IBM and partners show off 22nm SRAM

[epidemis]

Link: http://www.eetimes.com/news/latest/showArticle.jhtml?articleID=225200658

This tells us something on the disparities between AMD and Intel on process nodes.

SRAMs are a very basic IC feature, and a step that goes before getting logic down on the specific node.
In the past Intel has shown SRAM wafers with a new process 2 years in advance to commercial introduction. The latest wafer was shown during IDF in September 2009 on the 22nm node. That should mean they (intel) have roughly a 6-8 month head start on 22nm. However that's just a guesstimate, though we know intel plan to introduce ivy bridge (the 22nm shrink) in 2012q1, but it's not clear how fast GF could proceed from SRAM to logic.

 

[Idontcare]

The finfet sram stuff isn't for production, not at 22nm at least, just prototyping for data creation purposes. Very much like when you see press releases regarding EUV being used to make test chips years ago. 22nm will be planar cmos, HKMG.

Doesn't make the milestone any less cool or any less important, just saying this kind of milestone is not intended to be compared to the other kind of milestone you mentioned (pre-production wafer full of production-layout sram chips).

 

[Nemesis 1]

Link: http://www.eetimes.com/news/latest/showArticle.jhtml?articleID=225200658

This tells us something on the disparities between AMD and Intel on process nodes.

SRAMs are a very basic IC feature, and a step that goes before getting logic down on the specific node.
In the past Intel has shown SRAM wafers with a new process 2 years in advance to commercial introduction. The latest wafer was shown during IDF in September 2009 on the 22nm node. That should mean they (intel) have roughly a 6-8 month head start on 22nm. However that's just a guesstimate, though we know intel plan to introduce ivy bridge (the 22nm shrink) in 2012q1, but it's not clear how fast GF could proceed from SRAM to logic.

Ya Intel 32nm has me wondering abit. With the recent HPC announcements from intel , I am not real sure about Larrabee ,As a gpucpu. With the Marketing product not arriving on untill late 2011, After seeing the power consumption of the Knights ferry card and the fact we don't see Knights Corner until late 2011 on 22nm. Leads me to wonder if .

1) Intel is having problems with 32nm.

2) Intel is moving its Cadance up for the launch of a 22nm product. This kind of makes Sense for Intel to change its approach to Candance in this instance. Beings these are simpler cores to work with. Also if Intel succeeds in the HPC market and does bring a GPUCPU forward it would make sense to stagger New Arch. According to Anands article as I read it Knights Corn on 22nm in 2011 That already tells me if its correct that intel is bring 22nm ahead of Haswell

 

[Nemesis 1]

Ya Intel 32nm has me wondering abit. With the recent HPC announcements from intel , I am not real sure about Larrabee ,As a gpucpu. With the Marketing product not arriving on untill late 2011, After seeing the power consumption of the Knights ferry card and the fact we don't see Knights Corner until late 2011 on 22nm. Leads me to wonder if .

1) Intel is having problems with 32nm.

2) Intel is moving its Cadance up for the launch of a 22nm product. This kind of makes Sense for Intel to change its approach to Candance in this instance. Beings these are simpler cores to work with. Also if Intel succeeds in the HPC market and does bring a GPUCPU forward it would make sense to stagger New Arch. According to Anands article as I read it Knights Corn on 22nm in 2011 That already tells me if its correct that intel is bring 22nm ahead of Ivy Bridgel

Had to remove Haswell for Ivy Bridge

 

[Idontcare]

Nemesis I gather from your post that you are thinking Knights Corner is NOT Haswell?

 

[Nemesis 1]

Nemesis I gather from your post that you are thinking Knights Corner is NOT Haswell?

NO . I meant to say that Knights Corner according to available sources is due late 2011, Ivy Bridge Is do out on 22nm in early 2012. So it appears Intel will introduce 22nm with Knights Corner. Forget I mentioned Haswell. I meant Ivy Bridge,

If true that doesn't put intel that far behind 28nm for ATI/NV.

 

[epidemis]

The finfet sram stuff isn't for production, not at 22nm at least, just prototyping for data creation purposes. Very much like when you see press releases regarding EUV being used to make test chips years ago. 22nm will be planar cmos, HKMG.

Doesn't make the milestone any less cool or any less important, just saying this kind of milestone is not intended to be compared to the other kind of milestone you mentioned (pre-production wafer full of production-layout sram chips).

So the announcement isn't the same as the one Intel had at their IDF 09?

Excerpt from the 22nm SRAM announced at IDF 09

The particular chip is called a "shuttle test chip," as it tests different structures, including Static Random Access Memory (SRAM), logic tests, and so on. The basic chip has more than 2.9 billion transistors, with 364 Mb of SRAM, but also including the same transistors and interconnect features as will be used on 22nm CPUs, expected in about 2 years.

 

[Soleron]

NO . I meant to say that Knights Corner according to available sources is due late 2011, Ivy Bridge Is do out on 22nm in early 2012. So it appears Intel will introduce 22nm with Knights Corner.

Having read the AT article, I think he was speculating on Knights Corner's launch date based on 22nm timings, rather than the other way around. Intel haven't said otherwise, the only other site claiming to know its launch date is SemiAccurate, saying widely-available Larrabee will be 2012, either early or late depending on the source [indeed, Intel itself may not know precisely this far out].

22nm will be late Q4 '11 or early Q1 '12 just going on the usual pattern. I don't think Anandtech publicly stated more than we do about the precision on that.

However, the high-end (6 cores or more) Sandy Bridge will be H2 '11 so they won't want to launch two products in the same segment in the same year half. So I'm leaning towards 2012.

--

AMD have said 22nm SOI will be needed at GF "before H2 2012". So H1 2012 ramp for their 22nm too?

 

[Idontcare]

So the announcement isn't the same as the one Intel had at their IDF 09?

Excerpt from the 22nm SRAM announced at IDF 09

No it is not the same. Implementing finfet-based sram, and characterizing it, at node xyz has been done on every node since ~130nm. Doing it yet again at the 22nm is nothing new, nor was the publication itself intended to be news, but doing it at 22nm design rules and assessing the results is value-add which is why it was done in the first place (same motivation for doing it at 65, 45, and 32nm). This is pathfinding, purely exploratory.

You can't rule out an integration path with confidence if you haven't generated the data necessary to motivate downselection. Absolutely everyone in the industry does this stuff, it is just IBM that makes a business out of the licensing fees for their fab-club so they must do the marketing part (which is what this is) to demonstrate their technological prowess.

What Intel does at their dog-and-pony show is for their shareholders, and when it comes to sram they are showing a near finalized product design. Lightpeak and Larrabee were definite exceptions where they roll out the carpet to showcase stuff that likely isn't to see a market in its current form.

The glofo guys are not ready to showcase a 22nm sram wafer with production worthy integration at this point. Not their fault, they are working with 1/6 the revenue and budget.

 

[Nemesis 1]

Having read the AT article, I think he was speculating on Knights Corner's launch date based on 22nm timings, rather than the other way around. Intel haven't said otherwise, the only other site claiming to know its launch date is SemiAccurate, saying widely-available Larrabee will be 2012, either early or late depending on the source [indeed, Intel itself may not know precisely this far out].

22nm will be late Q4 '11 or early Q1 '12 just going on the usual pattern. I don't think Anandtech publicly stated more than we do about the precision on that.

However, the high-end (6 cores or more) Sandy Bridge will be H2 '11 so they won't want to launch two products in the same segment in the same year half. So I'm leaning towards 2012.

--

AMD have said 22nm SOI will be needed at GF "before H2 2012". So H1 2012 ramp for their 22nm too?

Thats good logic . But being I tend to read more into a thing than what may or may not be fact.
It is logical for Intel to move its 22nm fab process up for its GPUCPU part . It is also logical even tho its like spitting into the wind at this moment in time .To assume intel Highend in 4th/ 3rd qt next year sandy bridge would stall the introduction Of Knights Corner. I believe just the opposite would be the case. LRB and AVX on the same system sounds logical to me . But than again it was my Larrabee that seemingly has sank much like poor poor Itanic. Seems boomed to HPC but the Larrabbee elements are still in Knights Corner. SO I will hang on to the Idea Larrabee is just being polished . Skipping 32nm and going right to 22nm should make one think here . 45 was late hot power hog, 32nm would just get Intel into NV territory at present. 22nm tells me intel is trying to match up. SO Highperformance SB and LRB at the same time makes alot of sense to me.

 

[Idontcare]

Luck certainly favors the prepared, and with Intel's deeper pockets they can definitely win by attrition when it comes to node cadence and just biding their time as the gap between their leading edge node and that offered by the foundries (GloFo and TSMC) gets wider and wider.

 

[aigomorla]

Intel already has 22nm from what im hearing.

Its a matter on... IF they release it on time or not.

 

[Idontcare]

of course they already have 22nm...they showed the sram a year ago. 16nm is prolly only 12months out from sram.

In node development, roughly a four year process, the last year is spent ramping yield (this gets a lot of publicity) and more importantly/difficult the lifetime reliability metrics (rarely gets public mention).

So, could Intel ship 22nm based chips today, yes. They cost a shitload of money (if Intel wanted 50%GM on them) because the yields are crap and the mean-time-to-failure would be measured in weeks or months versus years and decades.

It's no different than where GloFo was with 32nm a year ago.

 

[IntelUser2000]

When the article says "conventional optical lithography", would that count for immersion as well?

 

[Idontcare]

Yes.

Conventional in this context means equipment that is currently used in production.

EUV would not be considered conventional. Immersion litho would be, but not the high-RI fluid immersion litho stuff that is still in development.