A node too far?
08/01/2005
Debates about semiconductor engineering often focus on the physical limitations of building smaller and smaller transistors or mass-producing them on larger and larger wafers. But the real limits are more likely to be determined by economics than physics. That’s the conclusion reached by analyst Dan Hutcheson of VLSI Research. “What you find is that industries pursue a technological thread only as long as it delivers a benefit to customers,” he contends. “The technology doesn’t stop; it’s the economics that run out of steam.”
Hutcheson made this observation during the keynote presentation at The ConFab event, which I had the opportunity to attend before officially joining Solid State Technology in June. Organized by our parent company, PennWell, the event brought together top semiconductor manufacturing executives and equipment suppliers for three days of discussion and collaboration.
To illustrate his point, Hutcheson showed how economic factors put the brakes on technological development in more mature business sectors. For example, in the commercial aircraft industry, the trend for nearly a century was to build bigger airliners. With planes that could carry more passengers, the airlines could offer seats at lower prices and make higher profits. But the formula depended on filling the planes with passengers, and after Boeing built the 747, the supply of seats began outstripping demand. There were no technological limits to building larger jumbo jets. But the industry downsized to smaller planes, which allowed airlines greater flexibility to optimize capacity and improve the bottom line.
Similarly, in the automotive business, for decades the quest was to build faster cars. In fact, in the 1950s, highway designers believed that roads would need to handle vehicles traveling more than 100mph. But when cars reached these speeds, it became clear that human beings are not very good at driving that fast. So, while there were no technological boundaries to building faster cars, the auto industry scaled back and settled on safer limits for commercial vehicles.
Are semiconductor manufacturers making wafers that are one size too big, or in shrinking chip features, are they going one node too far, or trying to get there too fast? Some trends are disturbing. For one, the cost of fabs has soared with each increase in wafer size and is now at about $3 billion for a 300mm facility. At the same time, the industry’s annual revenue growth has plummeted from more than 17% for the past several decades to 6.5% per year since the mid-1990s. However, Hutcheson’s analysis shows that economies of scale are more than offsetting the costs of building 300mm-wafer fabs, although going to 450mm fabs in another 10 to 15 years may be a different story.
Another factor of more immediate concern is that the cost of R&D has been skyrocketing as chipmakers have kept pace with Moore’s Law. In fact, for the past decade, R&D spending has been growing at 12% per year, twice as fast as revenues, according to VLSI Research. If R&D costs continue to climb at this rate, they will consume nearly half of all semiconductor revenues by the year 2020.
Of course, one way to deal with rising R&D spending would be to simply cut costs, allow the pace of development to slow, and hope that the market will continue to expand, perhaps like the automotive industry was able to do over the past several decades by raising car prices, notes Hutcheson. A major difference, however, is that cars wear out and can be replaced with essentially a fixed level of technology at inflated prices, whereas semiconductors last much longer and are replaced, at fixed costs, only when enhanced power render older models obsolete.
Another key difference is that unlike planes and cars, semiconductors are multipurpose devices that with further miniaturization and innovations - such as nanotechnologies, MEMS, biosensors, optoelectronics, and the like - will allow application developers to create a host of new applications. Thus, reducing R&D would likely shrink revenues further. And that, in turn, would mean even fewer R&D resources for the next round, which could create a feedback loop that would send the industry into a tailspin.
A more effective approach would be to find ways to make R&D more efficient and productive, such as by expanding on the successes that Sematech and other consortia have achieved in reducing industry-wide R&D spending, Hutcheson suggests. Other measures would include expanding collaboration across the board - between fabs, equipment makers, and component suppliers; between system designers and process engineers; and between rival companies on noncompetitive technologies.
Making R&D efficiency a top priority is critical to keeping the industry moving forward at the speed of Moore’s Law and ensuring that application developers are limited only by their imaginations, not by economics or the technology available to them.
 |
Phil LoPiccolo
Editor-in-Chief