LAS VEGAS Xilinx Inc. is talking to ASIC vendors about licensing its field-programmable gate array blocks to them. If such discussions are successful, ASIC companies will be able to include FPGAs in their ASICs and provide field programmability to their customers.
Wim Roelandts, president and chief executive officer of Xilinx (San Jose, Calif.) the company that invented the FPGA in the mid '80s gave the clearest indication yet that the company would license its FPGAs to ASIC companies during the final-day keynote Thursday (June 21) at the 38th Design Automation Conference here.
Roelandts talked about his company's discussions after his keynote, when Michael Riepe, a member of the consulting staff at Magma Design Automation Inc. (Cupertino, Calif.), asked Roelandts about the possibility of using Xilinx FPGA structures in ASICs.
"For several years we have been talking with ASIC companies about providing them with FPGA cores," answered Roelandts. He later said, "We believe it's the right way for the industry to go."
Speaking to EE Times after his speech, Roelandts said that "we have some discussions. We are certainly open to consider licensing."
When he was asked whether any discussions were close to a conclusion, he replied, "I don't say it is imminent. One doesn't know whether these talks will go quickly or slowly."
Given its history with Xilinx, IBM Microelectronics is an obvious candidate to take up Xilinx's FPGA architecture. IBM has already licensed its PowerPC processor core and the CoreConnect bus to Xilinx. The core and bus serve as the basis for Xilinx's Empower range of hybrid diffused cores and FPGA devices.
In July 2000, when the IBM-Xilinx deal was announced, it did not include an IBM license of Xilinx technology. But John Kelly, general manager of IBM Microelectronics, did say at that time that he was willing to consider placing FPGA cores onto ASIC systems-on-chip.
Although ASIC vendors have traditionally competed at the bottom end with FPGAs, the mixing of ASIC cores and FPGA fabric is part of a blurring of the distinction between ASICs and FPGAs that Roelandts spoke about during his keynote.
Hailing the entrance of his company's FPGA products into the mainstream, Roelandts went on to offer arguments showing why FPGAs are useful in reducing time-to-market, reducing the cost of getting to market and maximizing the time products last in the market. An essential part of this, Roelandts said, is the ability to do in-service equipment upgrades over the Internet.
"Sending hardware through the network; it might sound like science fiction, but it's really happening. Of course, we are just sending bit streams, but it expands a product's market because you can expand its utility afterward."
Roelandts said 30 to 40 percent of Xilinx customers now use this field upgradability in their products, compared with less than 5 percent three years ago.
But Roelandts spent most of his time focused on the platform-level hybrid devices that his company is introducing. He said that Virtex-II Pro would support diffused processors, sea-of-multiplier-type digital signal processors and system I/O structures at the same time.
"For Virtex-II Pro we plan the integration of a transceiver at 3 and 8 Gbits/second. A transceiver does the serialization and deserialization and clock recovery. So it will take a 3-Gbit/s serial stream and convert it into 32 parallel channels, which the FPGA can easily cope with, and do the reverse on the way out."
Roelandts said such transceivers would be the basis of the way Xilinx would support Infiniband, Gigabit Ethernet and Fibre Channel I/O directly on its Virtex devices. Roelandts added that the products would be "on the market in the next three to six months."