Model citizen: Electronic News spends some time with Silicon Metrics CEO Callan Carpenter - News - InterviewGale Morrison
Longtime EDA marketing executive Callan Carpenter, who is now CEO of Silicon Metrics, talks about the company, its competition with Circuit Semantics (CSI), the expanding role that library characterization and model creation has to play, and today's announcement that Silicon Metrics has just signed a multimillion-dollar contract with Agere.
Electronic News: How long has Silicon Metrics been around, and how long has there been a market for library characterization tools?
Carpenter: Silicon Metrics has been in business for five years. We received our venture funding in February of '98, though we were founded in July of '97. Then we shipped our first product in December of '98. And we've been focused in the space ever since.
But there's been a market for library characterization going back at least a decade, starting probably with Meta Software, at least as far back as their Master Toolbox tools. (Avant!, now Synopsys, purchased Meta in 1997.) Master Toolbox is for all intents and purposes a dead product today. That product has been largely set aside, though they did pioneer the market.
Electronic News: What's happening that's raising the profile of library characterization?
Carpenter: There is a difference between the market today and a decade ago. Let me back up.
There are two parts to what we do--the circuit analysis, i.e., the characterization, which is where we stimulate the circuits, take measurements, and derive certain electrical characteristics using Spice or Spice-like simulators; and modeling is the other half. What modeling entails is taking those derived electrical characteristics and formatting that into models that the various EDA tools consume, which is a lot harder than it sounds. There's multiple model formats. One's for simulation, for static-timing analysis, for synthesis, for timing driven place and route...
And you also have to verify those things. So the [quality assurance] of those is a big thing that we have to provide. They're somewhat distinct but related pieces.
The distinction between a decade ago and today is this: A decade ago, yes, everybody still needed the models. All EDA tools require models in some form or another. However, the technical rigor with which those models were characterized was not as great as is necessary today. Today there are many more complex physical phenomena in our deep submicron circuits, many more interrelated effects to take into account. There are very nonlinear delay models, power models, signal integrity models.
Much more thought has to go into the characterization today. The manufacturing processes are much more complex, and the margin for error in designs is much, much smaller. We're operating at much higher frequencies.
Certain issues, like leakage power for instance, used to be secondary or tertiary. Today that can be the primary consumer of power. Nobody was characterizing for leakage power a decade ago.
Electronic News: The competition between Silicon Metrics, your company, and Circuit Semantics has been very stiff. When Gary Larsen left CSI this summer, a lot of people said you're winning.
Carpenter: We do compete directly with CSI in one product area. That usually results in at least the two of us being in any benchmark for library characterization, and ours has been an overwhelming victory there. We focus on library characterization exclusively. They are focused in other areas. (Circuit Semantics has its own Spice engine, and a huge installation with the Intel Itanium teams). Their other areas take away from their ability to focus.
CSI does have customers at Intel. And Intel is a huge customer of ours as well. I'd wager they have spent more on us than CSI.
Electronic News: It's no small feat to get CAD dollars right now.
... CSI also has Hewlett-Packard. Agilent, which is the semiconductor spinout of HP, is a customer of ours. They use our I/O characterization tool. Their most recent product purchase was from us.
Carpenter: That's the thing that we thought was interesting in all of this. And Agere (which Silicon Metrics is announcing as a customer today) is a very clear example of this. We are seeing companies investing in [library characterization] as well as other technologies in preparation for a future semiconductor upturn.
When times are tough and manufacturing rates are down, that probably is the best time to retool, in terms of capital equipment and design automation. The retooling on the design automation side of things, that's where things like physical synthesis get a lot of airplay. Equally important, I think, is there's a lot of infrastructure that makes this work. These advanced libraries are the fuel that helps run those physical synthesis engines. If you are using these advanced timing closure design flows and these design flows that are going to take into account correct-by-construction signal integrity tools, you'd better have advanced libraries.
The Agere situation is telling in that here's a company [and] it's not a secret they lost 60 percent of revenue over the last two years. There's a company that is financially strapped, yet they just made a very large investment in this technology when they had an option in building it themselves.
It's also interesting that they are investing in something that has been dominated by internal tools. Even in the days of Master Toolbox, the majority has been serviced by internal tools--until now. And it's not just Agere. Major customers that have moved over--Hitachi and others--there's a trend toward outsourcing in this area to commercially available tools, and that's significant because it is so close to the process technology. It's down deep in the nervous system of design. It's very much the plumbing. Opening that up to commercial solution is a pretty big step. That's demonstrating the importance of being able to keep up with these changes, and it's hard for internal tool groups to do that.
It's hard not because they aren't smart but because it's very expensive. It comes down to a lot of companies saying, 'Should I invest my dollars in hiring engineers to build my next chip, or should I invest to write one more CAD infrastructure tool?'
Electronic News: Do the Big 3--Cadence, Synopsys, Mentor--see this? Do they need to?
Carpenter. I think that the Synopsys folks understand the strategic significance of libraries. If you look at the investment they've made in the library standard, clearly there's an understanding, perhaps better than any of the Big 3.1 don't think any of them have done a good job supplying tools, though. It's one thing to supply a library standard; it's another to implement that standard. Indirectly, both Cadence and Synopsys are investors in us [through their venture capital arms]. You could argue that Cadence and Synopsys are supplying tools through us.
Electronic News: Do you see the Intels, the IBMs, the TIs buying more, as a percentage, commercial EDA tools?
Carpenter: Again, it's not that these companies can't do what we do, it's more a question of is that the right thing for them to do? Not can they stop internal development, but should they? Commercial tools in characterization haven't been good enough, but they are now. It's just like 20 years ago, when people stopped developing simulation and schematic capture tools inhouse because the commercial tools got good enough.
Electronic News: At DAC and since then, silicon infrastructure and DFM and all of these design-to-manufacturing gap tools have been getting a lot of attention.
Carpenter: It's an area where you are going to see a lot more commercialization, and it goes under various names. Some people call it product engineering, or just "the white space between engineering and design." It's design optimization.
There's been a problem with DFM historically. You couldn't get design managers to care about DFM. Using that term was just the kiss of death. You couldn't get them to care. The tools budget was over with the design guys. Nonetheless, these are important areas. They really do start to fit in that gap between design and manufacturing. We need to be taking information from the manufacturing process side of the house and trying to make it work in design.
There are many more tools to come, I think, in that area in the next two to five years, which today are being handled internally. It's probably a pretty good growth area for design automation as an industry.
Electronic News: What other areas do you like for commercialization?
Carpenter: Another area where I would expect--and this is a trend happening for a number of years now--but I would expect to see a move away from internal Spice simulators. There's only a few left anyway. And they are very good, the internal ones, but I expect that they will give way over time to the very good commercial tools now on the market.
H-Spice (Synopsys') has been kind of the de facto standard. The commercial ones tend to focus on their niches and not really compete. For example, Cadence's Spectre tool doesn't really compete with H-Spice because Spectre is analog- and RF-focused.
But now there's a whole new class of Spice-like simulators from companies such as Nassda, Celestry and Synopsys. There are different options for trading off accuracy, performance and capacity.
Electronic News: What about this area of derivative cells?
Carpenter: Yes, there's a lot going on there, too. People like Prolific and Zenasis--there's a whole class of tools out there, doing more and more what you might call on-the-fly or in-situ physical circuit optimization. And we get tied into that flow. We happen to be partners with both because as soon as you start changing the physical library, you have to recharacterize and model it.
Traditionally in the library flow you have a fixed library and you design with that, and you optimize by the way you route.... Now, more companies are turning their attention to, what if I could change the underlying physical library? What if I could modify it to get some additional optimization? The answer appears to be yes. But now the old library is obsolete. Then you have to have tools like ours.
What's interesting for us is that takes our tool from something where our tools are used for the library to the use of it by designers. We think that's a market expansion opportunity for us. It's additional optimization for the designer, more market opportunity for us.
IC SPICE Market Share 2000
SOURCE: GARTNER DATAQUEST
Note: Table made from pie chart
COPYRIGHT 2002 Reed Business Information
COPYRIGHT 2002 Gale Group