I heard conflicting messages about this Kris (finFET vs FDSOI). Some say FDSOI controls leakage better than fitFET and is much easier to design. As such it will be the technology to succeed. Others say finFET is already the winner. I guess the proof of the pudding is in the eating....
Yes, they are very different technologies...finFET exploits silicon surface in a third dimension so you would expect every good Ion current but probably some difficulties in controlling Ioff...FDSoi takes advantage of thin silicon body so Ioff leakage is-likely easier to control but Ion values are not as good....so I think finFET will be better for high speed and FDSOI better for very low power...Kris
I guess it is possible that at some point finFET would be build on Soi substrate, personally I don,t think that would happen...looks too complicated...there is no reasons why the two consortiums would not continue developing what they put massive r&d investments already in
Kris, as of course as you know, FinFETs were originally developed on SOI - because it was easier! And there's that great animation on the SOI Consortium website that shows why that's still the case. But is easier necessarily in the interests of the foundries? If bulk FinFETs are more complicated but they get higher margins on them, it seems logical that's what they'll push...? Whereas for IDMs, they reap the savings (cheaper and easier) themselves. Not sure how good this trend would be for the industry, tho. On the other hand, we're starting to see some complaints from designers re: pain points in bulk FinFETS. Either way, looking forward to this IEDM for sure!
thank you Adele...no, I didn't realize that finFET was built on SOI substrate first!...since doing my research on SiGe devices while with U of Toronto in early 90-ties I have not been following basic transistor technology that closely...looking to hear from you on how finFET vs FDSOI debate continues at IEDM!...Kris
Changing technology at 10nm will be too late. FinFET introduction has not given the advantage the companies thought it would due to complexity and litho limitation among many other things but i guess until someone else prove that FDSOI is better then companies will reconsider.
I think your correct. FinFET sounded good but all the 3D process variation and high gate capacitance means many design blocks on the SOC have worse PPA. I think this is one of the reasons Apples 28nm A7 is much better than Intel's FinFET Bay Trail
re: self-heating -- the paper summary indicates that they're looking at 6nm (!!) FD-SOI, and finding self-heating "comparable" to bulk FinFETs. So maybe not going away, but certainly not a show stopper!
Here are some important facts about FDSOI technology. FDSOI was invented by IBM over a decade ago, but still not manufacturable at any technology node yet despite of enormous efforts and resouces were spent by IBM research and its International SOI consortium mainly because Soitec, the largest SOI wafer supplier can't deliver 7 nm thin SOI that is required for manufacturing of 28 nm FDSOI. AT 2011 SOI Conference at Phoenix, AZ Soitec announced that what it could deliver is 12 nm thin SOI wafer, not 7 nm SOI. For FDSOI at 20/22 nm an extremely thin 5 nm SOI is required to overcome the short channel effects or transistor leakage current. That is why FDSOI still not manufacturable even at 28nm today, and will not be manufacturable at 20/22 nm and beyond. FinFET is only technology in volume manufacturing for a couple of years at 22 nm by Intel, and 14 nm in late this year. TSMC appears to introduce its FinFET manufacturing in 2014. Beauty of FinFET is that it is extendable to the end of scaling according to FinFET physics. FDSOI can't contest with FinFET. There is alternative to FinFET today. Skim
Michigan0, the reason SOITEC delivers 12nm SOI wafer is not because they cannot deliver thinner wafers. It is simply because they are asked to do so. Any person familar with CMOS technology recongnizes that you need to consume a few nanometer to form STI (pad oxide needed before deposited pad nitride), then you need a few nanometer oxide for your I/O devices. Once you do the math, you realize that for a taget channel thickness of 6-7 nanometer you have to start somewhat thicker and this is exactly what has been asked from SOITEC and SEH. Unless you want to use deposited oxide for pad-ox and I/O devices (which is of course inferior to thermal oxide) this is what you'd need. Again, anybody that processed CMOS wafers knows that thermal oxidation is precisely controlled -- for our reference gate oxide was about 1nm thick with less than 5% variation before people switched to high-k.
As far as thickness control goes, in fact FDSOI has significant advantage over FinFET. The device is planar, which means you have a variety of well established methods to monitor thickness on as-received wafer and during processing, including ellipsometry and AFM. Metrology is a big problem with FinFET. Let alone the loading effects in depositing the spacer in SIT process and in etching the fins. Yes, FinFET is in mass production but I can say with enough confidence that it did NOT deliver the promissed 50% reduction in power that was claimed back in 2011 even after supposedly toc of Haswell.
michigan, so are you saying that despite much time there has not been enough progress in FinFET and this has led to lower number of papers in IEDM? But no other company apart from TSMC is presenting paper which means more companies are developing FDSOI. Don't u think.
My primary purpose was not to lower or higher number of FinFET or FDSOI papers in IEDM. FinFET is at high volume manufacturing now over 2 years at 22nm by Intel while FDSOI is not manufacturable at any technology node yet. Therefore, when FDSOI test chips become available, I propose the minimum requirements for measurement of the transistor electrical transfer characteristcs for both FinFET and FDSOI for fair comprison. The minimum requirements are posted in detail. SKim
note intel 14 Finfet now delayed. delay is more like a year for the volumn Broadwell release not the 3 months CEO stated. Broadwell will not be in volumn in market until end of 2014 and there is even talk Broadwell will be cancelled due to it being so late or a limited volumn release. my contact a copy manufacturing engineer says they are still not even involved in the manufacturing and he should of been involved 5 quarters ago.
A Book For All Reasons Bernard Cole1 Comment Robert Oshana's recent book "Software Engineering for Embedded Systems (Newnes/Elsevier)," written and edited with Mark Kraeling, is a 'book for all reasons.' At almost 1,200 pages, it ...