Programming The Future

By Ed Sperling -- EDN, May 6, 2005

Wim Roelandts, president and CEO of Xilinx, sat down with Electronic News to discuss the shifting balance between FPGAs and ASICs, the role of FPGAs in consumer electronics and supply chain gains. What follows are excerpts of that conversation.

Electronic News : Is business recovering?
Roelandts : Yes. I had been telling people that the slowdown last year was due to an inventory correction. It's now clear that is what happened. People were overbuying in the first half, cautious in the second half.

Electronic News : But they really didn't overbuy that much, compared with the 2000 downturn, did they?

Roelandts

: No. There's no comparison.

Electronic News : Is the supply chain being managed better than in the past?

Roelandts

: Absolutely. The sub-contract manufacturers and distributors are managing their inventory much, much better. What they are doing is pushing inventory back to suppliers like us, which in a way is better because we are better able to manage inventory. We keep most of the inventory in a die bank, which gives us much more flexibility in how we use this inventory. It's sort of like Dell having all these sub-assemblies and then customizing it at the moment the order comes in. It's build-to-order. As soon as the order comes in, we take it out of the die bank and package it the way the customer wants it and then ship it to them.

Electronic News : How much inventory do distributors have now?

Roelandts

: About 30 to 35 days, which means they turn inventory about 10 times per year. In the past they turned inventory about four times per year. Where inventory is still stored that's tough to measure is with the end customer. The subcontractors that deliver boards to the end customer do their own integration. That's where you often have excess inventory.

Electronic News : One area of growing concern among many companies are the European Restriction on Hazardous Substances (RoHS) rules. Will the RoHS restrictions have any impact on your sales?

Roelandts

: It's going to cost additional money, but I'm not so sure it's going to have an impact on sales.

Electronic News : But will boards start coming back with parts that aren't soldered correctly or which contain leaded parts.

Roelandts

: That could happen. It's a serious problem from an inventory management point of view. We have taken a position in favor of moving forward on this. The big issue is in consumer electronics. You're not going to find a Cisco router sitting on a dock somewhere. It's PCs and DVD players. The consumer market is really hot on this. The capital goods market is slower to react. It's also easier when you have five parts in a box to make sure they're lead-free than when you have 100,000 parts in a box.

Electronic News : FPGAs seem to be eating into the market for structured ASICs, while some of the mid-tier ASIC companies are eating into the sales of FGPAs when they begin shipping in volume. Is that a correct assessment of the market?

Roelandts

: It's happening to some degree. I'm not sure if there was ever a case where companies like AMI were ever successful in a Virtex conversion, though. It really is a complex product. There have been cases where they tried to convert it but didn't because of the clock system, which is very sophisticated. Can you take a PLD and do that? Yes, that has happened. But the more complex the product becomes, the more difficult it becomes to create an ASIC. With Virtex 4, less than 50 percent of the chip is programmable logic. All the rest is hard cores. You get the PowerPC processor and the 11Gbit transceiver.

As far as the structured ASICs and traditional ASICs, this is a market where we don't play. These are people who use our chips for prototyping. The structured ASIC has taken share from traditional ASICs, and vice versa, but not from us. Today, our largest chip is about 2,000 logic cells, which is about 4 million to 5 million gates, not counting all the hard cores. This is pure programmable logic. It costs about $3,000 apiece for a 90 square mm die. Six years from now it will be around $10. If you have a 4 million or 5 million gate capacity plus all the other stuff-transceivers and memory and you name it-for $10, why would you go to a structured ASIC or an ASIC?

Electronic News : That brings up an interesting point. Who can afford to stay on the Moore's Law road map? While it doesn't mean that companies will necessarily abandon ASICs, they may not be at the cutting edge because it costs too much.

Roelandts

:  That's correct. We have to be able to compete at 180nm. When we are at 65nm, we will be able to do that very well. We will be at 65nm next year.

Electronic News : Any technology hurdles along the way?

Roelandts

: I would be lying if I said there weren't any, but it's going to be a little easier than I thought. A lot of people felt that for 65nm we would have to move to new materials. The people who are behind the SIA road map decided to keep the 90nm transistor for 65nm. That means we don't have to go to new materials. The question is how much can you scale. This is where a lot of people get confused. We have reached the end of Moore's Law with silicon dioxide.

Electronic News : But only as far as scaling goes, right?

Roelandts

: Yes. But you need to move to new materials if you want to scale again.

Electronic News : Do suitable materials exist for gates in future chips?

Roelandts

. Definitely. But any time you move to new materials it is never clean. This is what happened with copper.

Electronic News : But aren't these high k materials soft and difficult to manufacture?

Roelandts

: Yes. No one has experience with it, it's really soft, there are different defect factors that nobody knows about. That's why 65nm is going to be easier. There is no new material. People will use new materials at 45nm. But that is at least three years away.

Electronic News : As we move down the Moore's Law road map, does the crossover point change to where it makes economic sense to use an FGPA versus an ASIC?

Roelandts

: It's profitable at 90nm at a half-million ASIC gates. At 65nm, it will be profitable at 1 million ASIC gates. ASICs will never disappear. They will always be the best solution for some applications. The question is can they build it in that technology. The problem with ASICs is that you have to have a pretty uniform structure because every chip has to work. What is happening at 65nm is your distribution gets wider and wider. To build an ASIC at 45nm is almost impossible today because the distribution is so wide, but you have to go back and make sure all the chips work. We can build it and test it, and charge more for the faster ones and less for the slower ones. Our cost structure is spread over thousands of designs. That is something ASICs cannot do.

Electronic News : Can you continue providing the tools necessary to build these chips, or will they get so complicated that Xilinx no longer can deliver them? You've been building some of the tools cost into the cost of the chip, haven't you?

Roelandts

: Absolutely. And the reason we do that is our architectures keep changing. Some of the schematics that we build today won't work at 45nm. We will change the architectures and we will change the tools. This is just the place and route tools.

Electronic News : Are you starting to see any success in that market among traditional EDA vendors?

Roelandts

: The company that is most successful is Synplicity. They have a focus to do synthesis tools for FPGAs. That's much easier than making synthesis tools for ASICs and getting them to work for FPGAs. Synopsys and Mentor Graphics have made a big effort to get in there, and they're starting to realize that the world is shifting. There are less and less ASICs, and they're struggling to figure out what they should do. FPGA users are not used to paying $100,000 for a seat. Synplicity doesn't charge that. They work with a much lower cost model, and they have very good tools.

Electronic News : Still, the vast majority of chip designers being hired these days are software engineers.

Roelandts

: That's true. [Traditional EDA vendors] are in a quandary. Their traditional customer used to do 1 million gates. Now they're doing 10 million gates and they claim the tools aren't good enough. There's a gap between the tools and the technology. Their traditional customers are screaming for more tools for ASICs, and no one is paying attention to FPGAs.

Electronic News : How about simulation?

Roelandts

: There are areas where these companies can do very well. Simulation is one of them.

Electronic News : What's the next step for FPGAs?

Roelandts

: The next step is really to make FPGAs disappear. Today our customers are hardware engineers. But FPGAs are programmable devices. If we can create a level of abstraction that appeals to software engineers, we can increase our customer base by at least 10x. That's really where our future is. As long as you have a set of interfaces that you can program to, you don't have to know what the hardware looks like.

Electronic News : How does the convergence of everything into consumer electronics affect Xilinx?

Roelandts

: Last quarter, 15 percent of our revenue came from consumer electronics. That has made the world far more complicated for ASIC vendors because every group has its own standards. You have all these things coming together -- networking, entertainment and computation. In automotive, what you have in your house you want in your car. All of these areas have standards, and all of these standards keep evolving. That's why programmable becomes so attractive. Companies don't know exactly what the customer needs. The digital consumer is only the first step. Five years ago when we said we were going after the consumer market, people said we were crazy.