"We're moving to 90nm with our eyes wide open, and we're confident with where TSMC is," said John Daane, chief executive of the San Jose-based programmable logic supplier. Daane's optimism stems in large part from Altera's closeness with TSMC, its exclusive foundry source for the past 10 years. TSMC last week presented Altera with a sculpture to commemorate the partnership, which has evolved from a pure wafer supply arrangement to a strategic technology collaboration.

"I think our relationship with Altera is preeminent in the area of technology collaboration," said Ed Ross, president of TSMC North America in San Jose. "Their R&D people are right in the fab with our guys, they're in the nits and grits of what's going on in terms of technology development, and there's complete visibility into every thing that's happening.

"The thing that really puts the fine point on it is that John Daane doesn't have to get involved in the operational stuff. There's no question that when issues come up--and they do come up--it's going to get resolved," Ross said.

It also helps that many of the technical hurdles manufacturers are facing with the 90nm transition were confronted by TSMC at the 0.13-micron node, he said.

"As we moved to 0.13, we changed lithography and we also changed materials, because we went to full copper interconnects and low-k dielectrics," Ross said. "The change to 90nm for us is just lithography.

"Altera was our first customer on copper. That was a huge debugging exercise for us," he said. "The risk--which took quite some time to take care of at 0.13--is under control, and I think the transition to 90nm will be non-problematic."

Despite much prodding by the investment community, Daane said Altera never considered bringing in a second foundry source to help mitigate potential supply disruptions.

The single-fab strategy is in stark contrast to that of rival Xilinx Inc., which tapped IBM Corp. as a backup to longtime foundry partner

United Microelectronics Corp., with questionable results.

A recent media report quoted Xilinx chief executive Wim Roelandts as saying he is dissatisfied with IBM's performance as a foundry, though a Xilinx spokeswoman said the comments were used out of context.

"There are companies that believe they need lots of suppliers. Our strategy is to have fewer suppliers and have strong relationships with those companies," Daane said. "The TSMC relationship is crucial to Altera."

Taking the 90nm step won't be straightforward, however. Indeed, it required a total rearchitecting of Altera's high-end FPGA family to overcome the technical drawbacks of the advanced process node--such as a five- to seven-times increase in leakage current. The reworking, Altera claims, will boost Stratix II performance by 50% and density by 75%, compared with its predecessor. Tools supporting the family will include the ability to make power trade-offs.

Cost is nevertheless a concern, which is why Altera will move its biggest, most expensive parts to the new process first, said Timothy Colleran, vice president of product marketing. Altera won't introduce its low-cost Cyclone II FPGAs in 90nm until at least 2005, when process yields are high enough to support low chip prices.

"We've got to make density and performance good enough that people will buy enough of the 90-nm products to bring down cost," Colleran said.

Others believe 90nm will enable lower cost points. Xilinx has said it plans to offer its low-cost Spartan-III FPGAs in 90nm by the end of this year.

Separately, Fujitsu Microelectronics America Inc. last week said it will move its AccelArray structured ASIC architecture to 90nm technology next year, a move the San Jose company believes will make the benefits of 90nm accessible to ASIC designers at a lower cost.