I already took a cursory look at the fact that the semiconductor industry is going to restructure, partially driven by the current economic downturn but mainly by the fact that almost all semiconductor companies are going to become completely fabless. What we have got used to calling IDMs (integrated device manufacturers) are just going to be large semiconductor companies that used to have fabs. This trend is driven by two things: the economic size of fab has got so large that it exceeds most semiconductor companies’ needs; and the cost of process development has got too high for any single semiconductor company (except Intel, TSMC and some of the DRAM guys) to be able to afford it.
Having a fab to fill means that a semiconductor company has a huge fixed cost that has to be amortized over all the wafers actually manufactured. This puts a huge premium on having the fab filled. Just like a hotel cannot inventory rooms, they are either occupied tonight or not, a fab cannot inventory wafer starts. Either a wafer was started or it was not, and a wafer not started is one that doesn’t carry its share of the overhead of depreciating and staffing the fab. So semiconductor companies have grown up to contain collections of divisions that together require all the wafers a fab can produce. If there are not enough then it is attractive to acquire further product lines.
Once a semiconductor company has no fab, then the particular collection of businesses that make it up have very little reason to be grouped into the same company. Further, it makes very little sense for a semiconductor company to pay a big premium to acquire a new product by buying a fabless semiconductor company since it no longer has a fab to fill and so doesn’t really have any economies of scale. Sometimes, as with TI and digital signal processing, there is company-wide expertise that cuts across a many products. But often not. For instance, it is interesting that TI was attempting to sell its wireless business (it gave up because it couldn’t get a good price) despite wireless having a significant DSP component.
One possible future scenario is that many of the semiconductor companies of today will disintegrate since they don’t have a lot of reason to keep product lines together. In fact the whole idea of a product line may start to be obsolete since so much of a chip is now externally sourced IP, both semiconductor IP and software libraries. I also looked at how some semiconductor companies are one-hit-wonders, with a successful chip that fills their cash position, that they then gadually burn through.
An unlikely place to look for parallels to chip design is the movie industry. Back in the middle of the last century, the studios were like IDMs. They had an entire infrastructure for making movies that had to be amortized by making lots of movies (to fill the studio, like filling the fab). Today, movies are not made like that. They are made by virtual companies that are put together expressly to make a single movie, almost everyone is a subcontractor not an employee of the movie, and the company is disbanded when the movie has been made and the profits have (or, often, have not) been distributed to the investors.
Chip design could go like that, with an individual chip being built by a team of subcontractors assembled for just that purpose and manufactured by a foundry, probably TSMC. If the chip makes a lot of money the investors get a return; otherwise not. associated with keeping a company together just because it had a hit product and no guarantee that the next product will be another hit. Better to distribute the profits and fund the next chip as a completely independent project. Every chip is a one-hit-wonder by design.
So, do you want to green-light my chip, Mr Spielberg?