News and New Products

Who will be left standing in DFM?

-- Electronic News, 2/9/2007

Electronic News/Electronic Business Editor in Chief Ed Sperling and EDN Executive Editor Ron Wilson sat down to discuss the future of design for manufacturing (DFM) with Jim McCanny, CEO of Altos Design Automation, Atul Sharan, president and CEO of Clear Shape Technologies, and Jim Hogan, a private investor and EDA expert. What follows are excerpts of that conversation.

Q: How many players will be left in the DFM market when the dust settles?
Hogan: This is part of a natural evolution. When there’s a big, climatic change, a lot of new species or new companies emerge. There’s been a compelling reason for investors to come in. There’s a lot of money spent on the manufacturing side and a lot of money spent on the design side. With DFM as an articulation point in the middle, it was like a land grab. Some of those things were good ideas, some were too narrow. People that can provide a platform for the design side or manufacturing side will survive. That will mean there are not too many—meaning one or two.
McCanny: I have a different opinion. There are lots of small companies and lots of point tools. As this matures, that won’t continue. But the majority of people who got funding are working on real problems and creating real solutions that are needed. I think there will be a high survival rate—maybe as aggregates or as acquisitions by some of the bigger players. The fallout will be relatively small. We get pulled into talking with many players in this space and we’re getting traction with customers. Some will do spectacular, some will do reasonably well.
Sharan: There were a lot of companies that were funded claiming to do DFM, so the first thing to figure out is who is really doing DFM. The bulk of DFM companies are actually improving the efficiency of existing OPC (optical and process correction) solutions or RET (resolution enhancement technology) solutions, and they’re doing it very marginally. I don’t think any of those are going to survive. Anyone who’s doing post-GDSII, competing with the mainstream OPC platforms, aren’t going to make it, and you’re starting to see that fallout. Then there are people on the design side who are furthering the rules-based methodologies, claiming to be DFM, but they’re not drawing on any manufacturing information. They’re not going to make it, either. The ones that are going to make it are the ones doing real DFM, moving from rule-based solutions to model-based solutions. They’re absorbing information the foundries are putting out and doing things the existing rule-based platforms cannot do.
Hogan: Saying you’re rule-based isn’t sufficient. The ‘D’ in DFM means you have to do design to be able to take rules and simulate or predict what’s going to happen, then fix things.
Sharan: That’s right. The ones who own the design side can predict what’s going to happen in silicon. You can’t do that with rules anymore because you’re sacrificing area and performance for accuracy.
Hogan: Everybody can get to 65 nanometers by brute force if you want transistors to operate at 250MHz. If you want to operate at 2GHz you’re going to have to extract all the potential that the process provides—that $5 billion worth of cement and steel in Taiwan.
Sharan: It’s the performance plus also time to market. That’s the advantage of DFM—extracting more from the process.

Q: Who owns that piece of the equation, the foundries or the designers?
Hogan: The foundries always own the behavior. The tools providers offer model formats and schematics, and that is populated by the foundries.
McCanny: The real confusion over DFM is terminology. Everyone is designing for manufacturing. The reality is that a lot of companies are working on the ‘M’ side. I do think there is a lot of promise in the ‘D’ side, as well, but a lot of it comes from how to extract this information from the manufacturing into something that can be digested by the design. It’s impractical to have the people doing design worrying about the end process.
Sharan: The fabs hold the valve. If they don’t feel it’s necessary, they won’t pass information to the design side. The design tool vendors have to react to that. Three years ago, the big debate was whether there would ever be DFM because the foundries weren’t going to release more information. Since then, TSMC has taken the lead in providing model-based information. There’s a good reason for that. TSMC lives off extracting value from the process platform. If they don’t maximize that, and they’re not flexible about it, they don’t make money. It’s already happening and the foundries have enabled it. Now it’s up to DFM companies to make that information useful.
Hogan: They’re in the business of manufacturing things. If they can express the behavior of their capability then it helps them attract more design. A couple years ago we wondered what it would take to get them to open up the keys to their kingdom. It became a competitive issue for them. They had to. They started giving people information they could use. One of the problems now is people have more information than they know what to do with.
Sharan: None of the existing design tools are able to use this information unless they come up with new analysis.
McCanny: If you can abstract information all the way up the process you can get away from some of these extreme margins the foundry has to impose to make sure what they take in will actually work. Right now they work off Murphy’s Law. If it can vary, it will vary in the worst possible way. They suspect maybe they’re being too severe, but they will only relax that in a way that has some mathematical rigor around it. You might get 15 to 20 percent more performance if you can reduce your power and reduce your leakage.

Q: Is DFM one solution, or does it mean different things to the designers and the manufacturers.
Hogan: As a designer I just want a black box. I want to take in the information the foundry gives me, use my analysis tools and my optimization tools to close my design. They are inherently different. With design, it’s whether I meet timing, power and functions. It’s binary. Manufacturing is a distribution. That’s been one of the problems getting DFM to go. We dealt with rule-based stuff for a long time. This stuff can’t be expressed that way. The first big step was to move from structure-based geometry to a model that talked about behavior. Once you have that, the information is populated by the foundry.

Q: Theoretically, DFM is supposed to be a bridge. But it looks the designers and the manufacturers are both claiming it.
Hogan: No, I think it’s just in the midst of being built.
Sharan: I think the bridge is already there because the foundry’s responsibility is to pass enough information so these models that predict the behavior can be articulated. The design tool vendor’s responsibility is to hide the complexity from the designer so they can still focus on timing and power. That’s exactly what’s starting to happen. I think the bridge is built.
Hogan: I grew up in Berkeley, and we always had the ugly side of the bridge. The side from San Francisco had lights and it was beautiful. The other side was the part that fell down. It’s the same in DFM. One side of the bridge is gold plated. The other side is put together with paper clips.
Sharan: If you look at the masks, you don’t know today if it’s a real defect or not. You don’t know if it’s going to affect your timing. Electrically driven mask inspection at 45 nanometers is going to be a must. There are different requirements for what the fab needs and for what the designer needs.
Hogan: When you hand off the design to the fab, you throw away all the electrical information. When KLA-Tencor does an inspection, they don’t know what critical paths are. They can have a lot more intelligent metrology if they can get more design methodology.
McCanny: The bridge has been built to the place where there is the most pain. It may not be a forward-looking approach that says the bridge should be built well so it lasts. The foundries are willing to provide information to avoid defects, but when it comes to models they’re worried about the cost and they don’t realize what the benefit is. There’s a little push-pull.

Q: Who’s going to own the bigger piece of the bridge?
Hogan: There’s a lot of money spent on design. About 18 to 20 percent of a new product is design. That’s what increases in budgets at design companies. They’re not spending on product engineers. The foundries will come north as far as it makes sense, probably into OPC.
Sharan: If you look at EDA in general, their valuations are terrible. Their valuations are two to three times P/E. Part of the onus is on the EDA companies themselves to get more value for all the wonderful technology they have. The consolidation is taking place on the foundry side, as well. TI, ST and Cypress are bailing out of the foundry business. That investment is rapidly collapsing into two or three foundry platforms.
Hogan: If you step back, maturing industries have this kind of profile. At the turn of the last century there were 100 automotive companies in Detroit. There were 5,000 EDA companies in Silicon Valley in 1989. There are two challenges here. One is that it is taking a lot longer for the processes to ramp because there is a lot more complexity. That puts economic burden on the foundries. It also opens the window longer for innovation, which allows large companies to take their existing technology and innovate linearly. The disruptive startup may not have enough time to see a return on investment in the window that’s available. That may be the overriding problem. Companies that have interesting technology get marginalized as the big guys extend the reach of their existing technologies—or if they do buy technology, they buy it for a lot less.

Q: Historically, in maturing industries, the consolidation happens around the strongest earnings flows. Does that suggest the EDA industry will be consolidated into companies that are making money, like the foundries?
Hogan: You can step back a little further and say EDA is an IT industry, and it’s just a matter of getting pulled up into the enterprise level as another piece of SAP or Oracle. It could happen.
Sharan: EDA companies throw out a lot of cash each quarter, too.
Hogan: That’s what you love about EDA companies. They have a lot of cash flow. That makes them attractive as an acquisition.
McCanny: The DFM space represents an opportunity for EDA. You can closely represent what you do in your design to what’s been manufactured as the number of foundries shrinks. EDA should be able to articulate that this is worth real money, and consequently the whole sector should be able to grow if managed correctly.
Hogan: The EDA companies get beat up pretty badly. But when you look at Synopsys and Mentor, at the handoff to the mask shop, they don’t discount. They hold their price. It’s not part of the big fab deal. They’re able to extract value disproportionate to the value they extract from physical implementation.
Sharan: If OPC was owned by equipment companies, it’s arguable that they would have been able to extract even more value. Part of the malaise is the inability to get more value. It begs the question why isn’t there more consolidation.
McCanny: It really comes down to what drives EDA, and I think it’s a bunch of engineers who enjoy a challenge. The problems presented by EDA are compelling for people who are into solving problems. The people who have worked in these large companies realize it’s hard to do that when you’re stuck in 40 hours of meetings every week. You get frustrated and you go off and build companies.

Q: Let’s roll this back. Is there an exit strategy for DFM companies?
McCanny: The reality is that small companies innovate, they do bring value, and they get absorbed.

Q: But who absorbs them?
McCanny: It’s still going to be the EDA companies.
Sharan: That may be a problem for any company in this segment, not just DFM. For the most part, DFM companies should get absorbed on the design side.