Print 44 comment(s) - last by flipsu5.. on Sep 25 at 4:14 AM

IBM's new breakthrough could allow its next generation of Cell processors to reach 22 nm before Intel does.
New techniques will allow tinier computer chips than ever before

The race to shrink circuits is a never ending one.  Having achieved 45nm with its Penryn processors, Intel is looking to move to 32nm next year for the die-shrink of its upcoming Nehalem processor.  Meanwhile, competitor AMD is struggling to get its first 45nm offering, Shanghai, ready for release later this year

However, while the back-and-forth between AMD and Intel often steals the show, the race for die shrinks has several other key competitors.  One is Texas Instruments (TI), who claims to lead Intel in 45nm deployment and performance.  While TI currently doesn't have much in the way of PC/server processor plans, IBM does.  IBM is looking to beat AMD and Intel in die-shrinks and pave the way for its processors to dominate a number of markets, including the PC/server space.

Having already shrunk its powerful Cell processor used in the PS3 to a 45nm production process, IBM is planning to quickly transition to 22nm.

At 22nm, the existing techniques for microprocessor manufacturing useful at the 45nm and 32nm nodes will become obsolete.  While currently lithography techniques cannot produce 22nm circuitry, IBM has developed a new approach called Computational Scaling, which will allow for this tiny-scale production.  The new technique uses advanced mathematical computation to adjust the shape of the masks and illuminating source during etching.

One key advantage of IBM's breakthrough is that it can advance from 32nm to 22nm without fundamental changes to its equipment.  While Intel and AMD are likely to devise similar techniques, the early breakthrough may allow IBM to seize a lead in shrinking, helping it to continue Moore's Law.

Advantages of such smaller processor are numerous.  Due to the shorter interconnects, and other factors, smaller die-sizes yield power savings.  Further, the tiny size will allow for heat reduction.  Finally, it may also allow for more cores in similar packaging.

Like Intel, IBM is planning to bring its 32nm processors to market in 2009.  However, whether it will adhere to Intel's 2011 deployment of 22nm or jump the gun is up in the air.

One thing that is clear is that IBM has plans to deploy 22nm server chips to compete with Intel's offerings.  IBM hopes to apply the new tech to its cloud computing efforts.  It hopes the greater efficiency and processing power afforded by 22nm will help to justify an increasing amount of business and private processing to be offloaded to cloud computing datacenters.

IBM also announced in August that it and its partners had developed a method of manufacturing static RAM (SRAM) cells at a 22nm node.

Comments     Threshold

This article is over a month old, voting and posting comments is disabled

C'mon Jason, dig a little deeper
By Amiga500 on 9/22/2008 1:46:23 PM , Rating: 3

At the start of the 2008 Intel Developer Forum, IBM, AMD, Freescale, STMicroelectronics, Toshiba and the College of Nanoscale Science and Engineering announced that they have developed the first 22-nanometer SRAM memory chips. These SRAM chips are the precursors to 22-nm microprocessors. IBM and its partners are looking to take the lead against Intel processors, which are expected to shrink to 32 nm in 2009.

Thus, the technology will be open to AMD if they can upgrade quick enough.

By TheFace on 9/22/2008 3:15:14 PM , Rating: 3
So who is the Axis and who are the Allies?

RE: C'mon Jason, dig a little deeper
By MAIA on 9/22/2008 7:21:10 PM , Rating: 2

IBM, AMD,...

That was exactly what i was thinking ...
Now, draw your new conclusions from here.

RE: C'mon Jason, dig a little deeper
By cheetah2k on 9/23/2008 10:24:19 PM , Rating: 2
I think its about time IBM bought out AMD. Share prices are low, AMD is picking up their game (albeit in graphics), and IBM have cash to burn.

While I know its just not IBM's strategic approach, such a purchase would enable IBM to take Server, Desktop and Graphics markets by storm and make the battle between David (aka AMD) and Goliath (Intel) so much more competitive. I'm confident that with funds and the right motivated management/staff, AMD would have the tools and resources to get out of the gutter, and back into the drivers seat...

I'm not a fanboy, but as an Australian I do love an "under-dog". However, if something doesn't change for AMD soon (including the US economy), with the AMD's headstone already carved and grave already dug (Intel's new 6 core Xeon), the future isnt looking bright at all..

By leidegre on 9/24/2008 3:41:49 AM , Rating: 2
This idea has some appeal to me. It actually sounds like something that really could (and maybe should) happen. Interestingly enough, the XBOX 360 features an IBM CPU and an AMD (formerly ATI) GPU. As they say, if you can't beat them, join them. As well, the CBEA/PS3 (Cell Broadband Engine Architecture) was at least co-developed by IBM. The PowerPC is pretty much gone, but IBM lives on through our consoles.

By Dasickninja on 9/24/2008 12:22:19 PM , Rating: 2
IBM doesn't want to have to get into the commodity CPU market again, especially with the way things are looking for x86 CPUs. They make far more money selling the technology and licencing that they did when they were a chip maker. It may be a good thing for AMD if IBM bought them out, but what really is IBM's interest in opening that can of worms?

By masher2 on 9/22/2008 2:20:38 PM , Rating: 4
At 22nm, the existing techniques for microprocessor manufacturing useful at the 45nm and 32nm nodes will become obsolete. While currently lithography techniques cannot produce 22nm circuitry, IBM has developed a new approach called Computational Scaling, which will allow for this tiny-scale production
It may not be clear from the above, but computational scaling isn't an entirely new approach supplanting current litho techniques. It is, in fact, a method that allows those current processes to continue to work at the 22nm node, rather than having to switch to entirely new methods such as EUV.

RE: Correction
By johnsonx on 9/22/2008 3:48:50 PM , Rating: 1
So you mean yet another sensational sounding Jason Mick article is much ado about nothing? Say it ain't so!

RE: Correction
By masher2 on 9/22/2008 4:02:38 PM , Rating: 2
I wouldn't go that far. Being able to stretch existing 45nm techniques to the 22nm node is a worthwhile development. Jason's wording, however, cast it as supplanting those techniques, rather than supplementing them.

RE: Correction
By johnsonx on 9/23/2008 1:24:43 PM , Rating: 5
Yes, I know, I was being a bit naughty. I guess the masses did not approve. Still, you have to admit, Jason Mick tends to word things in a way that makes them sound like more than they are.

By kilkennycat on 9/22/2008 2:24:50 PM , Rating: 2
IBM has a chronic history of being able to produce advanced-process silicon in very low-volumes successfully, but unable to QUICKLY transition any of their process-technologies to high-volumes with any decent yield. Not surprising really, because their silicon business is primarily focussed on IBM-specific targets, where low-volume is perfectly satisfactory. So what if the yield is 5% on an custom-design IBM server chip and only < 100K per year are produced ?

Intel, on the other hand, tailors its process-development for high-volume production from the get-go, and will not publicly announce a new process at all until they have proven confidence in satisfactory high-volume yields.

IBM has provided AMD with the process technology intended for their 45nm CPU. Notice the schedule delays? Nine months ago AMD (the late lamented Hector..) was promising 45nm CPU releases starting early in Q3 2008. Those releases were desperately needed by AMD to put serious Core2 competition in place before Nehalem arrived and regain some traction in the marketplace. That particular scenario has not happened and AMD's 45nm efforts are now going to run head-on into Nehalem.

Want to take a bet that the major problem causing the six-month delay is NOT still-unsatisfactory high-volume process-yield ? I notice that some AMD info *leaked* today from a Chinese site seems to indicate the release of a couple of 45nm processors in Q4 2008.


No indication as to whether shipments will start at the beginning or end of Q4. Since we are well into Q3 and since there have been zero public showings, public benchmarks or public shipment announcements of AMD's 45nm family, I suspect that only a very few will be shipped in Q4 2008, if any at all.

By TheFace on 9/22/2008 3:19:02 PM , Rating: 2
Interesting note too, the chips used in the PS3 (Cell), and the chips used in the 360 (PowerPC based) are both put out by IBM. PS3 was late for a variety of reasons, but the cell processor was a factor. 360 was less late but had it's fair share of processor related problems (although the graphics solution could be said to have caused much more).

By Etsp on 9/22/2008 11:13:55 PM , Rating: 2
The Wii's CPU is loosly based on the PowerPC, and is also produced by IBM...

Nice, but not enough
By fri2219 on 9/22/2008 2:03:26 PM , Rating: 2
IBM has always been good at announcing breakthroughs with things like Cu interconnects, SOI, and fast interconnects, but they've never been able to get decent yields on their wafers.

I'll believe their 22nm process exists in a commercially viable form when I see them selling PowerPC or Cell chips for less than $1000 a piece in bulk.

RE: Nice, but not enough
By FITCamaro on 9/22/2008 5:34:51 PM , Rating: 2
Well considering the PS3 has one of the most expensive variants of the Cell processor in it and it only costs $399, I really don't think the Cell processor is going for $1000 in lots of 1000.

RE: Nice, but not enough
By fri2219 on 9/22/2008 6:34:31 PM , Rating: 1
The $1000 comment was a (somehow cryptic) reference to moving the technology off their Power5 or Power6 chips.

IBM, Toshiba, and Sony have all had a horrible time producing the Cell architecture in a profitable manner at 65nm- their true cost of manufacture when you amortize R&D as well write-offs as has been consistently above the market equilibrium.

11nm.. isn't small enough for me..
By snownpaint on 9/22/2008 4:25:12 PM , Rating: 1
Until I can have my desktop in my cell phone, I'm not impressed. I have designated that space under my desk for that case space. Cooling and power isn't a issue.

I have a dual-core that is overclocked and have no problems running all kinds of processes at once and a really fast boot up, especially for a guy that is a "i clicked the button and want it now" kind of attitude. Even Crysis runs great at full res. and loads maps in 2 seconds.

RE: 11nm.. isn't small enough for me..
By masher2 on 9/22/2008 5:10:35 PM , Rating: 3
> "Until I can have my desktop in my cell phone, I'm not impressed"

The 10nm node should be about the time that happens.

By FITCamaro on 9/22/2008 5:35:17 PM , Rating: 1
It just won't last longer than five minutes.

Not being a smart ass, just curious
By Cincybeck on 9/23/2008 12:50:54 AM , Rating: 2
How did they decide on 130nm to 90nm to 65nm and so on? Is there a meaning behind these numbers or can they just pull a number out of their ass like "Let's make 35nm now and skip on down to 19nm later on." ..? Unless I'm failing on my math, which isn't out of the question.. It's not half the area is it?

By bridgeman on 9/23/2008 1:26:29 AM , Rating: 2
Yeah, each major process node halves the area. Feature size scales by about 0.7x, which scales the area by 0.5x (=0.7^2).

Good ol' IBM
By geddarkstorm on 9/22/2008 1:30:18 PM , Rating: 2
Good to see they are still continuing their tradition of innovation in the PC realm: these are some cool developments. Personally, I'm really looking forward to see what comes of that new RAM they are developing. Now that stuff sounds truly inventive.

Very impressive but...
By riku0116 on 9/22/2008 7:04:54 PM , Rating: 3
Can it run Crysis?

RE: Very impressive but...
By bterry62 on 9/22/08, Rating: -1
32 nm in 2009?
By flipsu5 on 9/25/2008 4:14:26 AM , Rating: 2
Haven't heard where Intel is putting its 32 nm fabs. By this time we should have heard, if they are planning 2009 32 nm.

April 2006 we already knew 3 45 nm fabs for 2007; today we can guess maybe one (D1D).

I don't know what this foreshadows for IBM/AMD; they don't have many fabs to begin with. Probably 32 nm slowdown as well.

By dagamer34 on 9/22/08, Rating: -1
By Joz on 9/22/2008 1:40:24 PM , Rating: 2
Its becouse going to 22 right off the bar more effiecent and cheaper then going to 32. Why would they go 32, if they can go 22?

By theplaidfad on 9/22/2008 1:41:16 PM , Rating: 4
A failure at reading comprehension has just occured, everyone hit the deck!

They are not skipping 32nm. Re-read the article.

By amanojaku on 9/22/2008 1:44:07 PM , Rating: 2
Good eye. The hardware will be able to move from 32 to 22nm chips with little modifications.

By amanojaku on 9/22/2008 4:36:24 PM , Rating: 2
3rd paragraph from the bottom confirms.

Like Intel, IBM is planning to bring its 32nm processors to market in 2009. However, whether it will adhere to Intel's 2011 deployment of 22nm or jump the gun is up in the air.

By amanojaku on 9/22/2008 1:41:59 PM , Rating: 2
I was pretty sure that AMD lends some tech from IBM when doing R&D for new manufacturing processes, thus IBM cannot "beat" AMD to the punch.

Why not? IBM doesn't have to give AMD it's 22nm process secrets, or even access to its fabs.

But why on earth would IBM be skipping 32nm to go straight to 22nm? Just seems weird to me.

Because AMD and Intel have been bringing Uzis to a knife fight. IBM decided to drop a bomb.

By theplaidfad on 9/22/2008 1:43:12 PM , Rating: 3
Great googly moogly!

Read the article, THEN post!

They aren't skipping 32nm!

By Oregonian2 on 9/22/2008 2:02:36 PM , Rating: 2
Nor do they say that they're ahead of Intel in terms of when parts come out of the door in production parts. Only dates mentioned are for 32nm parts where both Intel and IBM say 2009. One could interpret IBM's announcement just to say that they're not behind Intel and that they may be able to save money as compared to the way Intel is doing it.

By Master Kenobi on 9/22/2008 2:39:41 PM , Rating: 4
IBM is always about announcements of discoveries but never of commercialization. Intel has historically always been first to bring a new process or node to the mass market even when IBM announces they did it first.

By lifeblood on 9/22/2008 3:42:15 PM , Rating: 3
I see I am not the only one to notice that. IBM seems so quick to toot their own horn but they don't seem to actually get it out the door in into use.

I wonder if that's how IBM get's their money? Invent something cool in the lab, make a big deal about it regardless of how practical it is, then license it to someone to try to make it commercially useful. In this case AMD looks to be the company that has to take that cool theory and turn it into an actual money making product or process.

By Oregonian2 on 9/22/2008 5:41:54 PM , Rating: 3
Well, thing is that IBM's technology is for the most part only for "in house" use (although they made PPC IC's for Apple, some even with Motorola logo's on them, rumors have it). Their corporate focus is as a services company, not as a technology component selling one (even if they do that). Focus isn't where AMD and Intel's are. Their IC's are primarily a means to an end (fast servers providing services to customers), as opposed to AMD and Intel where the IC's are for the most part a major end product. Next major announcement on this for them may very well be for servers using the technology.

By melgross on 9/22/2008 1:43:08 PM , Rating: 4
AMD may share some tech with IBM, but it doesn't seem to do them any good.

Aren't you aware of what's been going on with AMD? Did you read the article?

AMD is always behind IBM and Intel by a good bit. Both IBM and Intel have been doing 45nm for about a year. Where is AMD? They promised to move to within 6 months of Intel's transitions, but it looks as though they will remain at least a year behind, or more.

How can you talk about them keeping up with IBM? They can't. They are a full generation behind, as usual. There's no reason to think that will change anytime soon.

And as a reminder, to read the article, perhaps you should try to understand what they're saying as well. At no time did they say that IBM was "skipping" 32nm.

By theplaidfad on 9/22/08, Rating: 0
By MrPoletski on 9/22/2008 7:27:57 PM , Rating: 2
I would rate you up too, if I hadn't just disqualified myself from rating you up by saying that.

By Goty on 9/22/2008 4:14:01 PM , Rating: 2
I guess it's fortunate that being "first" is synonymous with being "best".


Anyone remember Prescott? Yeah, I thought so.

By AstroCreep on 9/22/2008 6:18:56 PM , Rating: 2
I'm having flashbacks...

Uncle Prescott - don't touch me there!

By melgross on 9/23/2008 2:23:30 AM , Rating: 2
I hope you're not suggesting that AMD is "best", are you?

If so, it's good for laughs.

By CommodoreVic20 on 9/22/08, Rating: -1
By Tesseract on 9/23/2008 3:11:38 PM , Rating: 3
Exactly as fast as four mad horses can pull your limbs from your torso.

"A politician stumbles over himself... Then they pick it out. They edit it. He runs the clip, and then he makes a funny face, and the whole audience has a Pavlovian response." -- Joe Scarborough on John Stewart over Jim Cramer

Copyright 2016 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki