At last week’s CDN Live event, the Cadence team detailed their vision for the future of the EDA industry: not point tools for engineers but standards-based, end-user oriented application–driven design.
CEO Lip-Bu Tan pointed out that the $350B electronics industry is still growing fast despite the recession, with much of that growth driven by an explosion of applications. Meanwhile the EDA industry is ‘mature’, plagued by slow growth and uninspiring stock performance. There has to be a better way, and everyone in the industry is looking for it. Cadence has an interesting approach.
EDA360—first announced in April—looks to move Cadence beyond design tools to complete designs. The first step is to expand your IP portfolio. Cadence’s recent acquisitions of Denali and Taray were strategic in that regard. Synopsys, with more cash to throw around, is racing down this road through its acquisitions of Virage Logic and Coware, emphasizing IP that’s guaranteed to work together—a major pain point for chip designers. Mentor Graphics has been building its IP portfolio for years, focusing on the embedded space. Cadence hopes to quickly expand its IP portfolio through “increasing cooperation with [our] ecosystem partners,” including IBM, TI, ARM and TSMC.
“There’s an app for that!”
Having accumulated a lot of IP, what do you do with it? Design stuff, using your suite of EDA tools as a virtualization platform that can turn system-level models into optimized hardware/software systems to support a range of software applications. Cadence has some gaps to fill in its tool flow to pull this off, but EDA360 is a roadmap to more than just more tools.
The key word here is ‘applications’, which CMO John Bruggeman claims need to drive hardware/software design, a process he calls “application-driven system realization.” According to Bruggeman, the current design model is suboptimal at best: “If hardware is designed first and software is appended later, it is difficult to optimize at the system level.” No argument there.
Then in 2006, according to Bruggeman, Apple came up with a new model with its iPhone: design optimized proprietary hardware and software platforms that work together. Application developers had to work within the limitations dictated by these platforms, which arguably limited their functionality, but if they did so everything worked well.
Instead Bruggeman advocates starting with an understanding of the software applications that will run on a given hardware/software platform, defining system requirements and then working down to hardware and software IP creation and integration. This entails reconfigurable hardware driven by software, and not just the operating system. “The OS…has a very generic understanding of the underlying software,” so applications need to take direct control of hardware functions. That involves EDA companies delivering “an optimized IP stack—from bare-metal software to the physical layer—that allows visibility into and control of the hardware from the OS and the application.”
Bruggeman has been advocating silicon-aware IP—not to mention virtual platforms—since his days at Wind River; the company worked with semiconductor firms to develop software drivers to enable VxWorks to work closely with leading processors. Intel saw the value in closer OS/silicon integration, which is why they bought Wind River. At Cadence Bruggeman has extended the vision upward to the application layer. This is hardly uncharted territory, but here the solutions aren’t quite so obvious.
How do you determine application requirements in advance when there are hundreds of thousands of them competing to be on your iPhone? And what happens when you have 60 of them running at once in the background with five of them active? How do you design for that? The answer would seem to be a hardware/software platform that takes its orders from the applications and does, within limits, whatever is needed to service them, either singly or in bulk. That’s what the OS is supposed to do. Perhaps what we need is smarter OSs. The alternative could be “over-designed hardware,” which is what Bruggeman says his approach is designed to avoid. Bruggeman points to Android as an example of an open “multi-application platform,” which does clarify his frame of reference.
How Hard Can It Be?
A system based on a reasonable set of assumptions anticipating the demands made by a well-defined range of applications is not an unreasonable idea. What Cadence is proposing is basically an ESL tool suite customized to generate complete applications. To supply that they’ll have to supply not just the tools but also an embedded software infrastructure—with a full suite of tested, integrated and verified IP—plus an OS, middleware and reference applications. No small task, and not one that any EDA company is even attempting today, though you could argue that they’re moving in that direction. Cadence certainly is.
Supplying reference designs could put Cadence in conflict with its semiconductor customers, who have long had to supply reference designs—right down to the Gerber files—in order to secure design wins. National Semiconductor’s online WEBENCH tools enable you to design fairly complex systems and generate a complete BOM, including non-NSC chips. They’re moving into EDA turf while Cadence plans to move closer to theirs. Bruggeman doesn’t foresee a conflict here, but I see plenty of potential for one.
Being able to supply a range of reference designs is going to require a lot of systems-level domain expertise, something that is in short supply in EDA companies. Semiconductor companies have been acquiring it for years in very specific areas in order to win sockets with key customers. But more generalized expertise is time consuming and expensive to acquire. Xilinx has set out to develop a range of FPGA-based reference platforms for some high-margin industries (medical, automotive, mil/aero) and they’re doing a good job of it, but this is a steep hill to climb.
Are We There Yet?
EDA360 calls for application-driven hardware/software creation, integration and verification. Getting there involves a lot more than filling in some gaps in disjointed tool flows. Even bracketing the idea of application-driven design for the moment, this all requires a smooth ESL design flow from system-level modeling right through tape-out, with early models being able to generate detailed, integrated hardware and software designs. We aren’t nearly there yet, though we’re getting closer (perhaps asymptotically).
Truth be told a lot of the focus on applications is driven by the attractive business model. When you buy a TV, for example, Samsung gets money from you once every 10-15 years. If you buy an EDA tool, you pay annually for seats. But if your TV boots up and you can buy a lot of cool apps for it, Samsung gets a nice ongoing cash flow—and Cadence, if they designed the apps or licensed the IP used to create them. That and better stock valuations are the lure behind applications-driven design.
EDA360 isn’t the only way forward for the EDA industry, and in a lot of ways it’s a restatement of what everyone else has long been doing. From that point of view Cadence has dug it way out of a hole, is executing well and has belatedly joined the party.
To be fair the EDA360 vision goes well beyond what most companies are attempting, and some parts of it are fairly radical. Neither Cadence nor its competitors are currently able to execute on all of it, but it does reveal a reinvigorated company with a vision, one that inevitably will be clarified over time. Cadence has spent a lot of time trying to get everyone in the company to understand the concept and get on board. That’s a necessary first step and in itself a major accomplishment on the part of the company’s new leadership.
As Gary Smith said recently, “This is the best Cadence to date.” I quite agree.
[To watch video interviews with Cadence execs re. EDA360 click here.]