SAN MATEO, Calif. - Are pure-play foundries losing their edge?
The migration to nanometer process geometries presents a number of technical and economic challenges that some believe traditional chip makers are in a better position than foundries to overcome.
The intricacies of 90nm processing require a high level of process-to-product optimization that critics say is difficult under a pure-play foundry model in which hundreds of distinct chip designs are manufactured for scores of customers.
"Companies like IBM and Intel are arguably better equipped to solve the problems than Taiwan Semiconductor Manufacturing Co. Ltd.," said Len Jelinek, an analyst at iSuppli Corp., El Segundo, Calif. "IBM has a significant history of being an innovative company, and Intel has been very successful at bringing out leading-edge technology and spending a lot of money on research."
After IBM and Intel, Jelinek said joint ventures between integrated device manufacturers would be the next most likely to work out the 90nm problem, followed by advanced foundries like TSMC.
"When you get to 90nm and below, processes are going to get more complex," said a spokesman for IBM Microelectronics, East Fishkill, N.Y., which provides foundry services for a limited number of high-end designs. "If others are just into the cheapest possible way to crank out chips for lots of customers, they will be more and more challenged as time goes on."
However, fabless semiconductor companies characterized the issue as more complex than an old-fashioned rivalry.
"At 0.13-micron, the foundry model will absolutely prevail," said Gina Gloski, vice president and general manager of worldwide manufacturing operations at ASIC supplier eSilicon Inc., Santa Clara, Calif. "At 90nm, the problem isn't foundry vs. IDM. It's the base technology."
According to Gloski, the power being pulled through 90nm-based chips to achieve high circuit speeds will dramatically reduce thei r lifetime. Designers are addressing the problem by slowing down the circuitry, but this defeats the purpose of going to the next technology node, she said.
"Getting working silicon isn't the problem, it's getting working silicon that's actually giving you something better," she said.
Companies will eventually work through the problems, but it's no longer a given that foundries will lead the way.
After TSMC and fellow foundry United Microelectronics Corp. stumbled on the launch of their respective 0.13-micron processes, a recent decision by TSMC to postpone 90nm "risk production" was seen as a signal by analysts that the pure-play foundry model is under growing pressure.
Recent moves by prominent fabless companies also hint at trouble below the surface. While none made a direct connection to TSMC's or UMC's process problems, Nvidia, Qualcomm, and Xilinx have all recently opted to use IBM's foundry services for next-generation designs.
"Time to market is everything," Jelinek said. "A lot of people now are looking at IBM as a serious player."
IDMs, meanwhile, are developing technology with the pure-plays. But instead of committing production to their foundry partners, they are opting to make their chips in-house, which is putting a crimp in the foundries' ability to achieve the scale needed to fill future 300mm fabs, Jelinek said.
Motorola Inc.'s Semiconductor Products Sector, one of TSMC's largest customers, views the foundry as an integral part of its "fab-lite" manufacturing strategy. However, the company hasn't yet decided whether to give its 90nm volume business to TSMC or another partner, said a spokesman for SPS, Austin, Texas.
SPS is jointly developing 90nm process technology in an alliance with Philips Semiconductors, STMicroelectronics Inc., and TSMC at a pilot 300mm fab in Crolles, France. The facility is capable of supplying the companies with small-volume production and is expected to ramp by the end of the year.
"Right now, we all have internal factories that we will presumably transfer this stuff to," the SPS spokesman said, adding that TSMC would be the "incumbent" second source for 90nm production.
In defending its position, TSMC last week issued a statement that it has more than 20 90nm customer projects in the pipeline, and expects to start "risk production" in the third quarter of 2003 using copper interconnects and low-k dielectrics.
In an interview with EBN, the company denied it has fallen behind in 90nm development.
"The pushout you saw six months ago was based on demand-driven economics, not on technology--big difference," said Chuck Byers, director of brand management at TSMC-USA, San Jose.
Byers cited TSMC's mastery of 0.13-micron low-k, an effort others have since abandoned, as evidence of its technological prowess.
Still, analysts question whether the foundry business model will hold up in the 90nm migration, which coincides with a shift to 300mm wafers. The combination will produce twice as many die per wafer--creatin g more volume than most chip designs will ship in a year, according to analysts.
"You've got to have some serious sales in order to recoup the mask costs and processing costs," iSuppli's Jelinek said. "Where's the profit model for the fabless company that sells $10 million a year?"
Still, many foundry customers remain committed to the model long term. A spokesman for Agere Systems Inc., Allentown, Pa., said the company has 90nm designs in the pipeline at TSMC, and recently announced production using TSMC's 0.13-micron low-k process.
"No one else is doing that," he said. "The fact that we were the first to get out a 130nm low-k technology tells me foundries are at the leading edge, and it's absolutely a sustainable business model for us."
"Even the IDMs of the world are struggling" with advanced technologies, he added, citing a recent report that IBM abandoned its low-k pursuits after experiencing poor 0.13-micron yields.
SPS also believes foundries will remain competitive.
"They might be slowing down a bit, but on the other hand, they're becoming an integral player in joint development efforts because few companies can afford to go it alone anymore," the spokesman said.