Chips on Monster Wafers
Every decade or so, the semiconductor industry goes through a wrenching metamorphosis that alters its very foundations. The industry is in the midst of such a shift today, and this one is the biggest ever. Its hallmark is a king-size silicon wafer--the silvery disk from which tiny chips are diced. Used for memory chips, just one disk's worth could store roughly 5,000 sets of Encyclopedia Britannica. Or it could provide enough horsepower for a supercomputer. A few chipmakers are now processing thousands of these disks every day.
The new wafers measure 300 millimeters (12 inches) across, which may not seem like much of a jump from today's 200-mm (8-in.) wafers. But the 50% increase in diameter yields 2.5 times more surface area for etching chips, yet costs only about 20% more to process. That will help ensure a continuing drop in chip prices for years to come--affecting prices for everything that depends on silicon.
The most profound effect of 300-mm wafers, though, will be to reshuffle the chip industry's worldwide balance of power. And for all but the largest producers, the smartest choice may be to go "fabless," avoiding the manufacturing rat race. The price of a chip factory, or wafer fab, that can handle 300-mm platters can easily top $2.5 billion--and some plants are pushing past $3.5 billion. Only a handful of traditional chipmakers can justify such an investment because keeping one of these "megafabs" humming requires annual sales of at least $6 billion.
And yet, despite the semiconductor slump, many companies seem to be thumbing their noses at economic reality. By yearend, 15 megafabs will be in operation, including two from Intel and one each from IBM and Infineon Technologies (IFX ), plus several from contract manufacturers in Asia. Eight more huge plants are due next year. By 2007, as many as 40 megafabs may be up and running, with a combined capacity of 575,000 wafers per month. That's twice the projected worldwide demand, even after the chip industry recovers from its current downturn, calculates Morris Chang, chairman of Taiwan Semiconductor Manufacturing Co. (TSM ). So the stage is set for a bloody battle. "This isn't going to be a polite kind of high-tech revolution--it's going to be the French kind," predicts G. Dan Hutcheson, president of market watcher VLSI Research Inc.
Anticipating the bloodbath, many of the industry's Old Guard are holding off on 300 mm. For chipmakers such as Agere Systems (AGR ) (formerly Lucent Microelectronics), Fujitsu (FJTSY ), and Toshiba (TOSBF ), the stakes in this poker game are just too rich. In 2000, the best-ever year for chip sales, only 10 companies had sales in the $6 billion range. Motorola Inc. (MOT ) made the $6 billion cut back then--but not last year, when sales fell below $5 billion. Thus, when Fred A. Shlapak, president of Motorola's semiconductor unit, first eyed a financial analysis of 300-mm fab ownership, his reaction was a swift "Uh-uh. No way am I going to be stuck with one of those babies in the next downturn."
So most industry veterans are being forced to hatch a new business model. Of the Old Guard, only Intel (INTC ), IBM (IBM ), Texas Instruments (TXN ), and Samsung are actually operating a 300-mm plant. The others are either collaborating on joint-venture fabs or turning to contract chip manufacturers, called foundries, for up to 50% of their future production. Some chip stalwarts doubt they'll ever build another new factory on their own.
Because Taiwan dominates the foundry business, it's emerging as the big winner. Last year, the island nation's foundries processed about one-fifth of all the wafers carved into chips, according to Semiconductor Equipment & Materials International (SEMI), a Silicon Valley trade association. TSMC accounted for half of that volume, and Chang predicts that Taiwanese foundries will process 40% of the world's wafers by 2010. Meanwhile, hoping to emulate Taiwan's success, China, Malaysia, and Singapore plan to build 15 wafer fabs, five of them 300-mm behemoths destined to serve as foundries. As a result, SEMI expects foundries, located mainly in Southeast Asia, to be churning out half of all chips by decade's end.
While some existing 200-mm fabs will continue in production for a decade, they'll have a virtual cap on revenues. If any market served by a 200-mm producer gets large enough to attract a 300-mm rival, "bang--the game is over," says Trevor Yancey, vice-president for technology at market researcher IC Insights Inc. "A 200-mm producer simply can't compete on price." Since building a new 300-mm fab won't be feasible for midsize producers, Yancey adds, "the only real choice may be to go fabless"--farming out production to a foundry.
In addition, 200-mm fabs will be increasingly confined to yesterday's low-margin processing technology. Blame that on the ongoing market slump. Sales of chipmaking gear have plummeted, so equipment makers couldn't afford to develop tools that will "print" tomorrow's thinner circuit lines on both 200- and 300-mm disks. "Most vendors chose to support 300-mm wafers," says P. Larry Tolson, head of wafer fabrication at Texas Instruments Inc.
But for a while, at least, 200-mm fabs can buy new tools. Coming next year, these will trim lines on the most advanced chips from 130 nanometers wide to a mere 90 nm, or 1/1000 the width of a human hair. Individual chips can then shrink, more could be diced from each wafer, and chip costs would fall still further.
By late decade, however, 200-mm fabs won't be able to keep up with the latest technology--or supply key chips for the most sophisticated products. Circuit lines should shrink to 65 nm, perhaps by 2005, and then to 45 nm before decade's end. With lines that skinny, a pinhead-size chip could be crammed with more computing power than Intel's latest Pentium 4 microprocessor, and thumbtack-size chips could sport a billion transistors. Those chips will be dirt cheap because each 300-mm disk will yield thousands of them. And new applications will blossom. It might be economical, for example, to embed voice-activated cell phones in the tops of ballpoint pens.
With that many transistors to play with on each chip, designers will be able to create electronic gizmos that need only one piece of silicon. The much-ballyhooed system-on-a-chip should arrive at long last. The bad news: Companies that make auxiliary chips--to speed graphics or control peripheral devices, for example--may need a new game plan. That includes most of the 50-odd midsize chipmakers in the U.S., such as Cypress Semiconductor (CY ), Fairchild, National Semiconductor (NSM ), and Siliconix.
Systems on a chip could give a boost to the fabless camp. Last year, four fabless companies posted sales above $1 billion, led by Qualcomm Inc. (QCOM ), with $1.24 billion. And industry consultant Paul P. Castrucci, former head of IBM's 200-mm wafer fab in Vermont, says the advent of foundries actually makes outright fab ownership moot: "In many cases, you can do better, financially, without a fab."
But at least for now, few chipmakers are ready to give up their fabs. To old-line producers, a fabless semiconductor company has always been something of an aberration. "Real men have fabs" is the oft-repeated put-down from the 1980s by W.J. "Jerry" Sanders III, founder of Advanced Micro Devices Inc. (AMD )
Even before Sanders stepped down as CEO early this year, he was rethinking that position. His new outlook is tantamount to "real men have fab partners"--a stance certain to catch on among others who have been priced out of the 300-mm market. The soaring cost of factories has triggered a "tectonic shift" in economics, Sanders explained shortly before handing the CEO reins to Hector de J. Ruiz, former president of Motorola's semiconductor unit.
To compete in microprocessors with archrival Intel, AMD needs to tap the economics of 300-mm wafers. So AMD partnered this year with Taiwan's UMC Group and Germany's Infineon, formerly Siemens Semiconductors. Hutcheson of VLSI Research thinks AMD has picked the right partners. He credits Infineon with "doing more than any other company to make the 300-mm transition happen." Infineon cut the ribbon on the first 300-mm pilot line back in 1998. Now, Infineon and AMD are bartering their design and process expertise for 50% stakes in the twin 300-mm factories UMC is building in Singapore. The first, UMCinfineon, is due to open early next year, and the plant next door--run by a partnership called AU, from the initials of AMD and UMC--will come onstream in 2005. All three are cooperating on process-development work.
Motorola (MOT ), Philips (PHG ), STMicroelectronics (STM ), and Taiwan Semiconductor have a similar arrangement. Their 300-mm factory in Crolles, France, began dicing up the big wafers earlier this year. "Instead of budgeting 20% of revenues for capital spending, we'll spend just 10%," says Motorola's Shlapak, who dubs the strategy "asset-light."
Indeed, collaborations now permeate the industry because every facet of making next-generation chips is becoming too complex for one company to manage. Perfecting design methodologies and matching process recipes for circuit lines smaller than 90 nanometers can run into the hundreds of millions of dollars. That's what a Sony-IBM-Toshiba linkup expects to spend to develop the brain chip for Sony Corp.'s (SNE ) next PlayStation. Faced with such massive outlays, most chipmakers are staking out specialist positions and forming partnerships to share the financial burden.
Chipmakers that don't find some way to step up to the 300-mm plate understand the risks. They point to what happened to Japan's semiconductor industry. In the 1970s and '80s, the Japanese were hammering their U.S. and European rivals. By 1986, Japan emerged as the world's leading producer of chips. But then, the Japanese delayed spending on the metamorphosis that took wafers from 150 mm to 200 mm in diameter. Sensing an opening, the U.S., Europe, and Korea quickly jumped on 200-mm technology. By 1996, the U.S. had reclaimed the lead. Samsung Electronics blossomed into the world's biggest supplier of commodity memories. And Europe's silicon industry recouped its market-share losses. Even today, when it comes to chips, Japan is a pale shadow of its former self. Clearly, says VLSI's Hutcheson, "wafer size matters."
The jump to 300-mm wafers presents a stark choice. Companies that ignore the technology may not survive as major producers much after 2005. And those that invest in it may find profits elusive even longer. For many, going fabless may be the most sensible option of all.
|Corrections and Clarifications In "Chips on monster wafers" (Science & Technology, Nov. 4), the description of the usable area on a 12-inch silicon wafer should have said "up to 2.5 times as much" as an 8-inch wafer.|
By Otis Port in New York, with Irene M. Kunii in Tokyo, Bruce Einhorn in Hong Kong, and Andrew Park in Dallas