While AMD and IBM, amongst others, have been actively developing and - of course - using SOI for more than a decade, Intel has been adamant in its refusal to utilise SOI for mainstream chips - processorsr and chipsets; however, they very recently demonstrated SOI chips used for silicon photonics. They also had no plans on adopting SOI for their mainstream.
Yeah, but why?
The main lesson here is that underdogs must follow the trends of the market. Not that SOI technology is inferior. AMD has been using SOI whilst they had their own fabs, hadn't bought ATi and weren't making any very high performance chips in bulk silicon.
Now, they not only have ATi on board, they're also producing integrated CPU-GPU hybrids and on top of that, they've sold (well, spun-off) their fabs and are proceeding to distance themselves from any one foundry.
They didn't dump SOI because of some inherent technology inferiority, but because they can't find fabs to make chips in SOI. Right now, AMD is using both GlobalFoundries - their former fab spin-off - and TSMC, the biggest independent foundry right now, which does not have SOI facilities.
For obvious reasons, they can't afford to be designing and producing highly complex chips (i.e. CPUs or GPUs) in both bulk and SOI, simply because it's double the design time, double the debugging time and double the cost. They have to select one process and stick with that.
The choice of bulk silicon over SOI was, in one more way, unavoidable. What is now called AMD's graphics department, i.e. the former ATi division, had always been "fabless" and as such, had no expertise designing on SOI. Since the now unified company has to adhere to a single fabrication technology, the choice was between either switching the graphics division to SOI, or the CPU division to bulk silicon. Since right now, the graphics division is the one making headlines - or should I say *still* making headlines - for AMD, losing precious time switching them to SOI would be a mistake.
Moreover, SOI may be profitable for high-end products, where there's a sizable profit margin, but it's not exactly the money-maker, where budget designs are concerned. Right now, AMD's CPU division is 'trapped' in the "below high-end" market layers, but also, their graphics division may be doing extremely well on the high-end layer of the GPU market, the main bulk of its sales, however, reside in the same mid-range and below layers. Put simply, the SOI benefit would become a financial burden.
Fabs, on the other hand, don't invest in SOI because smaller outfits don't design for SOI, so they can't get clients to fill up their capacity. Only GloFo already has that technology and huge experience (as far as major foundries are concerned), but their only clients are AMD and outsourcing IBM.
So, is SOI dead? Actually, in the course of time, SOI has proven itself to be very interesting technology. Its merits are indisputable and there is reason to believe that, in some way, we haven't seen the last of that technology. Abandoning SOI for AMD is a strategic move, not a concession of defeat. Maybe when silicon photonics become prevalent, or maybe if (when) the bulk silicon technology hits an evolutionary "wall", foundries will turn to SOI to harvest the last possible shreds of performance and optimisation that that technology has to offer. For the time being, it'll be gradually limited to very specific audiences, much like the older Silicon-On-Sapphire, germanium or gallium arsenide and the very new graphene, molybdenite and black diamond substrates.
