SoC design is increasingly a viable alternative to the purchase of standard-product ICs for teams developing complex systems. Data from diverse sources seems to attest to that. Designs that once would have been undertaken only by the most advanced semiconductor houses or one of a handful of major ASIC vendors are now being done in startup fabless chip companies or inside the systems OEMs themselves.
But before we take it for granted that this is another of those technological paradigm shifts, it might be a good idea to take a second look. The apparent migration toward SoC design-and in particular, design using customer-owned tooling flows-may be in part just another distortion left over from the Web bubble.
During the dumb old days, venture funds were throwing money at anything that looked like a networking infrastructure startup. There was supposed to be infinite demand for bandwidth of anything-wireline or wireless-carrying Web traffic, or even hosting the intranet traffic necessary to support Web transactions. If one guy had an idea for a chip to do wire-speed IP switching, why not fund 20 chip developments? How about a dozen startups doing SoCs for digital cameras that could be used to put images on Web sites?
The problem with this is that SoC designs, as they are currently conducted, require a huge front-end investment. It makes sense only if you have a high probability of large returns. In the absence of divine guidance, every startup that enters an application area reduces that probability and so, eventually, none of the investments makes sense. If, in today's colder climate, investors decline to fund such startups, we will see a rapid drop in the number of SoC design starts. In fact, we are seeing such reluctance. And this may well reduce the number of startups and systems houses that undertake an SoC design. That's not great news for the EDA or foundry businesses.
The good news is that this drop will translate into increasing interest in alternatives to SoC design. One alternative is using more off-the-shelf parts, limiting the customer-specific content to something that can be done in a modest ASIC or an FPGA. Another alternative is to get serious about platform-based design: using off-the-shelf, configurable but application-specific standard products and differentiating with software. The platform design enthusiasts have been arguing their technical case for several years now, with questionable results. But the venture community may sit in judgment on the issue and find in their favor.
RSS
RSS| 
Digital| 
Mobile
