Over the past couple of years, during the chip industry’s worst slump ever,
Taiwan Semiconductor Manufacturing Co had an ace up its sleeve. That card was
Nvidia Corp, a Silicon Valley designer of graphics chips, including those in
Microsoft’s Xbox game player. TSMC manufactures chips for Nvidia and for
dozens of other semiconductor companies. Thanks to Nvidia’s soaring sales,
TSMC was able to stay in the black.
CEO |
However, TSMC recently hit a serious snag, and it’s pointing up the
vulnerabilities of not only Taiwan’s mightiest chipmaker but also its whole
contract semiconductor industry. So-called foundries such as TSMC and its
crosstown rival, United Microelectronics, serve as contract manufacturers for
scores of mostly American chip-design houses, such as Nvidia, Qualcomm, and
Xilinx. The two Taiwanese giants account for two-thirds of the foundry business,
which now produces some 10% of all the world’s chips. TSMC Chairman Morris
Chang and others have predicted that foundries will churn out 40% to 50% of all
chips by around 2010.
That outlook got tarnished when TSMC tried to produce Nvidia’s latest
design. To cram this hot chip with 125 million transistors, a hefty 56% jump
from its previous design, Nvidia used circuit lines a mere 130 nanometers wide.
That’s roughly 1,000 times smaller than a human hair and 30% thinner than
previous-generation chips. But the first version that TSMC produced last year
had big problems. For one, it generated too much heat. Fixing the glitches took
six months, and only now is Nvidia shipping the new chip.
TSMC’s stumble may prove costly. It prodded Nvidia to reevaluate its
association with the world’s No1 foundry. Last month, Nvidia announced it
would tap IBM’s chipmaking prowess for advanced semiconductors. Given Big Blue’s
track record in pioneering new silicon technology, explains Di Ma, Nvidia’s
vice-president for operations, IBM’s foundry "might have more to
offer" when it comes to the most advanced circuit designs.
Losing a portion of Nvidia’s world-beating chips is a major blow for TSMC.
Chang is counting on high-end products to help stave off hungry newcomers–the
rival foundries in China, Korea, and Malaysia that are putting pressure on
prices for low-end chips.
Indeed, making next-generation chips will involve knowhow that few foundries
possess, says Sumit Sadana, director of strategy at IBM Microelectronics. Chips
with 130-nm lines mark the start of a new phase in chipmaking. From here on out,
producers will have to marry innovations in materials with tweaks in basic
transistor design. Just buying the latest equipment for "printing"
circuits on silicon, which Sadana says has been the main focus of foundries, won’t
suffice for the 90-nm lines that will appear next year–and that are already in
pilot production at Intel Corp and Texas Instruments Inc
The foundries’ shortcomings have been clearly evident in 130-nm chips, says
G Dan Hutcheson, president of market watcher VLSI Research Inc. "The yields
were terrible," he notes. Joseph A Osha, semiconductor analyst at Merrill
Lynch & Co, terms TSMC’s 130-nm efforts a "debacle." To TSMC’s
chief rival, all this is a portent of dramatic change–"The days of the
high-margin foundry are gone forever," according to UMC Chairman Robert
Tsao. And since several of the top 25 companies in the US and Europe have bumped
into the same problem as TSMC, they will soon be forced to seek foundry
partners.
That should be music to the ears of Morris Chang–except for one thing.
Xilinx says that UMC and IBM, its two foundries, are "far ahead" of
TSMC in 90-nm technology. So Chang finds himself in the uncomfortable position
of playing catch-up as he tries to convince the chip world that TSMC will
continue to lead the foundry business.
By Bruce Einhorn in Hong Kong, Cliff Edwards in San Mateo, Caliornia,
Andrew
Park in Dallas, and Spencer E Ante in New York in BusinessWeek. Copyright 2003 by The McGraw-Hill Companies, Inc