SANTA CLARA, Calif. - As if a lousy business climate weren't bad enough, there's more to worry about for semiconductor manufacturers.
"The simple fact is that we won't be needing so many chip vendors in the future," Dataquest analyst Jim Tully said this week. "Many firms that are now fabless chip houses should seriously consider becoming IP intellectual-property vendors instead."
As for that next "killer app" the industry is holding its breath for, Tully says forget about it.
"We are always asking, 'What will be the next killer application?' " he said. "The truth is there will be no more killer apps. There will just be a growing network of hardware resources onto which increasingly compelling software applications can be loaded."
In a presentation, Tully based his less-than-reassuring outlook on a model of the end markets for electronic equipment, the needs of those markets and the rapidly changing technical environment in which all IC vendors-fabless or not-find themselves.
"We used to live in an industry that was driven by the personal computer," Tully observed. "We saw that change into an industry driven by the dot-com explosion. Now we are in the post-PC, post-dot-com era." Today, Tully said, systems-on-chip are consumed primarily by wireless handsets and consumer entertainment devices, with networking and storage applications bringing up a distant third. The fact that the sockets are essentially in consumer markets, he said, will cause profound changes.
"Consumer markets are fast and fickle," Tully warned.
Technical issues are also changing. The increasing integration of SoC products means that fewer and fewer chips are going into a system. And the spiraling cost of a design-both in nonrecurring expenses such as mask costs and in terms of design complexity and risk-are putting SoC design beyond the reach of more and more companies.
This has led, Tully claimed, to a decrease in the total number of chip design starts. "Starts peaked in 1997, and we predict that the decline will continue indefinitely."
So instead of SoC technology spreading across the industry, Tully painted an industry restructuring-and contracting-around a handful of vendors that could still afford to complete SoC designs. Many of today's fabless companies, he suggested, will continue to serve the industry, but as chipless IP providers rather than as semiconductor vendors.
Those that continue to produce silicon will increasingly design standardized platform chips, capable of serving a range of applications via reconfigurable hardware and software, thus spreading the multimillion-dollar cost of a design across a range of applications.
This reconfigurability also played a second role in Tully's view of the future. Increasingly, he said, the consumer markets that take SoC devices will be dominated by services, not hardware sales. New content, new features and functions, would represent a larger portion of the revenue than the hardware, until the SoC-based system simply becomes a delivery vehicle for revenue-producing services. OEMs, which now own most of the teeth in the food chain, will increasingly be squeezed between IC platform vendors that demand better margins, and content and service providers who demand their pound as well.
This scenario places a premium on the flexibility and extensibility of the hardware platform. And it discourages system architects from locking differential advantages into hardware. Hence, in Tully's view, the industry will gradually swing away from its tradition of starting a new SoC design for each new application, instead adapting platform chips to cover new opportunities.
The analyst suggested that the ideal SoC of the future would be a computing and networking platform, capable of adapting to a range of specific applications through reconfigurability. It could, on one side, configure itself for any of a wide range of communications environments and protocols, and on the other side would have sufficient headroom for a long line of enhanced user features.
Show me the money
Though not a view of the future that beleaguered semiconductor executives were eager to hear, Tully's prognosis correlated with their anecdotal experience. One manager at a wireless SoC startup who wished not to be identified spoke of the increasing difficulty of getting B-round funding. "Last year a lot of companies were able to get a first round. But now they all have designs that are just about done, and they need more money to tape out and produce chips. The money just isn't there unless you can show the VC three or four live customers.
"So there's a growing pool of companies almost ready to tape out, but trying to cut spending while they find enough customers to get another round," the manager related. "It's particularly bad in the network processor area, where there are whole designs that may never tape out."
Nonetheless, Tully's analysis left open some important questions. First, is the trend toward fewer design starts and fewer tapeouts indeed a structural phenomenon, or is it an artifact of the present popped-balloon market? Once another idea shows apparent growth potential, will investors rush to fund hundreds of indistinguishable startups once again, reigniting growth in SoC design starts?
A second question involves market life cycles. The scenario Tully describes is well-known for maturing markets. Growth slows, reducing investment. Differentiation becomes harder, and produces less revenue growth. Standardized computing platforms become capable of supporting most or all of the features that actually add market value, so the importance of vendor-specific hardware declines. And there is a shakeout among vendors.
Is Tully simply documenting the maturity of the existing SoC markets, then? Or will another set of consumer appetites appear that once again justifies the rush to build specialized hardware?
If the change is indeed structural, there is in fact a train coming down the tunnel. But if there is a next big thing out there that lies beyond the reach of existing platform architectures-even in advanced processes-that glimmer in the dark might be daylight after all.