The semiconductor industry likes a well-understood driver. The method of innovation involves three steps. First, find a potentially huge market in which the infrastructure development are already under way. Second, define requirements-often under the unwelcome leadership of one industry-dominating player. Finally, splitting into the smallest possible units-usually startups with great inventive talent, limited financial resources and no concept of marketing-innovate within the narrow boundaries set in step two, and then struggle to sell product to established OEMs. This pattern has been repeated for the aerospace market, the PC market and the networking bubble. In each case, the formula worked well. But none of those juicy opportunities are around today.

So the industry has thundered off in another direction, this time trying to stampede the consumer electronics industry into a box canyon. Consumer is a different beast, however. Aside from the obvious issues of zero margins and infinitesimal product lifetimes, there is one huge but ill-considered difference: There are no de facto standards.

That would not be the case except for the level of integration at which the semiconductor industry wants to design today. If it were still making LSI functional blocks on chips, there would be plenty of standards. But we aren't building little functional blocks. To have margins, the semiconductor industry needs to build systems-on-chip. And at the system level, each new consumer product is different.

Poorly organized
When semiconductor designers create an SoC, they are defining an end-market consumer product. But the U.S. semiconductor industry could not-with its myriad startups, almost total lack of vertically integrated companies and virtually nonexistent relationships with the real consumer electronics giants-be more poorly organized to define end products. Not only do chip designers have no idea what end consumers think, they also lack the marketing tools needed to turn a technically proficient product into a retail hit.

There are only a few ways around this problem. One is to cozy up to real consumer marketing companies. The relationship between PortalPlayer and Apple is a case in point, as is that between IBM-which achieved architectural and 90-nanometer magic with its Cell processor design-and Sony, which, unlike IBM, can actually sell boxes that use the thing.

Another idea would be for semiconductor companies to stay at the building-block level, keeping their prices low enough and their reference designs simple enough that small and midsize consumer OEMs could compete using multichip implementations. This would require a different approach to engineering-cost engineering, assumption of stacked packaging and other ideas to which the industry is not accustomed, along with intensive marketing on the ground in Asia, where the OEMs are. But for most semiconductor startups, it may be the best alternative.

Ron Wilson (rwilson@cmp.com) is semiconductors editor for EE Times