Will systems-on-chip (SoCs) become so expensive to design that people are going to buy chips off the shelf, and differentiate products through software alone? That's one question that was put before a panel of EDA industry experts at the DVCon conference Feb. 29, 2012. Short answer -- no, but we do need tools that can verify software apps.
The panel, titled "The Resurgence of Chip Design," was moderated by J.L. Gray, vice president and general manager of Verilab and author of the Cool Verification blog. This was the traditional DVCon "industry leaders" panel, except this time, there were no EDA vendor panelists (that is, executives from companies whose primary business is selling EDA software). Panelists included:
- Ted Vucurevich, CEO of Enconcert (and former Cadence CTO)
- John Costello, vice president of IC design, Altera
- Gary Smith, chief analyst, Gary Smith EDA
- Jim Hogan, founding partner, Vista Ventures LLC
- Victoria Coleman, vice president of emerging platforms, Nokia
Gray opened the panel on a personal note. A long-time IC verification engineer, he recounted being told that chip design was going to be relegated to a few niche companies and that everybody else is going to buy chips off the shelf and "write awesome software." He started to think maybe he was in the wrong industry. "But as luck would have it, in the past few years, we've been seeing a huge uptick of interest in building and verifying new devices, and an interest in how those devices interact with software."
Lowering the Costs of SoC Design
So, is there indeed a resurgence in chip design? Smith noted that IC design never went away, although there was a "dip" due to the high cost of SoC design, which was recently approaching $77 million for the average high-end SoC. But Smith said he discovered small groups of companies doing "multiple platform based designs," a methodology that can reduce that cost to $44 million. Asked what "multiple platform based design" is, he said that it starts with a base platform such as Snapdragon or OMAP, and then designers add full IP subsystems - essentially "platforms" in their own right - that are already verified.
Coleman said she was glad to hear about progress in this area, but noted that platform-based design is not easy. "You have to get the user experience you need at a power and performance that's tolerable, and often you have to do extreme optimization," she said. "It's really hard to meet things like power requirements in those platforms and do so with a performance that's acceptable."
Hogan noted that systems companies "don't differentiate on hardware too much," but many still want to develop their own SoCs. Why? "Because they would give up a lot of the value in their value chain to an SoC provider." Further, hardware can give advantages in performance, power, area, and cost. He noted, however, that "we get trapped into the notion of an ASIC fabric" while "FPGAs take the silicon risk out."
That was a perfect opening for Altera's Costello, who noted that "FPGAs provide a way to aggregate SoC design for customers who can't afford to do their own SoCs." More and more, he noted, "customers are looking for complete solutions, not just raw silicon. They want EDA software to design it, IP to go on top of that, and potentially software to go with it."
Verifying Software Apps
Vucurevich's startup, Enconcert, offers live streaming media applications such as BackStage Pass, an iPhone application that helps musicians create an enhanced live event experience. He talked about the deep "technical stack" of hardware and software that's needed to put these applications together. While it's easy to envision the applications, verification is much more difficult.
The problem? "Folks like Nokia, Apple, and Google work very hard to keep me away from what we used to do, and they fail miserably. There are many corner cases where the timing doesn't do what it needed to do, and the OS and HAL [hardware abstraction layers] are trying to keep me away from the fact that a timing constraint didn't work. All those wonderful layers of software that are supposed to insulate me make it virtually impossible to figure out what happened."
Coleman talked about "superbugs" that may be 50 layers deep. Static analysis tools that could dig them out are either unavailable or unused. "The EDA industry has focused on the hardware layers a little too much. Maybe it's time to think again about the notion of codesign," she said.
"The problem that's becoming apparent to everyone is in the context of the application running on hardware, but how do you verify that? You can try an instruction set type of verification, but what you really want to test is the scenario," Hogan said.
But Will They Buy Tools?
Speaking from the audience, an attendee noted that his company has a business in embedded software as well as EDA, and has taken technology developed initially for testing software and repurposed it for hardware verification. Now they're trying to bring it back into the software space, and "it's very, very difficult." One thing that killed UML [Unified Modeling Language] for embedded systems, he said, is that "every time somebody comes up with something innovative, somebody coming out of school develops it and puts it out into open source and kills the entire market."
Coleman responded that "technologies like UML have not kept up with the kind of software we write today." When she was at a previous job, she noted, "money was not the problem - I couldn't find the tools."
"As I talk to the big guys, they have such a problem with software that they are now shifting a lot of money they were spending on hardware over to software," Smith said. Hogan noted that "companies pay to get rid of pain. There's a lot of economic pressure to get things out, and that is what is going to drive us."
"As CEO of a company with a web-based application," Vucurevich said, "I wouldn't blink twice to pay 10% of what I'm getting from my customers for every application I ship, if it could guarantee that functionality was verified to a given level." When Gray commented that "these conditions could never be met," Vucurevich said he wanted "something I can measure and make a business decision around. I will attach value to that."
What we really need to do, Coleman suggested, is to "shift our way of thinking" with respect to software development and verification. "In all honesty, I think the EDA industry has been absent from that conversation. Maybe it's time you guys came back."
Other DVCon 2012 Coverage
DVCon 2012: Accellera "Town Hall" Meeting Explores Future of EDA Standards
Hardware/Software Codesign: Pink Elephants on Parade?
DVCon User Panelists: Is Low Power Design Worth the Costs?