Almost all high-tech companies go through cycles of heavy investment. But no sector matches the semiconductor industry when it comes to placing humongous bets. Every few years, major players invest in new fabrication plants, or "fabs"--the latest of which can cost upwards of $2 billion. But in return, the sector shows extraordinary sustained growth. Chip industry revenues have grown by an annual average of more than 15% since the early 1970s--despite cyclical contractions as severe as 20% in a bad year. And few businesses constantly overhaul products of such exquisite complexity, over cycles that are measured in months, not years.
What's most surprising is that people no longer find any of this surprising. So says Bob Helms, president of International Sematech, a 13-member research consortium that includes the world's largest semiconductor makers. Helms practically grew up with the industry, starting out as an electrical engineering student at Stanford University in the 1970s and eventually running Texas Instruments' (TXN) advanced research and development group, before taking the reins at Sematech earlier this year. In the past three decades, Helms has had a hand in a parade of innovations that--timed to the drumbeat of Moore's law--have halved the cost and doubled the power of chips on a steady 18-month schedule and given rise to a $290 billion industry.
These days, Helms points out, it's not just technology and investment that are transforming the chip business. The sector is evolving into a more complex ecosystem made up of a shrinking number of big players and complemented by a growing web of specialists that--increasingly--are located in Asia. Helms recently outlined his take on the industry's future to Industries Editor Adam Aston.
Q: How has the chip industry changed over the past decade?
A: Ten years ago, the integrated device manufacturers--or IDMs--such as Intel (INTC), Philips (PHGZF), and Texas Instruments--were even more integrated than they are today. TI, for example, made its own chip-manufacturing equipment. Others did everything from making materials--like silicon wafers--all the way up to putting together systems in a box.
These days you have what some people are calling dis-integration, or disintermediation, going on. The IDMs have steadily refined what they do well and grown more specialized. So now they mostly rely on equipment suppliers such as Applied Materials (AMAT) for their machines. And so-called fabless companies, such as Xylinx, do just chip design. They depend on foundries like TSMC [Taiwan Semiconductor Manufacturing Corp.] (TSM) to make the chips.
Q: What does this trend mean for the IDMs?
A: It's popular to say the IDM is a thing of the past. But that's not the case. Sure, there's a strong argument for specialists like TSMC. But many chip buyers, especially big buyers, understand that only the integrated vendors can reliably marry design and manufacturing under one roof. For the big IDMs, their scale is a competitive advantage. Their customers will continue to want that one-stop shopping: Here's the design, the packaging, guaranteed fab capacity, and top-notch technology all rolled up into one package. Yes, foundries accounted for 10% to 15% of industry revenue last year, and they'll have more capacity in 10 years' time. But the IDMs will continue to control most of the market share. In fact, the costs to research, develop, and install these advanced technologies will undoubtedly drive further consolidation in the semiconductor and equipment-suppliers segments.
Q: How does this affect the development of leading-edge chip-fabrication technology?
A: The technology cycle is now about two years. By this I mean the time between major technology shifts, like the move from 200-millimeter silicon wafers to 300-millimeter. And there are significant incremental advancements in chip technology in cycles of 12 months or less. Only the IDMs have the revenue streams to support the R&D investments--running into the billions--necessary to develop state-of-the-art technology and to put it into plants.
The one exception is TSMC, the world's largest pure foundry--it's driving to get near the leading edge. And the other foundries are trying, too. But you probably have to spend $500 million per year in R&D and in manufacturing process technology to be able to get close to the leading edge. And that doesn't even count the capital expense of buying the new equipment.
Q: In addition to TSMC, will we soon see other new players coming out of Taiwan or China?
A: There's a push on in China to jump-start the semiconductor sector. Fabs are being planned and built, especially around Shanghai. It makes sense in some ways: China is a big and growing end-user market.
Q: A series of new Asian players--first Japan, then South Korea and Taiwan--have been able to carve out niches for their semiconductor industries. With its market, skills, and ambitions, isn't China the next contender?
A: China will be a bigger player as time goes on. But look at what drives this business: End customers don't need to be near the point of manufacturing. Because chips are small and light, you can ship them around the world at a relatively low cost. And look where value is added in this business. Increasingly, it moves into the design space. The IDMs are the leaders here. The ones that can afford to keep their R&D investment up will continue to marry good design and silicon technology very fast. China can come up to some nominal market share but it doesn't concern me as a major risk to the semiconductor world as we know it today. That said, some of my senior colleagues are more paranoid than I am.
Q: So the industry is evolving, to some degree, toward more specialized players, and capacity is spreading into new markets. How does the ownership of intellectual property--the design of chips and how to build them--come into play?
A: If you look at the different approaches to making chips--within a foundry segment, or at the IDMs, or a fabless design company--each will have a different way to look at their most important intellectual property. The semiconductor business of 10 years ago was certainly simpler, and IP was valued differently. It's more complex today.
Consider just one extremely important aspect of the business: the accelerating product cycle. For companies that are making the R&D investment to produce new chip lines every two years, a very few months can be the difference between a company that hits the market and captures share and those that fall behind. IP is critical, but it's generated and used in real time. The biggest margins are in advanced technology, and the value of IP decays rapidly as you go from one tech cycle to the next. It used to be the big eat the small, but today the fast run over the slow.
Q: How can companies squeeze the most return from their R&D then?
A: Ten years ago, a company like IBM (IBM) might have kept a major R&D advancement close to its vest. Perhaps they would have developed a captive equipment supplier base and tried to reap all the benefits exclusively. Today, you're more likely to see a company develop a new technology first, then drive it to become an industry standard. That means licensing it out to competitors. You want your technology to become mainstream as fast as possible so that you can, in effect, share development costs.
Q: A final question: The chip industry is in its worst contraction ever, yet you seem optimistic. Do you know something we don't?
A: It's hard to judge when the upturn will come. There are just too many questions: the dot-com crash, the recession, and the horrific events of 9-11. I can say that it's the fastest downturn in history. Last year, about $206 billion worth of chips were sold. Even then, there was a feeling that there has to be a slowdown soon. For this year, I've seen projections as low as $160 billion and below. Still, this industry has grown by an average of 15% to 17% per year for the past 30 years. We're not finished. In some sectors, such as wireless, there's already inventory burn-off. How fast it picks up depends partly on consumers. But we'll be growing in the mid-teens soon.