@michigan. the Leti figure you're citing for FD-SOI Si thickness is *post-processing". The current and next-gen wafers from Soitec et al meet all the requirements. This was explained very clearly by Bruce Doris of IBM a couple years ago, who said (and it is still true), "The top silicon of the starting wafer (currently 12nm) has to be thicker than the final target thickness of the channel since some Si is used up in the process flow prior to final channel thickness definition. " http://www.advancedsubstratenews.com/2010/07/etsoi-substrates-what-we-need/ A lot has happened since you posted this comment -- I think you may find that the battle has just begun...
It sound like a battle between planer FD SOI by new IBM alliances versus FD Trigate FinFETs by Intel has been declared. The key word here is FD (fully depleted) because both 22-nm transistor for IBM and Intel must be fully depleted to suppress the leakage current or short channel effects. Recently, Leti at the 2011 SOI conference has showed that ultra-thin/un-doped 6nm Si film is required for 22/20nm FDSOI to suppress the leakage current. For the 14nm FDSOI an extremely thin 4nm or less Si film may be required. However, Soitec canít deliver such an ultrathin 6/4nm film in manufacturing. What Soitec can deliver is 12nm Si film and 25nm BOX for 22/20nm nodes. IBM and its alliance members also published FD SOI with 6.5/7 nm Si films at 2012 VLSI Symposium, but these are test chip data, not manufacture-able by Soitec.
Meanwhile, for FD finFETs the fin width (W) less than Lg (gate length) or W Lg is only required to suppress the leakage current. It means for 22nm node the fin W can have 21nm or less versus 6nm for FDSOI, and for 14nm node the fin W of 13nm or less versus 4nm for FDSOI. This is enormous advantage for manufacturability of 22/14nm FD finFETs compared with FD SOI. Furthermore, the trigate fins can be doped to adjust Vt, and manufacturability of the tri-gate fins dictates the device scaling. That is why Intel FD FinFETs is in high Volume manufacturing for several months, but FD SOI is not and will not be. The battle is over! S kim
A consortium for better mobile processor makes total sense. One of the many keys solution of reducing power consumption is by achieving complete integration. There may be more than 1 consortia formed due to competitive and political reasons. If it is acceptable to the industry, 1 is always better than multiple. Nonetheless, the industry and market will drive the direction.
An interesting comment from Moore,
"the current infrastructure works out to roughly one server to deliver content to 600 smartphones"
I believe it depends on the kind of server and memory capacity. In addition, the type of services will affect the number of smartphones, or in general, mobile devices a server can deal with. To my experience, a dual Xeon with 128GB of memory shall be able to deal with thousands of simultaneous connections. If Moore is talking about ARM based servers, I have no doubt improvement is necessary.
"While Intelís FinFET approach is not expected to enter production until the chip giant rolls out its 14-nm process technology" Isn't it true that Ivy Bridge is using 22nm Tri-Gate, and already in market?
"IBM lost relevance as a process technology pace-setter at least a decade ago."
10 years ago IBM screwed up big time pushing SOG (SiLK)in conjunction with copper interconnect.
UMC licensed the technology from IBM but dropped it because they figured out SiLK was not working.
And now gate first debacle - it's good for one node...
Of course Intel looked at SOI...and Bohr even agrees that some will use SOI but based on Intel's evaluation they rejected it.
4/16/2012 7:29 PM EDT
Mark Bohr addresses SOI versus bulk
One on One with M. Bohr
Since posting this comment, it's been pointed out to me that many people (JP Colinge, D.Flandre, researchers at Hitachi and many more...) did seminal work on fully-depleted/thin body transistors leveraging SOI. We think of this as new, but the industry's been quietly working on this for over 25 years!
Agree, This article talks about FDSOI as if Intel never compared this to FinFet or any other process technology options. FDSOI or SOI in general was known to semiconductor for long and its benefits and downsides are well documented. In fact if you ruled out the price/cost factor of SOI, there are other technologies which can deliver much better performance/power than SOI.
And you think the IBM alliance is the answer? IBM lost relevance as a process technology pace-setter at least a decade ago. Mobile chips tend to have slim margins, and the substrates required for FD-SOI are expensive, adding the equivalent of at least one metal layer to the overall price of a chip. I can see why Soitec is pushing this, but I am not aware of a single demonstrated benefit of FD-SOI, and several well documented downsides (self-heating, floating body effects).
Blog Doing Math in FPGAs Tom Burke 18 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 ...