Betting Big on Chips

Why TSMC boss Morris Chang is spending billions despite the tech slump

Semiconductor tycoon Morris Chang's name pops up in the oddest places on Taiwan. A billboard ad shows a stunning movie star holding an Ericsson personal digital assistant. "Morris Chang, you should change your PDA," it urges. Chang's name also turns up in media as diverse as a real estate promotion. It's a testament to just how widely revered the chairman and founder of Taiwan Semiconductor Manufacturing Co. (TSM ), and one of Asia's great tech visionaries, has become in his homeland.

Chang hasn't given his permission for any of the many ways his name is exploited. But he finds it amusing that, at age 70, he has cult status among Taiwan's young and trendy. Indeed, the pipe-smoking tycoon is generally unflappable.

These days, the sage of Taiwan's Hsinchu tech district needs every bit of his inner calm. In the thick of one of the chip industry's deepest swoons ever, Chang is betting billions to raise the ante on his competitors. TSMC is already the world's biggest foundry, or contract manufacturer, of semiconductors. And even as the tech economy crashes around him, Chang is building silicon wafer plants aimed at keeping his company at the forefront of technological change. Chang also is rolling out his vision for what he is convinced is the future of chipmaking: the "virtual foundry." The goal is to save time and money by making it easy for engineers from chip companies and TSMC's plants to collaborate on almost every step of the design and production processes--and do it all over the Internet. "The future is still very great," Chang says. "That hasn't changed at all."

Bullish talk, given the current environment. In the past, TSMC's flexibility and the broad diversity of products it makes for some 600 different chip customers have enabled it to glide through chip busts with minor damage. But these are not normal times. This year's tech crash is hammering sales of everything from personal computers and networking equipment to wireless phones. And that directly translates into plummeting demand for all sorts of chips, whether they be the graphic accelerators that power computer animation or the digital signal processors that enable a cell phone to handle voice and Internet data. Sales in the $225 billion global chip industry are expected to fall by 9% this year, according to market research firm IC Insights Inc. of Scottsdale, Ariz.

BAD TIMING. TSMC's finances are suffering accordingly, especially since Chang plowed $3.8 billion into expansion just before the chip crash. "If we had spent $1 billion less, we would be in better shape," he admits. Although TSMC reported a solid $1.9 billion profit on sales of $5.3 billion last year, ABN-Amro expects its first-quarter sales to drop 28% from the previous quarter, while profits will dive by 59%. A year ago, the company boasted a five-month backlog of orders. Today, that's down to eight weeks. Its fabs, which were running flat out a year ago, are now operating at less than 70% capacity. Industry sources say TSMC now has to give price discounts. Its stock is down 50% in the past year, to less than $3.

At the same time, competition is heating up. A few years ago, TSMC had only two major rivals: United Microelectronics Corp. of Taiwan and Singapore's Chartered Semiconductor Manufacturing. Now, new foundries are popping up in Germany, Israel, Malaysia, and China, often with U.S. and Japanese chip companies owning direct stakes. Plus, the new rivals are developing cutting-edge wafer production processes, flexibility, and quality control. "TSMC is still the 900-pound gorilla, but the others are gaining weight quickly," says Accenture semiconductor consultant Allen J. Delattre. In this game, offering innovative services is vital, he adds, because "good performance is becoming a commodity."

When it comes to divining the future of semiconductors, however, Chang is an old master. He has been involved with the industry from the ground up. After leaving Shanghai in 1949 following the Communist takeover, he earned advanced engineering degrees from Massachusetts Institute of Technology and Stanford University. Chang became manager of integrated circuits at Texas Instruments Inc. in 1966 and, during its heyday, helped build the company into the world's biggest chipmaker. After a brief stint as president of General Instrument Corp., he was lured to Taiwan in 1985 to head the giant government-funded Industrial Technology Research Institute. Soon after, he set up Taiwan's first semiconductor plant--but quickly concluded its chances of succeeding against the giants of Japan and the U.S. were slim.

So Chang pioneered a novel concept: a plant entirely dedicated to processing silicon wafers for other chip companies. By freeing them of the expense of building their own plants, TSMC helped spawn the explosive growth in "fabless" chip companies, such as Nvidia, Altera, and Xilinx, leading to rapid innovation in telecom and multimedia devices. Today, TSMC accounts for 40% of the $17 billion foundry business and serves nearly half of the world's 1,300 fabless chip houses. It operates or partly owns 11 wafer fabs in Taiwan, Singapore, and the U.S.

REBOUND MASTERS. The real test of skill in the chip business, though, comes in playing the downturns. In such a capital-intensive industry, the winners typically are the first to be in production with the latest products when demand takes off again, enabling them to make windfall profits that can be plowed into developing plants for the next generation. During the last chip slump in 1998, for example, TSMC kept building and picked up fabs from weaker Taiwanese foundries. So it was well-positioned when chip demand exploded in early 2000.

Chang is betting that the industry will start pulling out of its down cycle by yearend. Although he has trimmed his planned capital investments this year by almost half, to $2.2 billion, he is still pressing ahead with a number of fabs to carve chips from wafers measuring 12 inches in diameter, the future state of the art. TSMC has started making these wafers on a trial basis at a new fab in the southern city of Tainan. Across the street, it's building another fab for 12-inch wafers. Yet another, near its headquarters in Hsinchu, should be complete by yearend.

THINNER YET. It's a major gamble: Such fabs cost up to $3 billion each, about twice as much as fabs that process the 8-inch wafers that are now standard. The advantage is that 12-inch wafers can yield 2.6 times more chips, which should enable TSMC to lower its costs by 40%. Chang believes demand for bigger wafers will soar in the next few years as chip designers cram more and more functions onto a single sliver of silicon. "A lot of applications in the future will need larger dies," says TSMC President F.C. Tseng. Many in the industry see a trend toward integration of microprocessors, memory, graphics, and other functions into so-called systems on a chip. Therefore, the first foundries that produce such chips will be able to charge a hefty premium.

Meanwhile, TSMC now produces chips with some of the thinnest line widths in the industry. A "line" is a path for electricity. The narrower the lines etched onto a chip, the more a designer can pack onto it--and the shorter the distance that electricity must travel between transistors, vastly improving the chip's performance. TSMC already mass produces chips with line widths of .15 micron for such customers as Nvidia and Taiwan's Via Technologies Inc., the world's biggest maker of chip sets. TSMC says it is just slightly behind Intel in its ability to make chips with widths of .13 micron--and expects to crack the .10-micron barrier next year. "TSMC is still the leader" among foundries, says Via CEO Wen-chi Chen, who plans to start buying 0.13-micron wafers from TSMC for microprocessors this summer.

With no plans to retire--Chang kept up his frenetic pace of conference calls while in the U.S. in February to get married for the second time--he is determined to expand TSMC's lead in foundries. He's hiking his research and development budget by 70%, to $236 million this year, for new production processes. TSMC is also near the forefront in the drive to make chips from new materials such as wafers that use germanium, rather than aluminum or copper, as a conductor.

Chang's spending plans are all predicated on the belief that foundries are the future of semiconductors. Giant chipmakers like Intel, National Semiconductor, and Motorola are starting to outsource. Currently, foundries account for $12.5 billion in sales, roughly 16% of global semiconductor production. Dataquest Inc. predicts that foundry sales will nearly triple by 2005 and will account for 40% to 50% of chip production by 2010.

Not everyone is convinced Chang's spending binge will pay off anytime soon. Even assuming a recovery this year, prospects are "not going to be anything wonderful," says Janardan Menon, Dresdner Kleinwort Wasserstein's Asia technology research director. And new foundries elsewhere in Asia "all are nipping at TSMC's heels." Still, few in the industry have its financial staying power. Thanks to fat profits in the past two years and an $800 million stock issue last year, TSMC is sitting on $1 billion in cash. And its $900 million in debt is only one-quarter of its equity, leaving room for more borrowing.

The key to Chang's strategy for keeping TSMC in front is in services. Because of the technical help it provides chip designers, TSMC already is able to charge 20% more than less sophisticated foundries. Three years ago, the company unveiled Chang's vision of a "virtual foundry." Today, with a number of new online services, TSMC's business model is taking shape. Customers in California or elsewhere can work closely with engineers at the Taiwan foundry from the very earliest design phase until the moment the wafers enter production. Using online tracking services, they can keep tabs on a client's chip at any stage in the process. Two years ago at TSMC, when customers often flew to Taiwan to coordinate engineering and production, it typically took clients 12 to 18 months to get from design concept to production. Now, it can take as little as four months. Given the brief life cycles of today's chip innovations, the time saved can make the difference between big profits and failure. Although UMC and Chartered also offer online services, industry experts say TSMC is well ahead.

One customer that already has found the virtual foundry invaluable is Nvidia Corp., one of today's hottest fabless chip companies, whose products are used for 3D computer animation. Nvidia is making the graphics chips for the Xbox, the game machine that Microsoft Corp. hopes will soon challenge the Sony PlayStation. Sitting in his office in Santa Clara, Calif., Nvidia CEO Huang Jen-Hsun can log onto a TSMC Web site and track the production status of its chips in Taiwan. "We get daily feeds on where every single wafer is in the process," raves Huang. He can make late engineering changes and even cancel an order at the last minute without incurring a heavy penalty. "A lot of people would like to have our business," says Huang. But the "chemistry" he says his engineers have with those of TSMC is a major reason Nvidia gives the foundry some $500 million in orders every year.

TSMC is determined not to lose that edge. Instead of transmitting design data to customers in the U.S. a few times a day, President Tseng says the aim is to offer real-time, interactive engineering. "We would like customers to feel that they are running the fab," he says. To do so, TSMC is looking into setting up data centers in the U.S. and improving its Web portals.

The competition, however, is not standing still. New foundries are popping up in increasingly unlikely places. Take 1st Silicon, which has just opened in Kuching, Sarawak, in the Malaysian portion of Borneo. Sarawak's government helped finance the $1.2 billion project and guaranteed sufficient power and water. The facility is making 20,000 8-inch wafers a month using .25-micron technology. Sharp Electronics Corp., which has committed to buy $300 million worth of wafers, is the anchor tenant. While 1st Silicon is at the low end of the foundry business, manager Claudio Loddo insists it's sophisticated enough to become a player. "If somebody brought you here blindfolded, you would never guess where you are," Loddo says.

AMERICAN PUSH. Chang airily dismisses the upstarts, asserting they lack the services and resources to become serious threats. Still, as Web-based engineering and logistics become widespread, eventually even these fabs may be able to compete on services. Moreover, Chang realizes that customers may want to be less dependent on Taiwan, which is vulnerable to earthquakes and political turmoil.

One place where he is looking to expand is the U.S. Last year, TSMC took full control of its subsidiary there, WaferTech, paying $450 million to buy out the stakes of Silicon Valley chip houses Altera, Analogy Devices, and Integrated Silicon Solution. At WaferTech's fab in Camas, Wash., TSMC admits it has endured some rocky labor relations. TSMC also is looking to the U.S. as a base for more of its R&D. In 1997, the company had only 30 U.S. staffers, most of them in sales and marketing. Today, TSMC has 150 engineers in the U.S., and that's likely to continue growing as it looks for more innovative ideas.

Chang envisions a time in the next decade or so when chipmaking finally hits a wall. As TSMC and other manufacturers move to ever-narrower line widths, at some point the process will have to end. Without new breakthrough materials, "I think the semiconductor industry will become mature," Chang warns. Still, he adds, "I am virtually certain in the next 10 years something will emerge." No one knows what the chip company of 2011 will look like. But Chang is doing all he can to make sure TSMC is that company.

By Bruce Einhorn in Hsinchu, with Frederik Balfour in Kuching, Cliff Edwards in Silicon Valley, and Pete Engardio in New York

    Before it's here, it's on the Bloomberg Terminal.