NVIDIA and Xilinx outsourcing more to Taiwanese foundries

Xilinx originally planned to use IBM as the secondary source for its most advanced 90nm-based FPGA (field programmable gate array) – Spartan 3. However, as UMC continues to widen its production volume and yield advantages over IBM, Xilinx has outsourced all of its orders for 0.13-micron and more advanced processes to UMC, according to a high-ranking executive at Xilinx in Taiwan.

In addition, Xilinx has shifted a portion of its orders for 0.18- to 0.25-micron processes from IBM to China-based He Jian Technology, a UMC ally, the source added.

“We’re a little concerned about IBM not having much capacity for us, and that could cause us to look somewhere else,” a February 11 EETimes report cited Xilinx CEO Willem Roelandts as saying.

According to the source, UMC’s 0.13-micron and 90nm processes have been qualified by Xilinx and Motorola. Although operating at over 90% utilization rates, the Taiwan-based foundry has provided additional capacity to its major customers via He Jian and its Singapore-based affiliate UMCi.

Not only providing capacity support, UMC has not yet raised its prices for major customers such as Xilinx, sources said. UMC earlier announced plans to raise prices as demand picked up.

Similar to Xilinx, Nvidia has increased outsourcing to TSMC for its high-end graphics chips produced with a 0.11-micron process, which TSMC has shipped in small volumes, according to a source with Nvidia.

Given that production on a 0.13-micron process can handled with 0.11-micron technology more cost-effectively, Nvidia has delayed its original production plan at IBM, the Nvidia source said.

IBM’s strengths in non-standard processing technology and the companies’ concerns about maintaining adequate capacity have prompted Xilinx and Nvidia to continue pilot runs at IBM. However, they may consider seeking capacity from other sources, the sources indicated.