Why have they not moved to SOI already?
Look at Chartered Semiconductor's revenue versus TSMC's.
That gives you an idea of the market-size of SOI in the foundry business (excluding AMD and GloFo) that TSMC has neglected to go after by not bothering with developing/offering their own SOI process tech to date.
Relative to bulk-Si there just hasn't been enough market to make it worth TSMC's effort and resources to go after it.
Outside of AMD's use of SOI it is pretty much a niche application when you actually start enumerating the number of IC's in the world that use SOI.
But GloFo can be a game changer there by providing leading-edge design rule SOI process tech to customers of TSMC...now it is the customers of TSMC (who compete with each other, like NV and AMD or Qualcomm and TI) who must decide whether or not they
need to transition to SOI in order to stay one step ahead of (or catch back up with) their competition.
TSMC doesn't need SOI until its customers start bailing (or giving indication they are going to bail), until then TSMC determines where the bar is set for the foundry industry simply due to their marketshare.
@IDC
Isn't Global Foundries also supply constrained until their New York fab gets off the ground? At least that was the impression I was getting before the AMD breakup. AMD could have sold more chips but because they didn't have the fabs to manufacture enough to meet demand. At least the demand before the wheels fell off and the Empire struck back.
There are basically two fabs in Dresden, they are connected by a central admin office area and so they call them modules instead of seperate fabs but the relevance of this info to your question is that AMD had effectively mothballed half of their fab capacity in Dresden (one of the modules) when they transitioned from 200mm to 300mm. After the 300mm transition Intel was really putting the hurt on their marketshare and gross margins so they didn't bother to retool the other module.
So Glofo has fab space for sure, even without NY. That is why the timeline for the NY fab is so "loose and long". They are taking their sweet time with that fab buildout for good reason.
I was under the impression that the design cycle for GPUs was on the order of three years. Even if this precludes R800 from being manufactured at Global Foundries, it seems some test wafers should have been produced for AMD PR to crow about if a foundry shift was imminent.
Taping out the reticles for test wafers is costly, definitely not something you do just to make a few test wafers for PR reasons. The earliest we'd see anything like this would be 32nm provided AMD was actually taping out their next-gen GPU on GloFo's 32nm process.
Even then I'd expect minimal crowing from AMD as they transition to GloFo. It would be foolish to rub TSMC's nose in bad PR, burning that bridge, when for all AMD knows they could very well find themselves wanting to use TSMC's 16nm process instead of GloFo's in a few years time for any number of reasons.
What are the drawbacks of SOI? Despite offering better power efficiency, I was under the impression SOI offers poorer heat dissipation than the standard bulk process. This might make it less suitable for high power draw parts like GPUs... on the other hand AMD might simply be being cautious considering its financial situation.
Numerous drawbacks. Substrate cost is probably one of the largest. Intel produced some rather damning white-papers and conference presentations on SOI versus FD-SOI when they were making public their justifications for not following AMD and going with SOI.
better performance/watt on average. If this is correct would they not be attractive to ARM designers for the exploding smart-phone/smart-device market? Wouldn't the extra performance and-or battery life be a legitimate competitive edge for a handset manufacturer or are the costs of entry prohibitively high for the hypothetical prospective customer? Or some other issue that I am unaware of?
Facebook
Twitter