabs. 33 new fabs were constructed from  1984 to 1990. It was a bit of a crazy time. It is important to note here that while  export controls like CoCOM did exist,   the Chinese were little affected by them.  Most stuff they got was trailing edge stuff   that was not restricted. And the Chinese  - like the Soviets - were quite good at   evading restrictions, usually by  funneling them through Hong Kong. Funds were limited - at the time, Chinese  governments were importing equipment to   build its now-formidable electronics and computer  industries too. So this semiconductor equipment  

was old and often came second-hand. Factory  workers often had little clue how it all worked. Foreign visitors to the factories  noted the lack of good production   habits. Factory work was seen as an "iron  bowl" job. There were more people in the   fab than what was needed, and  most did not work efficiently.

A later report by the Chinese Semiconductor  Industry Association investigated the outcome   of 24 acquired lines and found that just ten of  them functioned. The growing Chinese electronics   industry - frustrated with weak domestic  IC production - quickly turned to imports. Really, only one of China's factories in   the 1980s actually performed up to  expectations: Wuxi Factory No. 742. ## Wuxi Factory 742 The factory now called Wuxi Factory  No. 742 began as the Jiangnan Radio   Factory in 1960 (江南无锡电器材厂), producing  Soviet-designed diodes in the city of Wuxi.

They later became a state-owned company  - Wuxi referring to its local city - and   then in 1968 merged with an industrial  school. They produced discrete devices   for use in military equipment like radios and  did it very well, winning several gold medals. In 1980, the Chinese government chose Factory  742 as a core factory for a new project referred   to as Project 65 - because it was part of  the Sixth Five-Year Plan from 1981 to 1985. In Project 65, the government would provide  about 270 million RMB of funds to acquire   two production lines from Toshiba -  a 3-inch wafer bipolar at a 5-micron   process node and a 4-inch wafer memory node.  Both then quite far behind the leading edge.

As part of the transfer process, Toshiba  hosted a few Chinese engineers at their   plant to learn recipes. The transfer went  smoothly and the line entered high-volume   production in 1984, producing  30 million wafers that year. After some early learning troubles, the  factory raised its yields to over 80% by 1985,   producing 70+ million integrated circuits  for televisions. By the late 1980s,   half of China's televisions were claimed  to have been made with chips from 742.

Soon thereafter, the factory realized  it needed to upscale its native R&D   capabilities. So in 1986-1987, the 742 merged  with Institute 24's Wuxi city branch to   create Wuxi Huajing Microelectronics  Company (無錫華晶電子), or just Huajing. The factory was reputed as Mainland China's  leading semiconductor company. They also   claimed several R&D milestones. Despite China's  isolation, the company received foreign experts,   had books and papers from IEEE,  and sent workers abroad to study.

In 1987, it became China's first company  to put an LSI-class integrated circuit   into production and also the first  to prototype a VLSI-class 64K RAM   chip. Huajing's successes as a smaller  factory convinced the Chinese government   to make it the central focus of their  next big semiconductor policy push. ## Project 908 As early as 1983, Chinese leaders had been aware that this first attempt to catch up  in semiconductors had not panned out.

They identified several reasons for this.  Investment in semiconductor facilities was   done haphazardly, leading to a lack of scale  and production. Yet it took years - until   1989 - for the Chinese State Council to  approve a new measure to address this. In 1989, Chinese IC consumption was 400  million chips. Domestic production across all  

its scattered fabs was just 114 million. Not to  mention those chips were horrendously backwards. Policymakers decided to emulate the  successful Koreans and make Chinese   chaebol of their own. They cleared the decks,   establishing five major semiconductor companies  in China's northern and southern regions. Two were in Shanghai, one was in Shaoxing City  of Zhejiang Province, one more in Beijing,   and the last was Huajing. These would be  China's semiconductor national champions.

Thus in 1990, we have Project 908. 908's  goal was to demonstrate to the world that   China can build a leading integrated  design manufacturer in semiconductors. It had several parts but its biggest and most  significant package would grant Huajing 2 billion   RMB to upgrade its existing 5-inch wafer line to  a 6-inch wafer line - then considered a leading   edge technology. It would run a 1 micron or 0.8  micron node, producing 12,000 wafers a month. That is a lot of money. In 1989, the  mainland's per capita GDP was 1,538 RMB  

or $400. So 2 billion RMB would be equivalent  to the annual production of 5 million Chinese. But despite all those resources, Project 908  was a resounding failure. The new production   lines were obsolete before they even  activated. In 1997, the fab had capacity   for 12,000 wafers a month but was only making  just 800 each month. Huajing fell into crisis. ## 908's Failure As a component for a finished good,  semiconductors had always been a   state-funded thing in China - heavily reliant  on government funding. Huajing sprung from a  

government factory and Project 908  was a government-initiated project. So naturally 908 and others of its like -  there was a 907 project too which suffered   similar issues - were government-led initiatives.  This made it difficult for Huajing to succeed. Any   investment decision - like for a litho machine -  required government-sign off, which wasted time.

And local governments fought bureaucratic  battles over how to do things and what   was to be done. Memorably, 26 agencies  locked horns over where to do a single   dredging operation. You might imagine how such  dynamics would affect a major strategic decision   like whether to pursue DRAM or ASIC logic chips. When Samsung was grappling over a  similar decision in the mid-1980s,   there was a circle of executives  debating over which to pursue. But in the end, a single person  made the final call to go into   memory - Samsung Chairman Lee Byung-chul.

Not the case with Huajing and its  various government backers. In the end,   they decided to produce both DRAM  and ASICs in equal quantities,   leading to a situation where neither  was being produced very well. Huajing's management were hands-on people,  and at first they struggled to handle this   titanic project. So they devolved some  authority to smaller operating units,   but this in turn encouraged empire-building  and vastly exploded the company's scope.

By the late 1990s, Huajing  consisted of 26 factories, firms,   and branches managed by 9 departments.  It was not only scattered across the   whole semiconductor supply chain from  materials to design to wafers to assembly,   but also had its own hospital, day care,  school system, and community centers. There was also an issue of talents.  Taiwan and Korea benefitted from   waves of returning talents - particularly  electric engineering talents - who trained   in the West. At the time, China did not have this. Beyond these internal issues, there was a major  external one. Due to its participation in the   Korean War in 1950, the People's Republic of  China was subject to export restrictions on   leading edge United States technology. Same goes  with Taiwan, another plausible transfer partner.

Nevertheless in 1994, the Chinese government  negotiated a technology transfer deal between   Huajing and Lucent Technologies. Lucent - the  former Bell Labs - would train engineers and   transfer a node process plus IP blocks for  an advanced telecom switching technology. On the surface, it seems like a huge boon. It  wasn't. The government negotiated the deal,   meaning that Huajing had no  say on what to transfer over.   Huajing's managers knew the company was  nowhere near ready and wanted a cheap,   low-tech technology to start with before  working their way up. But the government   officials chose for them, and steered towards  a flashy technology Huajing could not digest.

Furthermore, what they got they could not use.  Huajing's engineers, mostly self-taught, were   good at reverse-engineering existing products.  When they got the Lucent IP and technologies,   they had no idea how to integrate it  into what they were already doing. Thus,   the technology transfer process did not take.

And then there was a customer issue. When  Taiwan's ITRI spun off its first chip company UMC,   they helped source customers - tapping a  growing network of Taiwanese electronics   companies. They provided product IP. They  wrangled Taiwanese business tycoons to invest. By contrast, the 908 project never really  considered the problem of competing for   and getting customer orders. Why  would they? In the old times,   the planned economy guaranteed demand and there  were no imports. But times were different now. All this means that it was not until  1995 that the first production lines   began construction. Production did not start  until 1997-1998. By then, the 6-inch wafer   technology they received from Lucent was out  of date and left unused. A massive debacle.

Huajing had took out a billion RMB of  government loans to pay for its high   tech fab. In 1997, the company began  to struggle from high interest rates,   turning a staggering loss of  240 million RMB that year. With the company in crisis,  Huajing struck a deal with   Taiwan. Several Taiwanese with experience in  the semiconductor industry agreed to run it. Due to cross-strait rules, they founded a  shell company in Hong Kong called Central   Semiconductor Manufacturing Corporation or  CSMC - sounds familiar - and took over Huajing. They stabilized the company and then  sold it to a Chinese SOE in 2002. Today,  

it is Wuxi China Resources Huajing  Microelectronics. They make commodity   discrete power electronics and trade  mostly on the power of their storied name. ## Project 909 By 1995, it was clear that Project 908 was  a failure - another black eye on China's   ongoing semiconductor journey. Everything  went wrong and a new approach was needed. Policymakers, listening to advisors,  correctly determined that 908 failed   because of its government-led status.  The government's central role and  

Huajing's former life as a SOE crippled its  ability to move fast in a dynamic market. In November 1995 - even as the 908 dumpster fire  burned over in Wuxi City - the Chinese government   took its third and most ambitious yet crack  at the semiconductor apple - Project 909. At   the time of its initiation in March 1996, it was  China's largest semiconductor state project yet.

909 wanted to raise a national champion from  birth - Shanghai Huahong Group (华虹半导体有限公司). It   would ameliorate all of the problems faced  by China's prior forays into semiconductors,   bring in and digest foreign technologies,  and dominate the domestic market. 909 was funded with over 10 billion RMB directly  from the State Treasury and Shanghai's municipal   government - at least, we believe so. To  be honest, can never be too sure where the   money comes from with these projects.  Nevertheless, they had a blank check.

To cut red tape, the company had China's serving  Minister of Electronics Industry Hu Qili (胡启立) as   its chairman of the board. The mayor of Shanghai  was on its lead working group, and Shanghai's   various governmental departments were told  to give 909 the greenest of green lights. Huahong's core achievement would be the  construction of a 8-inch memory wafer   fab running a 500 nanometer process,  aided by another technology transfer   deal with Japan's NEC. This joint  venture would be called Huahong-NEC. Again, export restrictions were  not an issue. In part because of   the technology’s lagging sophistication  and then-warming relations between the   United States and the People's Republic of China.

Lastly, the government reserved a market for  Huahong-NEC. The Ministry of Electronics Industry   researched the market and selected a rapidly  growing product in China - IC cards. In 1997,   the State Council reserved  that market for Huahong-NEC. ## 909's Fate Did Huahong succeed? Huahong-NEC constructed its fab within  two years. Per the requirements of its  

technology transfer deal, NEC would  provide an early sales channel,   so the venture started selling 64-megabit  DRAM sold under NEC's brand name. In 2000, its first full year of operations,  the company turned a profit of 350 million   RMB. The company improved its yields  from 50% at the start to nearly 90%. But there was a catch. Japanese workers from  NEC were the ones actually running the fab,   not Huahong personnel. And  they did so in such a way   to give Huahong managers no insight into  what was going on inside their own fab! By contrast, when Taiwan's ITRI struck their  legendary technology transfer deal back in 1974,   they laid out a complete transfer process - soup  to nuts. The deal was not considered complete  

until Taiwan finished a replica fab of their  own, on Taiwanese soil, staffed with Taiwanese. So the transfer was not complete. Without that,  Huahong could not upgrade the fab. When the   company experienced a large operating  loss during a memory crash in 2002,   Huahong canceled NEC's participation  and took full operational control. They then struck a deal with another  company called Jazz Semiconductor,   a startup aerospace foundry spun  off from Rockwell Semiconductor.

Jazz's founding CEO was a  Chinese-American returnee named Shu Li. This weird triple-venture then acquired foreign  IP to develop China's first contactless smart IC   cards for transportation and payment: The Shanghai  Public Transport "One Card". They also switched to   a foundry model, which probably more contributed  to a rebound in revenue in the coming years. Jazz was later acquired by the Israeli foundry  Tower and sold their interest in Huahong in   2011. And after a bunch of mergers, Huahong  is today China's second largest semiconductor  

foundry - first place loser, in my book - with  much of their production at the lagging edge. ## Circular 18 Huahong's profits might have been up and  down. But China was starting to learn. In 2001, China entered the  World Trade Organization,   liberalizing their semiconductor markets. Now  that policymakers had their national champion,   they can build a broad industry behind  it by attracting foreign investment. In June 2000, the State Council passes  a new policy called Circular 18,   or “Several policies for  encouraging the development   of software industry and integrated  circuit industry" (鼓励软件产业和集成电路产业发展的若干政策) Circular 18 offered significant tax incentives  for every semiconductor company in China. And  

most importantly, it did not matter whether  they were domestic or owned by foreigners. In 2000, Richard Chang sells his foundry company   Worldwide Semiconductor to TSMC.  Soon thereafter, Chang goes to China,   raises over a billion dollars, and starts  SMIC using basically the same business model. SMIC has gone on to be China's  leading semiconductor foundry,   and a force to reckoned with. China's first true  

blue semiconductor national champion  to go alongside Hua Hong and Nexchip. At the same time, there is a massive flurry  of semiconductor FDI activity from foreign   companies entering the Chinese market.  Applied Materials, AMD, Philips, Motorola,   ASML and more announce billion dollar investments  into China. The industry is up and running. ## Conclusion For additional reading, I really recommend  this dissertation by Susan K Mays,   "Rapid Advance: High Technology in  China in the Global Electronic Age". Mays, who seems to be at UT Austin now,   did some incredible work here  and I just want to call that out.

Anyway. There is an impression that China did not  care much about semiconductors in the past. And   now that they are facing export bans, the  West has suddenly motivated them to care. Perhaps there is some truth to that.  But the People's Republic of China has   been grinding on semiconductors since  the 1960s. They have thrown billions   into it and adapted their approach and  strategy multiple times over the decades.  

It seems to me like they have  always cared, with a capital C. To note, it is hard to ignore the fact  that so much of China's accelerating   progress today comes from its existing ties  to the global semiconductor world at large.   The absorption of new knowledge  - and equipment - from abroad.

As those ties and connections start to change  and maybe even sever, we enter uncertain parts   of the playbook. I wonder what happens  next to China's semiconductor industry.

2024-07-04

Show video