SAN JOSE, Calif. In a bid to be the first company to deliver production-ready chips using the latest process technology, IBM Microelectronics said it has taped out field-programmable gate arrays designed by Xilinx Inc. that use 90-nanometer process rules.
The devices are set to be evaluated by Xilinx and are a precursor to early engineering samples, which Xilinx expects to deliver to its customers starting in March or April. A version of IBM's PowerPC microprocessor from the company's own product portfolio will follow shortly after.
Bijan Davari, vice president of technology for IBM Microelectronics, said that IBM has "a really good shot" at being the first to begin producing 90-nm-based chips in mass quantity.
Xilinx said it is also confident it will be the first to deliver FPGAs based on the latest process technology. By shifting to the 90-nm node, the company said it can shrink the size of its FPGAs by 50 to 80 percent, increasing the number of dice produced per wafer and boosting yields.
That means Xilinx will be able to bring down the price of a 1-million-gate FPGA roughly 17,000 logic cells to $25 by the second half of 2004. The company declined to disclose the product family that will be the first to exploit the new manufacturing technology, though the low-cost Spartan and high-end Virtex families are first in line, said Sandeep Vij, vice president of marketing at Xilinx.
The first 90-nm-based FPGAs will be produced on 200-mm wafers at IBM, although Xilinx plans to eventually switch to 300-mm wafers when its volume production kicks in. The company has already started shifting to 300-mm wafers in earnest with foundry partner United Microelectronics Corp. (UMC) and has started receiving 300-mm wafers from IBM for FPGAs based on 130-nm technology. By next quarter, Xilinx expects half its wafers will come from 300-mm lines. Xilinx will also tap UMC for production of the 90-nm devices starting in the second half of 2003.
FPGAs lead the way
Using FPGAs to prove a new process technology is a first for IBM. SRAMs were once favored because they are based on an array of redundant cell structures, which helps in identifying defects. PowerPC microprocessors have also been used to spearhead process development because they fetch a high price and usually need the latest process technology to stay competitive in performance.
FPGAs have many of the same attributes, but the companies were able to go a step further. Xilinx developed a variant of an FPGA structure, called a defect monitor vehicle, along with software tools to help IBM keep better track of defect density.
"You can detect any defect down to a single bit failure at every metal level. For an SRAM it takes three days to a week with repeated bit maps. For microprocessors it's a matter of weeks. In this structure it takes a matter of hours," Davari said.
Xilinx's FPGAs, which make heavy use of interconnect, should also benefit from IBM's use of low-k dielectric materials around the metal lines, considered a key ingredient to enhancing chip performance, said Richard Sevcik, Xilinx's senior vice president of FPGA products.
Sevcik said Xilinx has been working with IBM for the past six months on silicon-on-insulator technology, adding that it is "very likely" that the company will incorporate SOI into future FPGAs manufactured at the 90-nm node.