SAN JOSE, Calif. Open Source: EDA's next wave, or yesterday's hype? That was the question Larry Nagel, proprietor of Omega Enterprises (Randolf, N.J.) posed to a panel of researchers, tool developers and a user during the Sunday evening (Nov. 10) session at the 20th annual International Conference for Computer Aided Design (ICCAD-2002) conference here. Specifically, Nagel charged the panelists with documenting the problems, frustrations and solutions if any they had encountered in developing or using open-source code in the electronic design automation world.
First to reply was Andrew Graham, formerly president of SI2 and currently a consultant. Speaking of his experiences in overseeing the development and distribution of open-source code for the Silicon Integration Initiative, Graham said that the record was clear that open-source was the way to go for rapidly proliferating a piece of infrastructure software, such as an interface or a library format. He cited about 3,500 downloads for the organization's Open Library application programming interfaces (APIs) as an indication of this.
But Graham admitted there were issues. Specifically, he said that maintaining integrity of versions and keeping to a predictable migration path were major challenges for open-source code, and that those issues could be addressed only by either a controlling monopoly rendering the openness rather moot or by an organization that focused and organized the inputs of users in the market. He described SI2's process for doing the latter: starting out with a pair of architects in the proverbial locked room, then gradually passing control of the code to a small change team, then to an approval process by a coalition of interested organizations, and finally releasing the code to the community. He warned that creating a license was a major hurdle that for example, click-through licenses were virtually unenforceable against corporations. The bottom line, Graham said, was to somehow establish the trust of the user community.
Andrew Kahng, professor of computer science and engineering, electrical and computer engineering at the University of California, San Diego, took an altogether more militant stand on the subject, based on his work with the UCLA physical design tools project.
The development of a physical design tool suite under an open-source MIT-style license, Kahng said, had demonstrated that not only infrastructure elements but also actual tools could be built in the open-source world. The suite was serving both academia and industry, he claimed: academia as a test bed for further algorithm research and as a benchmark, and industry as a prototyping platform, a benchmark and, ultimately, as a "Plan B" in the case of a project in trouble.
In a powerful summary, Kahng warned that the entire EDA industry and the associated academic community, taken together, were too small and too poorly funded to solve the massive problems of extreme submicron physical design without the collaboration and the protection from reinventing the wheel provided by the open-source approach. "It's relatively easy for academic researchers. We are supposed to publish," Kahng said. "But open-source is a discipline that industry will have to learn as well."
The industry representatives on the panel, Kevin Kranen, director of strategic programs at Synopsys Inc. (Mountain View, Calif.), and Stuart Swan, architect with Cadence Design Systems Inc. (San Jose, Calif.), both argued that open-source software was for necessary common infrastructure only, and then only when it was necessary to have a common base on which to build competitive products.
The two representatives discussed their respective experiences with the Liberty Library format and with the SystemC initiative. Both cited struggles to get internal legal departments to agree to release proprietary source code without compensation, and more struggles with licenses. Kranen discussed how Synopsys had decided to considerably loosen its open-source license, originally based on the IBM public license. The company's goal was to abandon clauses requiring compatibility of developed software and to open-source any software derived from the licensed code. Competitors had objected to both clauses, he said.
Kranen emphasized continual support and interchange between users as a tool to keep people involved in the open software. Swan, who said Cadence decided to go open-source with SystemC in the hope of avoiding yet another language war, emphasized the need to confine control of the code to a small group and to ensure that all the participating organizations shared a common set of goals.
A somewhat different point of view was advanced by Greg Spirakis, vice president of design technology at Intel Corp. (Santa Clara, Calif.). Spirakis agreed with the EDA representatives that open-source code had a role in infrastructure, but not in the tools themselves. "Our experience indicates that operating systems and APIs should be open-sourced," Spirakis claimed. "But we see no need, for example, to make our device models available to others."
That was somewhat different from the view held by Colin McAndrew, director of enabling technology at Motorola Semiconductor Products Sector's Analog-Mixed-Signal Technology Center. McAndrew, who said that basically his business was acquiring simulation models, decried the environment in which a plethora of device models had to be interfaced, one by one, to the many simulators used within Motorola. "We want open-source to propagate models across simulators," McAndrew said. He identified two separate problems: the creation of new models and their interface to simulation tools.
On the creation side, he warned, "Everyone expects models to be free. But they aren't free to develop: they cost millions. And that leads to problems. I can guarantee that you aren't using the best models. The best model was developed by some researcher who can't get the support to distribute it. You are using a model that was developed by someone with the bucks to promote it as a standard." McAndrew claimed that developing models in an open-source environment would overcome much of that limitation. "Basically, the more help, the better," he argued. "But you do need a Bozo filter of some sort."
He also said that having a common open-source interface between models and simulation environments would be a huge help to users. This would avoid the problem of having to interface each new model to each of a dozen important simulators.