Silicon Gold Rush
Taiwan’s chip makers can’t resist the lure of China, the world’s largest consumer of semiconductors
In just 15 years, the farmland of the Pudong district of Shanghai has disappeared under gleaming skyscrapers, monuments to, among other things, the fastest-growing semiconductor industry in the world. Within this charmed circle, the fastest-growing start-up is Semiconductor Manufacturing International Corp. (SMIC). In the five years since its founding, it has become the leading chip maker in China, the next big chip-making power in the world.
Ironically, China’s ascendance in semiconductors is almost entirely attributable to assistance from Taiwanese technologists. SMIC itself was founded by a Chinese-American raised in Taiwan, and many of its employees are from Taiwan. Hundreds of them, just weeks off the airplane, are bedding down in the company’s massive employee community, complete with gym, supermarket, beauty parlors, and a school for 1200 children.
SMIC is not the only foundry with roots in Taiwan that has set up shop here. A report by a U.S. chip industry group, the Semiconductor Industry Association, in San Jose, Calif., estimated that in the second half of 2001 alone, more than 3000 engineers left Taiwan to work in China’s semiconductor industry. Far more have evidently arrived since then. In Pudong, where three of SMIC’s foundries started operation two years ago, the Minnan dialect of Taiwan can be heard in local bars and restaurants.
You’d think the outflow of expertise from Taiwan to the mainland would make the Taiwanese very, very afraid. After all, they invented the foundry model—the manufacture under contract of chips by one company for another—and their government has spent hugely on incentives to secure so much of the global market.
The idea was for the foundries to grow, even at a loss, in order to attain economies of scale that would deter others from entering the business. And so far, the plan has worked. Taiwan controls nearly two-thirds of the worldwide sales of the US $20 billion foundry business, a business that is nearly a tenth of global chip sales. What’s more, the fraction of chips made in foundries is projected to grow steadily for years to come.
Yet whatever their government may think, Taiwan’s engineers aren’t afraid about ceding that lead to the mainland; they’re here in China, in the thousands.
“I seldom if ever talk to a non-Taiwanese person when I visit SMIC,” says Michael G. Pecht, a mechanical engineering professor at the University of Maryland, in College Park, who has written extensively about Asian electronics industries. He adds, “I’m not talking to senior management, and I’m not talking to the factory workers—I’m talking to the people who know how to make things work and understand the technology.”
In fact, besides being home to Chinese foundries that are at least partially staffed with Taiwanese workers, China also now boasts shimmering new fab facilities built and operated by Taiwan-based companies. Foremost among these is Taiwan Semiconductor Manufacturing Co. (TSMC), in Hsinchu, the contract chip-making giant founded in 1987 [see photos, "A Piece of the Action”].
It’s tempting to look at the relationship between TSMC, the largest foundry chip maker in the world, and the smaller SMIC and see in it an ironic reversal of their respective countries’ big-brother-little-brother status. Viewed from a global perspective, TSMC’s worldwide sales of $7.6 billion last year dwarfed the $1 billion in sales of SMIC. But within China itself, SMIC is the bigger player, selling half the chips made there. So with the foundries, little brother is growing up fast, and big brother is acutely aware of it [see photo, "In the Chips”].
It is just one of the many twists in this complicated, high-stakes, fast-moving game. Even the basic parameters can be hard to fathom: China provides incentives; the engineers come from Taiwan to help staff local start-ups; Taiwanese companies feel compelled to set up subsidiaries in China to compete with the start-ups other Taiwanese are leading; then Taiwan’s government passes laws to slow things down, partly because of acute diplomatic sensitivities but mainly out of concern that mainland China is going to eat its lunch.
Given that sobering long-term prognosis, why do Taiwanese companies keep flocking to China? First, China offers sweeteners. Until recently, China tantalized outsiders by taxing China-produced chips at a significantly lower rate—up to 14 percent less—than foreign-made chips. But that helping hand was withdrawn after China joined the Geneva-based World Trade Organization three years ago. Now China relies on more informal understandings, which are harder for outsiders to criticize. The most important of them is the unofficial requirement that at least 30 percent of the total value of electronic goods made in China be supplied by local companies.
Second—and more important—China is where the growth is. This year its electronics industry will buy chips worth $59 billion, more than any other country in the world, according to the research firm iSuppli Corp., in El Segundo, Calif. Less than 5 percent of those chips are now made locally, leaving the other 95 percent up for grabs [see graph, ].
No company understands those economics better than TSMC. Its founders invented contract chip making in the late 1980s, when they noticed that the cost of a fab—now generally a 10-digit figure—was spiraling in relation to other industry costs. They realized chip makers that could no longer afford to own their manufacturing capacity would have little choice but to rent it.
In an interesting parallel to the role of Taiwanese engineers in China in this decade, it was technologists who returned from the United States two decades earlier—TSMC founder Morris Chang among them—who gave Taiwanese foundries their start.
These foundries bred new chip makers, the so-called fabless companies. Later they began getting business even from the biggest players, as such companies as Intel Corp., in Santa Clara, Calif., and Motorola Inc., in Schaumburg, Ill., came to realize that sometimes it made more sense to rent a bit of spare capacity than to build a fab that wouldn’t be fully exploited.
Throughout the 1990s, as tech-related exuberance built to a crescendo, TSMC thrived. It now has 11 foundries—8 in Taiwan, 1 in Singapore, 1 in Washington state, and 1 in China. The Chinese plant is run by a subsidiary, TSMC Shanghai, which won approval from Taiwan’s government in February 2003. Y.C. Chao, the president of TSMC Shanghai, in the city’s Songjiang suburb, says he is now ready to take orders.
In the competition between the two companies, convolutions abound. Start with SMIC’s charismatic chairman, Richard Chang. China-born but Taiwan-bred, he graduated with a degree in mechanical engineering from National Taiwan University, in Taipei, in 1970 before earning a master’s degree from the University at Buffalo, the State University of New York, in 1973 and a Ph.D. in electrical engineering from Southern Methodist University, in Dallas, in 1986. After working for 20 years in Dallas at Texas Instruments Inc., he returned to Taiwan to found the Worldwide Semiconductor Manufacturing Co., quickly selling it to none other than TSMC. Then he moved back to China, in 2000, to found SMIC.
By 2003, SMIC had three plants running in Shanghai. And last year, the company bought a plant in Tianjin from Motorola. Earlier this year, the company began operating one more in Beijing, and two more Beijing plants are being readied for production. At his spacious steel-and-glass headquarters in Pudong, 15 kilometers east of the old, historic part of Shanghai, Chang can gaze out at an employee village that makes up for the deficiencies in local infrastructure by providing apartments, supermarkets, sports fields, and that school for 1200 students, who can study in English, Mandarin, or both.
Lately, the investment seems to be paying off. Slowly but surely, SMIC is creeping closer and closer to TSMC’s cutting-edge capabilities. SMIC’s best plants, in Beijing, can now make chips with 90-nanometer wires on 300-millimeter wafers—which is the state of the art. Those are the parameters typical of fab facilities used to make gigabit memories and chips for cellphones. Chang points out that SMIC makes chips only for commercial applications, not for military ones.
In other aspects, though, SMIC still lags. Its Pudong plants produce chips with wires only as narrow as 130 nm on 200-mm wafers, good enough for controllers for DVD players, sensor chips for low-end digital cameras, or decoders for high-definition television sets. One of the three Pudong plants supports the other two by connecting circuit elements with copper, the metal of choice; it also offers the increasingly obsolete standby, aluminum.
SMIC’s engineers have not yet mastered the technique of putting a thin veneer of silicon over a layer of silicon dioxide, a technique that lets circuits run faster than those on ordinary silicon wafers. Nor does it yet embed germanium atoms in the upper layers of silicon wafers to strain their crystalline structure, tremendously speeding up the passage of signals. Still, matters have improved significantly over what they were a decade ago, when China’s semiconductor makers were laying down wires 10 times wider than those used in the most advanced processes.
Meanwhile, TSMC isn’t resting on its laurels. Chao plans to focus his production on local Chinese markets, building chips for Chinese customers and also for international customers who want to sell to Chinese electronics and appliance companies. He says that at the beginning of this year, the plant was cranking out 5000 wafers per month—a third of its planned capacity and only a small fraction of the 120 000 wafers that SMIC’s foundries can produce.
But production numbers don’t tell the whole story about the competition between SMIC and TSMC. In that battle, TSMC is prohibited from deploying its most potent weapon—superior technology—to full effect. As a Taiwanese entity, Chao’s company is not allowed to produce circuits with wires smaller than 0.25-micrometers. But Chao expects that the Taiwanese government will soon approve more advanced technologies, capable of making wires with widths down to 0.15 um. Once that happens, he says, the company will install the equipment and ramp up production rather quickly.
Chao tries to put the best face on the restrictions: the resulting chips are good enough for cars, toasters, power controllers, and analog circuits, though not for most portable applications—including cellphones and laptops—which require, above all, the low power and compactness that the most advanced technologies offer.
“We spend according to the market needs and according to the level of approvals from Taiwan,” he says. There’s plenty of demand in China for chips built with older technologies, he adds, and as demand for more advanced technologies grows, TSMC will be able to keep up with it.
Other experts disagree with Chao’s outlook, noting that local customers are already paying for better chips than he can provide. SMIC data supports this view. In the third quarter of 2004, the company produced more than two-thirds of its chips at 0.18 um or less. While TSMC and other firms tagged with Taiwan’s flag may lobby their government for permission to use more advanced technology, the express political point of the export restrictions is to limit trade—and the economic point is to limit transfer of technology.
Trace It back to Gen. Chiang Kai-shek, the defeated nationalist leader who, with other Chinese refugees, moved to Taiwan and set up an alternate Chinese state in 1949. His policy was “no contact, no negotiation, no compromise” with China, a line long reciprocated by the mainland communist government.
But this mutual animosity between the two governments is not altogether shared by Taiwanese and Chinese engineers, who maintain an allegiance to Chinese culture that transcends political systems and financial considerations.
After Chiang’s death in 1975 and the coincident opening of China to the world, however, Taiwanese investors began to skirt the rules, originally by dealing with China through shell companies in Hong Kong.
The shell game continues with a vengeance in the form of transplanted operations masquerading as independent start-ups. In the foundry business, they are organizations funded and run by experts from a particular Taiwanese company who disavow any connection to it. The most widely known of those is He Jian Technology (Suzhou) Co., about a 2-hour drive from Shanghai.
He Jian was founded in 2002 by former executives of United Microelectronics Corp. (UMC), in Taipei, the world’s second-ranking contract chip maker. A 2003 report by the Semiconductor Industry Association says bluntly that many in the industry believe “He Jian’s Suzhou operation constitutes what will at some point become UMC when the pretenses are dropped.”
An industry executive based in Taiwan who wishes to remain anonymous confirmed that. “All of He Jian’s key senior managers come from UMC; the fabs use equipment from UMC; if the managers want to quit He Jian and return to UMC, they can,” he says. This past winter, in fact, the Taiwanese government began investigating the UMC-He Jian link. UMC denied that it violated any laws, claiming that it provided only consulting services to He Jian and did not transfer any technology. In a recent statement, however, UMC’s chairman, Robert Tsao, acknowledged that he hoped to acquire He Jian if the government relaxes rules on China-bound investment. Presumably, he would acquire it on good terms.
SMIC, too, has been charged with some sleight of hand. Chang asserts he built the company’s technological base by collaborating with Fujitsu, Infineon, Toshiba, and other foreign clients. But TSMC alleged in a lawsuit last year that SMIC had purloined its industrial secrets by hiring away dozens of its employees.
The suit was settled early this year, when SMIC agreed to pay TSMC $175 million over six years. As part of the agreement, the two companies will share their patents. It’s not a lot of money for either company, notes Bill McClean, president of IC Insights Inc., a market-research firm in Scottsdale, Ariz. But SMIC’s leaders will benefit the most from the agreement. They’ll now be able to focus on business and reassure customers who might have hesitated to buy chips incorporating intellectual property of dubious provenance.
Much of the antagonism will abate, of course, if it turns out that there is plenty of profitable business for everybody. However, SMIC, TSMC, and other prospective entrants to the market differ on how fast demand for their wares will build. And some wonder whether China’s foundry rush makes sense to begin with. While SMIC’s Chang builds plants at breakneck speed, TSMC’s Chao worries that China is building too much production capacity. He believes that for the next several years most of the demand for logic chips will be filled by far more advanced manufacturing plants overseas. “Intel alone,” he points out, “sold $5.8 billion worth of chips to China last year.” That’s almost three times what the indigenous companies sell.
Dynamic random-access memories are also likely to come mainly from abroad for a while. “I don’t think the Chinese will be able to compete against [the Korean company] Samsung or [Boise, Idaho-based] Micron Technology in dynamic memory,” says the University of Maryland’s Pecht.
Foundries, by their nature, yield thin margins, don’t create a lot of jobs, and do just as well no matter where they stand. Ed Chang, deputy director of TSMC’s China operations, is the first to concede that it is no cheaper to make chips in China than in Taiwan. He notes that although land, labor, and construction tend to cost less in China, electricity and other infrastructure cost more. But over the long term, as the Chinese economy grows and the chip industry matures, labor costs will rise and electricity prices will fall. “We expect that the cost of making ICs here will be a few percent lower than in Taiwan when the supporting infrastructure matures,” says Chang.
Meanwhile, the start-up foundries are reducing their risk of failing by tying their fortunes to local officials. Not that they have much choice: the officials are necessary every step of the way. They let you know it, too. According to the unnamed Taiwan semiconductor executive, municipal governments may ask firms to make a donation to some project, but “the donations aren’t voluntary, they’re required.”
Still, competition among regions—not unlike that seen in the United States whenever a corporation thinks out loud of relocating—makes officials bend over backward to help their chip makers. Last summer, for example, when Shanghai was experiencing rolling blackouts, local officials told SMIC not to worry, that its electricity would not be cut off. And it wasn’t.
For TSMC Shanghai, help from local officials has come in the form of land. Its plant is built on a much more generous parcel of land than Chao recalls from his days in Taiwan’s industrial parks. Chao, an avid golfer, likes to say that even using a No. 1 driver, “you couldn’t hit a golf ball from the street to our building.” When he suggested to local officials that the wide expanse could be used to build a second plant, they told him that he really needed more land and that they had already set aside another section to the west for expansion.
Besides cultivating the locals, managers must, of course, keep their customers happy. And in China, famous for its intellectual piracy, that means working doubly hard to protect customers’ secrets, principally circuit-design data. (They also work hard to protect their own manufacturing secrets, as TSMC’s lawsuit against SMIC attests.)
To keep industrial spies from sneaking secrets out of the plant, the TSMC foundry in China bans its employees from bringing their personal laptops into the office; back in Taiwan, the same rule holds. At SMIC, employees must sign an agreement not to reveal its customers’ or its own intellectual property. Most employees in the foundry are barred from using the Internet, and even senior managers aren’t allowed to bring their own notebook PCs into the company. Only two people have access to the data for making the masks used to etch a customer’s wafers.
Even with the incentives and the skyrocketing electronics industry, China’s semiconductor industry will grow only slowly, says IC Insights’ McClean. That’s because many of the ICs sold come not directly from foundries but from fabless semiconductor companies, which design ICs and sell them into the electronics market, relying on the foundries to build them.
Today, China’s fabless semiconductor industry is immature. Although the number of these companies has grown from fewer than 100 in 2000 to about 500 in 2004, most observers believe that few of them are likely to survive, because they lack experience in IC design. As more Chinese designers return from overseas and go to work for local fabless companies or start their own, that situation may change.
But the University of Maryland’s Pecht says China’s foundry strategy will finally make sense when its fabless companies and its electronics and appliance manufacturers use China’s own domestic fabs to bootstrap their way to world-class operations. “What these companies are going to see is that if they design those chips internally and have them built by Chinese semiconductor manufacturers, they can make their products much more cheaply,” he says.