For the past 11 years, Wim Roelandts has been the face of Xilinx Inc., a premier example of a Silicon Valley startup that crossed the chasm to success. In August, Roelandts announced plans to retire as CEO, though he will remain chairman and continue to be active in the industry. The 62-year-old executive was firing on all cylinders when EE Times' Rick Merrit caught up with him for a wide-ranging discussion about his company and the industry.
With Roelandts at the helm, Xilinx grew its annual revenue from roughly $450 million to $1.8 billion, despite a downturn that rattled the entire industry. Roelandts, who describes himself as "Flemish and stubborn," joined the fabless FPGA vendor after a 30-year career as a systems executive at Hewlett-Packard Co., thus becoming an example of the risk-taking Valley culture he so admires.
Roelandts also has been the face of the fabless semiconductor sector, via a four-year stint as chairman of the Fabless Semiconductor Association. In addition, he sits on the board of Applied Materials. And on Nov. 14, he was named chairman of the Semiconductor Industry Association, becoming per- haps the first CEO of a fabless company to hold that post.
EE Times: How is the search for a new Xilinx chief executive going?
Wim Roelandts: We made it public to bring candidates out that you otherwise wouldn't think about. We have narrowed it down to the top four and are doing in-depth reference checks.
EE Times: What is the next phase for Xilinx?
Roelandts: Clearly, in the low-volume infrastructure space, programmable devices are pretty much dominant in systems such as routers and basestations. But these products are perhaps 25 percent of the market, and 75 percent is high-volume consumer devices--PCs, cell phones and so on. Today, about 17 percent of our revenue comes from consumer and automotive. It was zero percent six years ago.
A few months ago, we got the highest order we ever received: 8 million units for a single socket. We are pretty regularly getting million-unit orders for a socket now. I can see a stage where anyone who does logic design will consider Xilinx. We are perhaps halfway there.
Instead of driving high clock rates, we use parallelism to get high performance, which means the power/performance equation is in our favor compared with microprocessors. But the big challenge is software. You want to make the [FPGA] software [more] transparent. I am convinced in the next five years or so you will be able to transfer from C code to gates without modification.
EE Times: What's the outlook for the fabless model?
Roelandts: With the announcements from Texas Instruments and STMicroelectronics, it's clear the medium-sized chip makers are going fabless. The fabless model is one of the core business models. But the fact that design is done in one company and manufacturing in another is going to cause issues of integration. The other issue is determining who can afford the technology.
Moore's Law will end for economic reasons before it ends for physical reasons. A 45-nm fab will cost $4 billion to $5 billion, and a 32-nm fab will cost $6 billion, because an immersion stepper is 50 or 100 percent more expensive than the previous generation.
EE Times: Will this drive more fab alliances?
Roelandts: Yes. The question is how many of these alliances will exist over time. There will not be many.
Sooner or later, UMC [United Microelectronics Corp., Xilinx's main fab partner] will have to join one of these alliances or create one.
The Japanese have been trying to create [an] alliance for years without much success. In my experience, if there are too many players, it becomes a danger. IBM now has five or six partners [in its Common Platform alliance], and that is a question mark, but they have probably thought about it. I would not be astonished if Toshiba [another Xilinx fab partner] joins IBM, because the two companies have links.
There will certainly be a Japanese alliance of some kind, there will probably be an alliance around TSMC and there could be one around UMC, but you are down to the fingers on one hand.
EE Times: What does this mean for the semiconductor equipment sector?
Roelandts: The big issue has been that the transition to 300 mm has not been very profitable for the equipment industry. That's partly because it took a long time to roll out the technology and partly because each time you move to bigger wafers you need less manufacturing, so you are kind of digging your own grave.
Now there are rumblings about 450-mm wafers. I saw some statistics-- I don't know if they are true--that with 450-mm wafers at 32 nm, you only need seven fabs to produce all the transistors used in the world today. If you are Applied Materials, that's not much of a market. The equipment industry can't afford to go in this direction with so few players.