The outcome of the talks could determine whether Xilinx will continue to rely on IBM as a second production source for its most advanced FPGAs. The leading programmable logic vendor is now using Taiwan's United Microelectronics Corp. (UMC) for the bulk of its production output, including devices built on 300-mm wafers and its first 90-nm FPGAs, known as Spartan 3.

Xilinx wants to maintain its policy of having two foundry sources, but IBM has been slow to ramp up production of its most advanced parts. "We're a little concerned about IBM not having much capacity for us, and that could cause us to look somewhere else," Roelandts said in an interview during the company's 20th anniversary dinner party here.

Maintaining adequate production capacity is emerging as a central issue for Xilinx and other chip makers this year as the demand for semiconductors picks up. Xilinx, which began volume shipments of Spartan 3 last month, has indicated that it will not be able to meet early demand for these low-cost FPGAs. Later this year, it plans to introduce its first high-end Virtex FPGAs based on 90-nm design rules.

Roelandts said there have been many internal debates about whether to tap other foundries. "We've not made any decision yet but probably it's something we need to look at seriously," he said.

One big problem that IBM is facing is that it isn't pushing through as much volume as UMC, which makes hard to improve yields. IBM "is chasing a galloping horse," he said. "Every day UMC gets ahead."

One solution Xilinx is proposing is to buy parts from IBM at a lower cost in order to stimulate more production, Roelandts said.

Aside from their foundry relationship, IBM and Xilinx are involved in a technology sharing pact that gives Xilinx rights to use IBM's PowerPC architecture and IBM rights to use Xilinx's FPGA fabric.