Science & Technology
`REAL MEN HAVE FABS'
Odd as it sounds, chipmaker Actel Corp. doesn't make chips. It sells them, all right, and its designs for so-called programmable logic devices are hot. But the nine-year-old company farms out production to true chipmakers with spare capacity in their wafer-fabrication plants. This strategy was a winner for six years. Actel's sales took off in 1988--and rose nearly 50% just in 1993's first nine months. Then the downside of going "fabless" showed up. Manufacturing went awry at one of its foundries, and Actel couldn't find a pinch-hitter. Some $3 million in expected fourth-quarter sales went down the drain, virtually wiping out profits.
Actel is hardly alone. "The fabs are full," laments CEO John C. East. And that could signal the end of an era. It means that 250 fabless chip companies--the startup stars of Silicon Valley for a decade--face a growth crisis. As their chip-production options shrink, it will be hard for them to avoid a retreat. At the least, the character of tomorrow's semiconductor industry is going to change.
The fabless wonders have thrived by exploiting a capacity glut, mainly in Japan. There, memory-chip makers expanded greatly in the early 1980s--just before demand waned. Many of those fabs were at 60% of capacity as recently as 1992, according to consultant VLSI Research Inc. By using them, U.S. entrepreneurs avoided the main hurdle for a chip startup: the tens or hundreds of millions in wafer-fab costs. A new venture could thus devote its resources to innovative designs, such as faster clones of Intel Corp.'s microprocessors or multimedia chips that vastly improve graphics on personal computers. By pioneering these cutting-edge products, fabless companies grew faster and earned higher returns than established chipmakers--giving them clout way beyond their 5% share of the $77 billion world chip market.
Now, everything is changing. The boom in advanced PCs crammed with memory chips, coupled with stingy capital spending in Japan, is boosting utilization of Japan's older fabs to 80% or more. Worse, surplus capacity in the most modern fabs--those that print circuit-line widths of 0.8 micron or less--has shriveled to 3%, says VLSI Research. That's bad news because the designs of the fabless folks usually require advanced production processes. Says G. Dan Hutcheson, VLSI's president: "All the conditions that made fabless companies possible are coming to a close."
"NO LONGER FREE." Hardly anyone is being spared. The list of chip companies that have had production delays since late 1993 include Catalyst Semiconductor, Cyrix, and Exar. Foundry problems at Seeq Technology Inc., which once had a fab but closed it in 1990 to pinch pennies, contributed to a decision to sell its business in EEPROMs, or electrically erasable, programmable read-only memories.
Years ago, as the fabless trend picked up steam, Advanced Micro Devices Inc. Chairman W.J. "Jerry" Sanders III was skeptical. "Real men have fabs," he growled. Now, many fabless companies concede his point. "Silicon [production] is no longer 'free,'" says Barry L. Cox, president of fabless Weitek Corp. So Cox and his counterparts are hastily devising survival strategies. One is simply to avoid designing chips that demand the latest manufacturing processes, thus avoiding fabs where capacity is tightest. That can't work for long, though. For instance, Catalyst Semiconductor Inc. had to upgrade to a more cost-efficient process for its flash-memory chips to stay in the price war between Intel and AMD.
Other stopgap measures abound. Xilinx Inc., for example, is boosting inventories to make sure it doesn't run short if foundries fall behind. Last September, Cirrus Logic Inc. promised to buy a set number of wafers over three years in return for a guarantee of foundry capacity. Some fabless folks are even buying into factories. In February, Actel and Brooktree Corp. each invested $10 million for a small stake in an $850 million fab that Chartered Semiconductor Manufacturing, a Singapore foundry, will start building this month. Because the Singapore government is providing low-cost funds, says Brooktree Chairman James A. Bixby, "we get more leverage" from the investment.
USED BLUES. Cirrus Logic--the largest fabless chipmaker, with $535 million in projected sales for the fiscal year ended Mar. 31--may soon become a full-fledged partner in a foundry. With its dozen wafer suppliers running at capacity, Cirrus plans to join IBM in revamping an underused Big Blue fab that once specialized in chips for mainframe computers. The East Fishkill (N.Y.) plant will provide about a third of Cirrus' needs starting next year. Even an existing chip plant isn't cheap, however. Cirrus says its share may cost a substantial portion of a recent $137 million stock offering.
More such deals are in the works. Chipmakers with fabs, such as LSI Logic Corp. and Cypress Semiconductor Corp., report being besieged with proposals for "strategic relationships." These don't always work. Cyrix' deal to trade design rights to its Intel-compatible microprocessors in exchange for foundry time at Texas Instruments Inc. blew up in a dispute over deliveries. Still, Steve Della Rocchetta, Chartered's marketing vice-president, says he can't return all the calls from companies eager to invest in its new Singapore foundry.
The fabless companies could yet get a respite. Actel's East optimistically predicts that new, government-subsidized fabs in Korea and Taiwan, as well as in Singapore, could add fresh capacity as early as next year. The relief may be short-lived, however. Craig R. Barrett, Intel's chief operating officer, says that because world-class fabs now cost $1 billion or more, the industry is investing far too little to meet even conservative projections of sales growth through 2000. And offsetting any short-term gains, Hewlett-Packard Co. and AT&T Microelectronics recently decided to stop providing foundry services.
INVEST OR GIVE UP. Progress in semiconductor technology only makes matters worse. Fast microprocessors such as Intel's Pentium family require companion chips that can keep up. So circuits have to shrink to speed signals along. But designs are now packed so tightly that electrical errors, or shorts, are a chronic danger. Avoiding them often means tailoring manufacturing to each chip design, which tends to rule out generic fabs.
The upshot could be grim: Put up--invest in a fab--or give up. Frederick L. Zieber, president of industry watcher Pathfinder Research Inc., thinks more fabless companies will vanish, especially those with single product lines. Cirrus Logic has itself acquired six small fry since 1990. Even if fabless outfits survive, they may have to mortgage their futures. Like many true chipmakers, LSI Logic now insists on technology rights in return for factory capacity.
So, barring a breakthrough that overturns chipmaking economics, large companies seem destined to dominate more than ever. Fabless companies can only hope that by maintaining a blistering pace of innovation, they'll win favor--and fab time--from their bigger cousins. Because even the high-stakes crowd needs a constant supply of clever designs to stay in the chips.TABLE: THE FABLESS WONDERS
A sampling of companies that design chips and outsource production
Company Revenues (Millions)
Main product Fiscal 1990 1993
Programmable logic $21 $60
Programmable logic 78 140
Mixed-signal chips 70 96
CHIPS & TECHNOLOGIES
Chip sets 293 98
Multimedia, communications 107 355
Microprocessors NM 125
Analog and other chips 88 146
Programmable logic 65 103
Video and graphics NM 113
Networking, communications 45 33
Mixed-signal chips 61 83
Graphics accelerators NM 80
Controllers 58 36
Programmable logic 50 178
NM=Not meaningful DATA: HAMBRECHT & QUIST INC.
Robert D. Hof in San Jose, Calif.