Fabless Semiconductor companies: is intellectual property reuse creating a new model?

Over the last few months several people have suggested that a new business model is emerging for fabless semiconductor companies. (See, for instance, this article on Qualcomm or the comments at the very end of this one.) The subject came up again yesterday in an interview with design services and IP provider S3 (whose site is here.)

To being with S3 is a hard company to categorize. From a purely product point of view, they provide attractive mixed-signal IP. But they also provide physical design services, design management services, and can take a set of requirements for an SoC and run with them from the beginning of a project.

It is this latter—and growing—phase of their business that led marketing director for systems ICs Bob Tait to speculate about the future of the fabless model. In a way, fabless chip companies are a vanishing breed—not because they are going away, but because they are divesting what used to be their defining expertise, chip design.

“Clients using our physical design services increasingly require our project management capability as well,” Tait says. Presumably this reflects the growing difficulty of separating decisions made early in the design from their consequences during layout and extraction. It may also reflect the growing difficulty of just getting into the physical design phase with a data set that can yield a working chip.

But Tait went on to say that clients are also increasingly relying upon services firms to do IP integration and system verification for them. They aren’t spending their time on designing commodity RTL blocks, or even on selecting them from IP vendors. “We see less and less work going into RTL design these days,” Tait says, “but more and more work going into IP reuse—integration and verification of somebody else’s blocks.”

The endpoint of this trend, he speculates, is that a fabless firm will comprise only three things: a good idea for an application, a piece of key intellectual property that will differentiate a chip in that application, and a key customer relationship to help pay the bills and take the chip into the end market. All the rest—understanding the system-level design, deriving the SoC requirements, gathering the third-party IP and integrating the chip, and most vitally creating the software and reference design—is non-value-add from the fabless vendor’s point of view, and not worth hiring for.

The fabless semiconductor company, then, becomes just another stage in the pipeline, no longer containing any expertise not directly related to their bright idea. Maybe we need a new name other than “semiconductor” company. Any suggestions?