It's time for academic researchers in reconfigurable computing to stop relying on commercial field-programmable gate arrays (FPGAs) and tools and to form a university consortium to develop open, simple FPGA hardware and software more suitable for research.
So says John Wawrzynek, professor of EE and CS at the University of California, Berkeley, who recently chaired a panel discussion of academics and industry representatives on whether such a call to action is necessary.
"Due to the proprietary nature of commercial arrays and the fact that the devices aren't really designed with computation in mind, it's hard for the community to build their own tools to support them," Wawrzynek says. "This has created an ease-of-use issue that isn't well-addressed by industry."
Andre DeHon, an associate professor at the University of Pennsylvania's School of Engineering and Applied Science, agrees that a certain amount of innovation needs to come from universities, which he says are currently limited by FPGA vendors' toolsets.
"It's academia's role to contribute ideas to the space," DeHon says, "especially in those areas that don't make short-term economic sense for industry to pursue but may be highly valuable in the long-run."
But Mike Hutton, principal design engineer in the office of the CTO at FPGA supplier Altera Corp., cautions that to design a single open-source FPGA requires the academic community to agree on process, capacity, features, and other tradeoffs. "I'm not sure that's possible," he says. "Research targets a wide variety of problem-domains and I don't see how one device can support them all."
Besides, he adds, "Altera spends $100 million to make a single new physical FPGA which is a tremendously complicated, expensive undertaking. Designing a chip would be a very ambitious project for academia. It's a much better idea to drive academic research towards having a strong set of open-source modeling tools instead of actually building a physical device."
As an example, Hutton points to the VPR software developed at the University of Toronto in 1998 "which has been used as the base for research on FPGA architecture and software for the last 15 years. Altera uses a similar methodology to research new products long before they are physically fabricated."
Nevertheless, Wawrzynek sees the need for not only open-source tools but for an open-source hardware design that is free for anybody to use.
"I would argue that commercial FPGAs are perhaps overly complicated in the same sense that microprocessors back in the early '80s became overly complicated," he says. "What we may need is a complete reset; what is the minimal simple array that we need to get a reasonable efficiency? We need to come up with a set of benchmark applications that are the target set."
Wawrzynek says that if there's enough interest in creating a consortium, he'd be willing to send a white paper to government agencies, like DARPA and the National Science Foundation, to determine whether they'd be in favor of moving the project forward. But, he says, first he would need a few additional senior people to climb onboard. He's anxious to hear from them.
Paul Hyman was editor-in-chief of several hi-tech publications at CMP Media, including Electronic Buyers' News.
No entries found