Programmable Notes

New Fabless Models Changes gnaw away at the fabless model as companies seek more control over wafer capacity by Larry Waller

The "fabless" method of operating in semiconductors surely has proven a big winner for programmable companies. In the past five years, such firms as Xilinx and Altera have built very profitable businesses at the $500 million sales level, without the vast expense and unceasing challenge of running their own chip fabrication lines.

Instead, following the fabless model, wafers are produced to specific company designs by outside suppliers (either foundries dedicated to this business or by other semiconductor firms that devote production to outside customers). And other smaller fabless outfits­up to perhaps 100, some say­are also doing well.

So, with all that fabless momentum going for these companies, why are so many of them now pouring hefty investments into new world-class chip fabrication facilities? There have been many such announcements since mid-1995 and more are expected.

One obvious answer has to be that additional global semiconductor production capacity, in the sub-micron sizes especially, is now so hard to find that only by investing into foundries or partnerships can future production needs be met. That reason is near the top for virtually all participants. Furthermore, most projections on world semiconductor growth see chip demand outrunning the ability to supply it for perhaps two more years. However, other seers warn that the chip supply-demand equation should start to balance out much sooner. But capacity availability still would stay tight.

There is yet another factor at work that confirms some earlier opinions about what could force changes in the fabless model. One of the fabless scene's strongest proponents, Bernie Vonderschmitt, founder and CEO of Xilinx (San Jose, CA), pinpoints what caused him to change his mind over the past year. "It was the logistics of dealing with too many fabs," he says, along with the mushrooming growth that took Xilinx so swiftly to "north of a half billion dollars (in sales)." Administering the flow of which products to what foundry and, even touchier, maintaining second sources for all of them, amounted to "a hell of a problem about where to place them."

The solution for Xilinx is investing $150 million to get upwards of a 25% stake of a $1 billion fab in a joint venture with United Microelectronics Corp. (UMC, Hsin-Chu, Taiwan). Before it goes on-line in 1997, UMC will support Xilinx with other fabs. Also, Xilinx continues with several foundries, notably Japan-based Seiko-Epson, which presently supplies about 65% of its needs. As to any worse capacity crunch in the offing, Vonderschmitt says, "put me down as someone whose top worry is not a shortage."

The biggest fabless firm, PC component supplier Cirrus Logic, Inc. (Fremont, CA), with more than $1 billion IC sales, experienced foundry growing pains similar to Xilinx . It started to look for other manufacturing answers "beginning at $500 million sales and dealing with 14 foundries," observes VP Tom Rigoli. "It gets very unwieldy, also we needed more ready access to world-class fabrication technology," he says.

Cirrus Logic has the biggest fab investment program of all the formerly fabless outfits underway. In October, it announced plans to invest about $2 billion over the next five years in manufacturing. It is expanding an existing manufacturing pact with IBM Corp. in East Fishkill, New York; setting up a new joint venture with AT&T Microelectronics of Allentown, PA; going into partnership with UMC; and broadening its deal with Taiwan Semiconductor Manufacturing Company Ltd. (TSMC, Hsin-Chu, Taiwan).

Observers must listen carefully when these fab deals are described, since the financial arrangements and terminology vary widely between companies. Altera, for example, carefully points out it does not own any fab facilities, but at the same time is making investments that nail down some $1 billion in wafer capacity by decade's end. Altera is committing $123 million through 1997 to assure future wafer purchases from TSMC and earmarking another $125 million for a joint venture with TSMC and others to build a US-based wafer fab.

Eric Cleage, VP of marketing, notes that Altera was the first programmable firm to take a stake in someone else's chip plant. "Remember that in 1990 we bought 17% of Cypress' Texas plant." The company recognized early on that these investments were necessary to "have some control and influence over the direction of technology, and assurance of on-going capacity."

Among the formerly fabless crowd, none have bumped more sharply against capacity limits than Lattice Semiconductor (Hillsboro, OR). This company, which depended solely on long-term partner Seiko-Epson for its wafers and put up $42 million in advance payments two years ago, got an unexpected jolt during the summer. Seiko's message was "don't count on any wafer increases after 1996." For firms growing at the programmable industry's 30% and upwards rate, this is not welcome news.

Along with other fabless companies, Lattice has invested in another world-class fab to be built by UMC with production starting in late 1997. "It's fair to say things are transitioning," says corporate marketing director, Stan Kopec. "That's the new fabless model...to be part owner."

If anyone gets credit for being right on the money in discerning in advance the recent unfolding of fabless investments, Bill McLean, vice-president of market researcher Integrated Circuit Engineering (Scottsdale, AZ), can claim the prize. Several years ago after studying the tea leaves, he predicted that at $500 million revenues, fabless operations would start to split the seams. "Things go haywire at that level," he now observes, with logistical and scheduling snarls as knotty as an airline's. "Companies have to be investing in some kind of partial ownership," he says.

Few officials in the business foresaw a fabless barrier at $500
million sales, notes John East, CEO of Actel Corp. (Sunnyvale, CA) "Five years ago, we never thought a fabless company could grow that big," he observes.

Therefore, the big question now facing participants has to be­ with all the changes taking place among leading companies, is the so-called fabless model still workable?

The answer, as phrased by Altera's Cleage, is one with which nearly everyone agrees. "The fabless concept is still sound," he states. The proof is that even with heavy investments in wafer capacity, his firm (and most others) still will depend on foundries for a large segment of production for the foreseeable future. Altera expects to be buying half its wafers in 2000.

Executive director of the Fabless Semiconductor Association in Dallas, TX, Jodi Shelton is another who sees the concept continuing to thrive. One reason is size; of the 100 or so fabless firms, most are small operations who'll have to buy wafers for years. "We're encouraging foundry expansion since things are tight and some still have trouble getting a couple hundred of wafers a month," she notes. The current move of companies to invest in their own fabs has more to do with the extremes associated with heavy demand rather than a waning of the concept, in her view. "Is it only something that happens in a capacity-strained environment," she wonders.

Market analyst Rhondalee Rohleder of Pace Technologies (Scottsdale, AZ) also is a staunch believer. She thinks the number of fabless outfits will actually grow "because capital investment in Class I fabs becomes unattainable for all but the largest of semiconductor manufacturers."

While the outlook appears rosy for firms taking big positions in joint ventures, one veteran official who has looked closely at them has some doubts. These concern how highly competitive firms will get along in sharing the big new fabs when inevitable production, management, and allocation problems occur. In view of continuing forecasts of demand outstripping supply through the end of the decade, these joint ventures could turn into battle royals over who gets served first and with the most allocations.

Larry Waller is a contributing editor for Integrated System Design.

To voice an opinion on this or any Integrated System Design article, please e-mail your message to: michael@asic.com.

integrated system design  February 1996