The panel--loaded heavily with a decided European viewpoint--generally agreed that large integrated device manufacturers (IDMs) hold an distinct advantage over pure-play foundries in quickly ramping production of complex system-on-chip (SoC) designs that encompass a broad range of functions, beyond just leading-edge CMOS logic. The majority of executives on the panel also questioned whether any of their companies could ever use foundry wafers to make a decent profit in high-volume, leading-edge products.

"It is virtually impossible to make money with foundry wafers at the leading-edge technologies and at high volumes," asserted Fred Shlapak, president of Motorola Inc.'s Semiconductor Products Sector.

Motorola chip's unit, based in Austin, Tex., has been pursuing an aggressive strategy to outsource a large percentage of its semiconductor manufacturing in recent years, but the group's new management team--including Shlapak--hinted recently that the company could be moderating its previous foundry stance to become more balanced (see September feature story). After today's panel discussion, Shlapak said Motorola's foundry target is now 12-15% of its total frontend capacity--much lower than what was once described as a 30-to-50% goal. Contract assembly manufacturers are being used for 38% of Motorola's chip packaging and testing needs, he added.

Shlapak, a 30-year veteran of Motorola who has served as president of the company's European semiconductor group, said foundries face a growing problem as requirements increase for specialized processes beyond leading-edge digital CMOS. "Where many of us do business wireless and communications markets, it is virtually impossible to compete at high-volumes in baseband technologies from foundries and still make a profit," he said. "It's impossible."

Generally agreeing with Shlapak were Pasquale Pistorio, CEO of STMicroelectronics, and Ulrich Schumacher, CEO of Infineon Technologies AG. Disagreeing with the notion that silicon foundries will be greatly restricted in leading-edge system-on-chip (SoC) technologies was Arthur van der Poel, president and chairman of Philips Semiconductors. The Dutch chip maker is partnered with silicon foundry giant Taiwan Semiconductor Manufacturing Co. Ltd. in a new joint-venture fab, called Systems on Silicon Manufacturing Co. Pte. Ltd. (SSMC)--in Singapore.

"I believe they foundries can play a certain role," said the Philips semiconductor president. "If you take the most complex mix of technologies--from digital to very high-frequency analog--they might not play that role," he admitted. "But there is more than just standard CMOS processing underway in the foundry business. The foundries must include the system-on-chip spectrum--in a certain way--otherwise, they will end up with only modest growth opportunities and not be able to serve their customers.

"They will try to do their best to extend their territory, and they will certainly succeed, at least halfway," added Arthur van der Poel, hinting that he also believes IDMs can do a better job in some circumstances producing their own next-generation system-on-chip designs.

The panel members--made up of executives from Samsung, Motorola, STMicroelectronics, Philips, Infineon, and Europe's Medea industry program--generally agreed that next-generation SoC will require an increase in functions on a chip--including mechanical devices, such as microelectromechanical systems (MEMS) for sensors and accelerometers, as well as optoelectronics. But even the current range of SoC technologies is beyond the scope of foundry suppliers, said Infineon's Schumacher.

"We're not saying foundries are not at the leading-edge of CMOS logic--they are, whether its three months later than some IDMs, or not, it doesn't matter," Schumacher said. "System-on-chip is not a main logic process. It requires embedding all kinds of memory on chip. It's RF radio-frequency, and other technologies. You don't have foundries today providing all of these technologies.

"You cannot get leading-edge RF supplies from foundries, or even power chip supplies from foundries," said the Infineon president. Schumacher--and several of his colleagues--also expressed concern about the costs at foundries.

"In the last 18 months, I would say the cost of a foundry wafer is two times higher than your own manufacturing costs," Schumacker said, referring to higher prices for processed silicon resulting from recent shortages of capacity. But even when foundry supplies become plentiful again, he said, foundries must still make at least 30-to-35% return on investment. That return on investment will make wafers unprofitable for major chip house in a number of highly competitive markets, he said.

"You must be a bad manufacturer to have higher prices than foundry supplies," Schumacker added.

STMicroelectronics CEO Pistorio argued that foundries can be used strategically for certain products, but he also maintained that "you must master your own manufacturing" in the system-on-chip era. Nearly everyone on the panel agreed.

Motorola's Shlapak said fabless semiconductor companies have been able to succeed in recent years by concentrating on high-value intellectual property (IP) for specific system functions, but the result of this strategy begins to weaken as those suppliers to move into new markets or attempt to sell system-on-chip products.

"The fabless companies are obviously important to the overall technology base from a system point of view. If they have a strong IP portfolio for system functions the manufacturing technology might not be that important," said the Motorola semiconductor president. "But as a company migrates from the CDMA segment to GSM, it is impossible to compete with technology from foundries," said Shlapak, referring to the intense competitive pressures faced by chip companies in the Global System for Mobile segment of cellular phones.

Schumacker said it is likely to become harder for pure-play foundries to be competitive in a number of new chip technologies, which will be required for next-generation communications and wireless products. "It is significantly easier to develop 0.18-micron CMOS than develop indium phosphide, gallium arsenide, or silicon germanium technologies," he argued. "It will not be so easy because it is not just a question of money but it's also a question of expertise."