Practical Chip Design

A panel discussion sponsored by a global technology interest group, Monte Jade, explored the future of the semiconductor industry last week, by looking into the question of industry maturity. Panelists addressed three aspects of the maturity of the semiconductor industry: technology, structure, and finance.

On the question of whether the semiconductor industry was approaching technological maturity, Intel fellow Shekhar Borkar stated what turned out to be the majority view. "Semiconductor technology is far from mature," Borkar, director of the circuit research lab at Intel, said. "We are barely starting to see quantum effects in design. This technology has ten years to go."

John Chen, vice president of technology and foundry operations at nVidia, was more circumspect, adding "I agree that Moore's Law will continue for five to ten more years. The thing that will change is that we won't get there by pure scaling any longer. But the underlying physics still has a ways to go."

But Handel Jones, president of research firm International Business Strategies, offered a note of caution. "I have to disagree," Jones said. "Designers are seeing major problems with leakage at 45 nm, and at 32 it will be worse.

"From here on, the cost equation changes as well. According to our models, the cost per gate will only drop about 5 percent between 32 and 22 nm, and by 18 nm, it will actually have begun to rise again. Memory becomes problematic even faster: by 22 nm it no longer pays to have large memory structures on-chip. Further, interconnect is not scaling with device critical dimensions—we will have to have new architectures to even employ these advanced processes."

Will these forces—a roadmap, but increasing difficulties and decreasing productivity returns—force a restructuring in the industry? The panelists examined whether the semiconductor business was maturing into a typical stable industry dominated by a few giants with their networks of subcontractors, or if it was still in flux.

"In one respect, there are signs of maturity," Jones said. "Year-on-year profits are declining, and we are seeing consolidation. But in another respect, this is not behaving like a mature industry at all: R/D is rising as a percent of revenue—some companies are spending over 30 percent—and we see huge untapped market potential in China. Even market share is still dynamic: Taiwan is taking share, primarily from Japan."

Di Ma, vice president of field technical support at TSMC, also suggested that classical maturity really didn't describe what was happening to the structure of the industry. "There is consolidation," he said, "and VC investment levels are holding flat. But there are also many new ideas bubbling up. And we aren't really seeing consolidation so much as continued disaggregation—companies in the industry are clumping together by collaborating, not by merging or acquiring."

Chen argued that some consolidation is now inevitable, and that in fact we've reached the point where the industry probably won't see much else in the way of structural change in the next five years. But he warned the audience not to mistake growing structural maturity for impending death. "The fact that will sustain the industry is that there is no viable replacement for semiconductors on the horizon," he said.

On the question of financial maturity, the panelists again had different points of view. "Right now, the cost per transistor continues down," Borkar argued. "As long as we continue to deliver more value per transistor, we will have growth."

Chen admitted that many application areas were now past their period of rapid growth. But looking to the future, he said, meant asking who really needs all the transistors we will be producing. He suggested that value in the future—and hence, high returns—might come from innovations on the systems-design side, as architects learned to use enormous transistor budgets, rather than from the process and implementation sides of the house. "There is still very high growth to be had in the medical area, for instance," he said.

But Chen said it was vital for semiconductor companies to move up the value chain and claim more of the revenue from the value of the systems. "We have to shift the distribution of wealth," he said. "We worry about the cost per transistor leveling off—but we don't have to have lower costs to have growth. Look at a $600 consumer GPS. There's $5 worth of silicon in it. The semiconductor industry has been too complacent about inventing whole new applications, and being creative in what they are offering to users. IC vendors have to move up into the systems space." Jones added later to this point by saying that there was a major area of innovation open to semiconductor companies today: software, which in the past IC vendors have thought of as a support cost.

Then Jones went on to observe that finance and innovation are redrawing the global semiconductor map. Today in the NAND business, for example, margins are so poor and capital requirements so huge that the market may belong to the two or three companies with the most willing investors—and today that will mean sovereign wealth funds and government direct investment more than healthy capital markets.

"China will have a memory business in the future," Jones said. "Maybe not next year, but within the next five years. Middle-East money may fund a foundry as well. But just having the capital will not be enough. It's hard to buy your way into the semiconductor industry today without a pool of engineering skills and a strong local market. China has both, but the Middle East may not." Ma agreed: "The bottleneck in the industry is still engineering talent. That's why you don't have to worry so much about outsourcing."

Asked by the audience to give their views on 450 mm wafers, the panelists again were mixed in their outlook. Ma said that the wafer-size decision would be based on the needs of key customers, period. But Intel's Borkar observed that it all came down to cost. "You get twice the number of dies from a wafer. Could you save building a fab by going to 450 mm? That's the question."

Jones saw the 450 mm question in the context of industry consolidation. "According to our cost models, by 22 or maybe 18 nm, you will have to be on 450 mm wafers in order to have adequate costs. But that will be a very exclusive club: Intel, certainly, and maybe Toshiba or possibly ST Microelectronics. I don't think we will see any foundry except TSMC make the move."

So in effect, the panelists saw the pressures that lead to maturity growing around the semiconductor industry. But continued technology headroom, the industry's ability to substitute collaboration for consolidation among its mid-level and small companies, and the possibility of major shifts as huge new markets develop and enormous new sources of capital appear could make this a very dynamic business despite the pressures.