Open-source code creeping into chip design

Open-source code creeping into chip design
SAN JOSE, Calif. — The debate about the role of open-source software in chip development will go on for years. But an interesting snapshot of the current state of the controversy emerged from a panel discussion at the ICCAD conference here Sunday (Nov. 10). Panelists from the EDA industry, academia, standards bodies and Intel Corp. all suggested — whether they knew it or not — that open-source is insinuating itself into the process.

The overall picture came into focus when Andrew Kahng, professor in the computer science and electrical and computer engineering departments at the University of California, San Diego, responded to an audience question from Daniel Gajski, director of the Center for Embedded Computer Systems at the University of California, Irvine.

The image the two academic researchers offered was of a multilayered software environment for chip design. Layers began at the bottom with operating system software, and then worked their way up through infrastructure such as databases, libraries and application programming interfaces (APIs) to the design tools themselves: simulators, synthesis tools, physical-design and analysis tools. But these latter tools, Kahng implied, should be seen not as monolithic structures, but as stacks that have their own infrastructure, user interface and plug-in "application" modules.

In conversation after the ICCAD panel Kahng and Gajski both proposed that open-source code would gradually spread from the lowest common levels — such as operating systems — up through the infrastructure layers and into the application realm of the design-automation tools themselves.

This view was anathema to the representatives of EDA vendors on the panel, who insisted that open-source was for infrastructure only. Intel's vice president for design technology, Greg Spirakis, agreed. "We see no need for our device models to be open-sourced," Spirakis said.

Yet at the same time Spirakis appeared to confirm that the process Kahng and Gajski described was already taking place. He cited Linux as a model of what open-source code could be, offering both a single package for users and a variety of sources. Spirakis showed data indicating that by moving from an unnamed proprietary Unix version to Linux on Intel platforms, the company had saved nearly a factor of five on the lifetime cost of ownership of servers while improving performance by up to a factor of two.

"By breaking the Unix monopoly we were able to free up lots of dollars to spend elsewhere — for instance on more boxes, each of which required more EDA tool licenses," he said.

Other speakers, without accepting the notion that open-source was infiltrating the stack, also traced the infiltration. Reports on the success of open-source programs in library development — OLA and the Liberty format — in generation of device models and in a language standard (SystemC) were all cited.

But these speakers, too, for the most part maintained that open-source code should be limited to infrastructure. The value of open-source, they argued, was that it is the only way to explicitly define a standard required for interoperability among multiple vendors — as in the case of a library API or a generator of models for different simulation engines.

UC's Kahng begged to differ, saying it was only a matter of time before open-source became a fundamental trend in tools as well. "If we all keep working independently and duplicating each other's efforts, the EDA industry is too small to hold up its end of Moore's Law," Kahng said. "We will have to have open sources and cooperative development to solve the problems we are facing now in very deep-submicron design."

Telling downloads

As evidence, Kahng cited a UCLA project that is producing a complete physical-design tool chain of open-source tools. He noted that the tools, available under an open MIT license, have been widely licensed by university researchers, who use them as a framework in which to test new point tools and as a benchmark against which to compare new algorithms.

More to the point, Kahng said that EDA developers and such semiconductor powerhouses as IBM, Intel and Philips have also downloaded the tools. In industry, Kahng claimed, these open-source tools are serving not only as benchmarks but also as platforms into which proprietary point tools can be dropped, and as a complete "Plan B" solution should a commercial tool chain cave in on a critical design.

This trend, he suggested, will grow as the challenges facing back-end tools grow, to the point where it is assumed that a physical-design flow will comprise a framework of open-source tools holding a combination of open-source, vendor and in-house modules.

Such a turn of events might not be unfamiliar to advanced-chip design teams, which have always worked with a mix of vendor tools, proprietary tools and vendor-negotiated workarounds. But for the electronic design automation community the shift could have profound implications, leaving big players looking much less like Microsoft and much more like Red Hat.