"Fab-tight" is great wording. SEMI's World Fab Forecast talked about exactly that end of November. See article: "Fab Capacity Back in the Black - uncertain outlook for New Fab Construction" http://www.semi.org/en/MarketInfo/ctr_042104
I agree with Peter, at best there are maximum 7 foundries that can afford a new fab. With so much risk and such a high cost why enter that market? I bet that 7 will become 5 or maybe 3 in 10 years out. Look who builds airplanes: Boeing and Airbus. The rest is just small planes, niche markets. Semiconductor processing will be the same eventually...Kris
But not many companies have the process R&D to get into the foundry market at the leading edge. As you highlight besides TSMC there is Samsung, GlobalFoundries, then there is Intel.
The only other companies i see able to put down a shell are Toshiba, Hynix and maybe Micron who are working on the memory front.
And that's it. Seven 'active' large-scale chip companies. The rest are milking the fabs and processes they have already developed.
If Penn is right, such a fab-tight era would be a boon for new foundries eager for market share...
Judging from their capex plans, i bet Samsung and GlobalFoundries are doing their best to supply customers who can't get their sweets from TSMC...
It is all about economy. Each company will have to honestly calculate their cost of wafer price from internal fab vs. external foundries. Product lines that can get cheaper from self-operated in-house fab line should continue to do so. All others should seek a more economical wafer source which is foundry.
Key word here is "honest cost calculation" which needs to include good and bad times when the internal fab is under utilized.
I disagree with the hypothesis that fab-lite is just a transition to fabless. Companies that make a variety of products that use different technologies have diverse fab needs. The analog mixed-signal group, for example, is always going to be using larger geometries than the group that makes processors & SoCs.
As a business trying to maximize your return on capital, which fabs do you invest in? The ones that build your mixed-signal products? The ones that require the latest process node? The ones that build the products with the most sales? Maybe instead you invest in those that build products that cannot be easily moved to a foundry, and leverage the foundries to handle upsides in demand for those products that the foundry is able to build for you.
Just because you have a diverse supply chain and use a mix of internal and external fabs does not mean you are in a transition to fabless.
This is not surprising at all to me. The cost of a new fab is more than most companies can bear and it does not make sense for the few that can afford it. Moving your wafer production to foundaries just makes sense. The problem is going to be, who gets priority when the fabs are full? The biggest customers will. The small startups and fabless companies will be at the mercy of the foundaries. But then, they always were.
Blog Doing Math in FPGAs Tom Burke 5 comments For a recent project, I explored doing "real" (that is, non-integer) math on a Spartan 3 FPGA. FPGAs, by their nature, do integer math. That is, there's no floating-point ...