MUNICH -- Semiconductor foundries will play only a restricted role in the system-on-chip business, according to a panel of CEOs at the Electronica 2000 conference here last week, who claimed that SOCs are so complex and resource-intensive that "only a few semiconductor manufacturers will survive" in that business.
Ulrich Schumacher, Infineon president, summed up the panel consensus that foundries will play little role in making advanced SOCs. "It's almost impossible for a foundry to make complex SOC with embedded DRAM and embedded flash with different process technologies. Foundries will continue to do commodity high volume simple SOCs."
Pasquale Pistorio, president of STMicroelectronics, concurred that "you can't outsource complex SOCs. Very few foundries have the expertise and resources to do complex chips."
Fred Shlapak, president of Motorola Semiconductor Products Sector, felt foundries "are very important and contribute to the technology base for commodity high volume products. But if you get into complex custom SOCs, it is virtually impossible for foundries to produce and make a buck," he said.
Slightly more optimistic on the subject was Arthur van der Poel, president of Philips Semiconductors International, whose parent firm is a major investor in Taiwan Semiconductor Manufacturing Co. and its related foundries and also a large user of foundries. "Foundries can play a role in complex SOCs. Many are very large companies who know they must be able to take the most complex mix of technologies to make future SOCs. Otherwise they know they will have modest growth. They will do all they can to succeed in SOCs, and I think they will get half way there, if not all the way."
The generally gloomy assesment of the role of foundries in SOC was quickly challenged in Silicon Valley by representatives of those foundries. "Those comments sound as if they were thinking about foundries from five years ago," said Jim Kupec, president of UMC (USA), Sunnyvale. "They may apply to some backwater foundries, but they certainly don't apply to UMC. We offer broad SOC-type technology including RF, analog/mixed signal, metal-to-metal capacitor, and embedded DRAM. In fact, we are co-developing technology with IBM and Infineon for 0.13- and 0.10-micron.
"Foundry is competitive with the internal costs of many integrated device manufacturers (IDMs), because of economy of scale and the efficiency of foundries," Kupec added.
Foundries will play a vital role in SOC through collaboration with IDMs, said Ana Hunter, vice president of worldwide EDA and design services for Chartered Semiconductor. "That also fits in with what Chartered has been doing all along with our EDA partnerships.
"We certainly have a focus on the communications market, specifically RF CMOS as one of the key process modules we have developed for system-on-silicon," she said. "Our underlying approach is to develop basic CMOS processes and do add-on modules for system-on-silicon." On the issue of cost, Hunter contended that time-to-market factors often play a bigger role in whether a chip company uses a foundry for SOC, and that foundries can provide an advantage there by collaborating with IP providers.
"We're engaged with customers that are doing large-scale SOC designs, including LSI Logic," said Steve Della Rocchetta, executive vice president of sales and marketing for Silterra Malaysia, Sunnyvale. "We plan as part of our technology roadmap to offer RF capability. I don't believe that (SOC) is the exclusive domains of IDMs."
ST"s Pistorio said even IDMs have a short time to make comfortable margins on complex custom SOCs. "The customer dictates to us what price they want to pay. We are all competing aggressively to get the business, so we must try to help the customer reach their price target."
The chip presidents agreed that customers almost never will use multiple sources for an SOC chip. "Time to market is so critical for them that they can't afford to take time to get a second source," said Motorola's Sclapak. Pistorio said customers may compete different SOC designs among various chip makers, but only a single winner will be selected for each socket, and that company will become the sole supplier for the life of the socket.
The panelists laid out an extensive list of daunting challenges any SOC maker will face as IP gets more complex, software becomes more dominant, processes more sophisticated and costly, design more paramount, and resource investment skyrockets.
Shlapak said just trying to cope with the exploding number of gates on a chip is a formidable task. "We are moving from 100K gates per mm2 to 250K gates per mm2 to 500K gates. We must get very innovative in how we use all these new gates. We can't afford to just follow the traiditional approach of integrating a lot of separate IP cores on a chip. We must partition (SOC) chips very differently."
All the chip executives agreed that SOC business is growing, but they hedged on details, citing the inability and wide variations in defining what constitutes an SOC. Philips' van der Poel said if you define SOC very rigidly, the category may comprise 10 percent of the chip market. If defined very loosely, SOC could account for perhaps 30 percent. Most of the executives, however, concurred that their firms are investing two-thirds of their R&D resources in the SOC area.
Additional reporting by Robert Ristelhueber