GlobalFoundries Outlines 22 nm Roadmap

[Idontcare]

GlobalFoundries Outlines 22 nm Roadmap

At the Fab 2 groundbreaking last week, Tom Sonderman, vice president of manufacturing systems technology, told reporters that GlobalFoundries "is closing the gap with Intel" on leading-edge technology introductions.

"By going to a shared technology development model with IBM," he said, "we are closing the gap. We were a year behind Intel at the 45 nm node, and that difference will be cut significantly at the 32 nm generation. By 22 nm, there will be no difference. It will be in the noise level."

http://www.semiconductor.net/a...ines_22_nm_Roadmap.php

Lots of other sound-bites in the link relating to various process technology choices they have made for 22nm. I quoted the part I figured would rank as the most interesting the forum goers here...22nm being released same time as Intel.

The question - or concer over the possible spin - I have here is that GF doesn't sell products like Intel, they sell foundry space. So when GF compares themselves to Intel are they saying their customers (AMD, etc) will be shipping 22nm IC's at the same time that Intel begins shipping their 22nm IC's? (meaning GF's 22nm node itself will have already debuted, gone thru risk production, and be production ready for volume at the same time as Intel's 22nm)

Or is GF trying to be sly here and say their customers (AMD, etc) will have access to first silicon on 22nm (post tapeout, but pre-respin/pre-risk production/pre-volume ramp) at the same time that Intel is shipping their 22nm products to Newegg?

There is a difference, but I can't tell which way GF is trying to spin this "closing the gap" angle. I hope it means Bulldozer 22nm shrink (I have no idea on the codename) from AMD will be in stock and available to ship from Newegg within 30 days of Ivy Bridge being in stock and available to ship from Newegg.

If it doesn't mean that then its really a waste of everyone's time for GF's to imply that this will be the case IMO.

edit: fixed link


edit #2: adding link to Anand's article on the AT frontpage regarding GF's roadmap and fab background:

Globalfoundries Starts on 22nm Fab & Announces First non-AMD Customer

Until this morning, Globalfoundries only had a single customer - AMD, but that just changed with the announcement that STMicro would be using their fabs.

http://www.anandtech.com/cpuch...ts/showdoc.aspx?i=3614

 

[Eeqmcsq]

I'm going to assume that GF's "close the gap" goal is for their customers to release 22nm stuff at practically the same time as Intel's 22nm stuff. Whether they make that goal only time will tell. For AMD's sake, and the sake of competition, I hope they succeed at matching Intel by 22nm.

 

[KingstonU]

Over the last 3 or so years I definitely remember reading a lot more news about 22nm from IBM/AMD than 32nm, even though 22nm is further away (2012 versus 2010 for 32nm)

 

[ilkhan]

kingstonU: true. But intel doesnt have a history of tooting their horn like IBM/AMD has to do. They say they have SRAM, they say samples, they say done. Etc.

Intel 22nm: Didnt we figure AMD would just be releasing 32nm on early 2011 the other day? <12mo to go 32nm->22nm would be impressive.
yeah: http://forums.anandtech.com/me...id=28&threadid=2319424

 

[Idontcare]

Originally posted by: ilkhan
kingstonU: true. But intel doesnt have a history of tooting their horn like IBM/AMD has to do. They say they have SRAM, they say samples, they say done. Etc.

Intel 22nm: Didnt we figure AMD would just be releasing 32nm on early 2011 the other day? <12mo to go 32nm->22nm would be impressive.
yeah: http://forums.anandtech.com/me...id=28&threadid=2319424

Yep...the words they say versus the implications of what must be done to achieve what the words mean create a startling declaration on behalf of GF...

We have a near identical node-cadence aggressiveness (in posture thus far, not in reality) coming from TSMC with their repeatedly stated roadmap of releasing 28nm in 2010.

 

[VirtualLarry]

Isn't this rate of introducing die shrinks, going to bleed AMD and IBM red? I mean, they have to spend millions (?) on investing and developing this technology, and they've got such an aggressive shrink timeline, that they aren't giving time for them to sell those products, and make a return on their R&D investment.

Is parity with Intel's mfg tech of such importance, that they are going to put profit on the sidelines? Or am I reading this wrong?

 

[firewolfsm]

Originally posted by: VirtualLarry
Isn't this rate of introducing die shrinks, going to bleed AMD and IBM red? I mean, they have to spend millions (?) on investing and developing this technology, and they've got such an aggressive shrink timeline, that they aren't giving time for them to sell those products, and make a return on their R&D investment.

Is parity with Intel's mfg tech of such importance, that they are going to put profit on the sidelines? Or am I reading this wrong?

But the fact of the matter is that they won't make that profit on parts made on older node processes. Without ramping up faster, they simply cannot compete at a profitable level.

Plus, with AMD and IBM sharing the expenses on the same research it shouldn't be as difficult to concentrate on selling processors at the same time.

 

[Fox5]

Originally posted by: VirtualLarry
Isn't this rate of introducing die shrinks, going to bleed AMD and IBM red? I mean, they have to spend millions (?) on investing and developing this technology, and they've got such an aggressive shrink timeline, that they aren't giving time for them to sell those products, and make a return on their R&D investment.

Is parity with Intel's mfg tech of such importance, that they are going to put profit on the sidelines? Or am I reading this wrong?

If they're competing head to head with Intel, their products won't sell for much if they can't keep parity. Had Phenom II launched around the time of the 45nm Core 2's, AMD might be pretty healthy right now. I assume IBM's products got a similar beating.

Also, since GF now sells fab space to anyone, there's big money in being the world leading fab. They can offer cheaper costs (eventually) and higher performing parts as well. Additionally, they may be able to optimize their process tech for low-power consumption better than TSMC, they have more experience with coming up with fabrication technologies rather than just circuit design to lower power consumption.

 

[BTRY B 529th FA BN]

Why does IBM have to worry about Intel by helping AMD? Or are they? I mean by sharing whatever they are sharing they, im sure, are interested in something but what? I can't believe they want to keep the industry from being monopolized or out of balance in favor or Intel. Thus what in return will be accounted for? Money? That'll prolong AMD's share profit in the market ....O am i seeing this in short terms while the plan is long term?

 

[deputc26]

Originally posted by: VirtualLarry
Is parity with Intel's mfg tech of such importance,

Yes. Intel is a year ahead right now which means that half the time Intel is competing with AMD, they can make twice the chips for (close to) the same cost. This is a massive advantage that AMD/IBM cannot keep giving to Intel if they plan to survive in the long term.

 

[Idontcare]

Originally posted by: BTRY B 529th FA BN
Why does IBM have to worry about Intel by helping AMD? Or are they? I mean by sharing whatever they are sharing they, im sure, are interested in something but what? I can't believe they want to keep the industry from being monopolized or out of balance in favor or Intel. Thus what in return will be accounted for? Money? That'll prolong AMD's share profit in the market ....O am i seeing this in short terms while the plan is long term?

IBM and AMD are not competitors, but both IBM and AMD compete with Intel. At a superficial level (i.e. not reality but lets make it anthropomorphic and go with it anyways) you could view it as an example of the adage "the enemy of your enemy is your friend".

That and IBM, just like AMD, can't afford to develop all the leading edge process tech on a timeline that remains competitive with Intel's timeline...so by creating the IBM fab club (formally referred to as their alliance ecosystem) and enabling all the smaller fish (IBM, AMD, Chartered, ST, Freescale, NEC, Toshiba, Samsung and others) can pool their respective R&D resources into a much larger R&D budget which is then competitive with the scale and scope of Intel's.

It's a cost-effective way of commoditizing process technology R&D, if you will, in that it removes process technology itself as a contributing differentiator to the product differences that compete in the open marketplace between the alliance members. No different than how we view the business models of Nvidia and ATI/AMD, for example, when comparing two of their GPU's which were fabbed at the same process node and foundry (e.g. 55nm TSMC)...only in this case the alliance members want to retain control over their inhouse manufacturing processes (possibly for cost reasons, also for supply stability, among others) so they jointly create the process technology but don't actually use each other as foundries per se.

 

[drizek]

In the past, AMD could get by with a lesser manufacturing process because the architecture compensated for it and allowed them to remain competetive with intel. Since Core 2 however, that has not been true, with AMD having arguably an inferior architecture(phenom II just about has parity I would say). Global Foundries is now its own company, and they can't be propped up by architecture any longer, even if AMDs was superior. They need to remain competitive and catching up with intel is absolutely necessary for their survival. Now is as good a time as ever to get aggressive. Phenom II is doing reasonably well, probably better than any other product AMD releases between now and Bulldozer. If they don't invest quickly and heavily in RnD, they will continue to lose money every quarter, making it harder and harder to catch up.

 

[deputc26]

Excellent explanations drizek and IDC, I was just wondering how AMD had managed to be competitive with Intel for so long with inferior process tech. I had shakily decided that their fabs must have been significantly cheaper to operate; architecture superiority makes much more sense.

 

[taltamir]

making roadmaps and MEETING DEADLINES are two completely different things. I have no faith in them being able to release 22nm at the same YEAR as intel.
Let me rephrase, I have no faith in AMDs piss poor management to do so. Bleeding talent and money while having horrible management with CEOs that look at a company as their own personal cash cow without regards to the company's longevity (which is only because the board and owners let them get away with it) is not a way to success...

Besides, IDC, you taught me that nowadays the nm label is just a label...

 

[IntelUser2000]

It won't happen. The roadmaps say Bulldozer in 2011 will be the first 32nm product, and we are going to see faster Phenom IIs as year progresses(2010 in this case). There are roadmaps for a 3.8Ghz Phenom II(yes 3.8GHz).
See, the first thing about being a good business is being profitable. I'm pretty sure its possible that Intel can release 32nm products by late Q3 or early Q4. Like that will happen. There were reports that manufacturers asked Intel to delay Calpella to clear inventory on the existing, Montevina platforms.

Delays are certainly possible, but everything earlier really isn't practical. And delays on previous generation platforms compound with release on the next generation platforms. You want enough time to get profits out of the investment(in this case, lithography) before moving onto the next one.

I used to think differently but after looking back all these years, its practically impossible.

The time that AMD was closest to Intel in process technology in the last 15 years(1998-now) was in the 90nm technology era when AMD was mere 6 months behind, but that had to do with the delay of 90nm Prescott CPU. Now the difference between them is the usual 11-12 months. The average difference between each generation for both Intel and AMD is 23 months. The shortest between each generation was 22 months.

Oh, and for those that don't know. For the full-node process(for example 65nm vs 65nm) TSMC's actual products are available on the market a full year behind Intel products. TSMC might say they are ramping earlier or w/e, but nothing earlier on the real product.

Something changing after 15 years of being similar?? Quite unlikely.

AMD won't be that bad off as some might think since there are no faster clocked Intel parts coming(actually maybe a minute increase with a 3.46Ghz i7 EE). I'm willing to say not even with Sandy Bridge we will see faster clock products.

 

[Idontcare]

Link to Anand's article on the AT frontpage regarding GF's roadmap and fab background:

Globalfoundries Starts on 22nm Fab & Announces First non-AMD Customer

Until this morning, Globalfoundries only had a single customer - AMD, but that just changed with the announcement that STMicro would be using their fabs.

http://www.anandtech.com/cpuch...ts/showdoc.aspx?i=3614

Originally posted by: taltamir
Besides, IDC, you taught me that nowadays the nm label is just a label...

You raise an excellent point to the conversation taltamir...the real question here is not when will 22nm AMD products debut; but rather the question is will they debut on a competive process technology? If GF's releases their 22nm node at the same time as Intel releases their 22nm but the xtor specs and overall performance metrics (power, clockspeed, etc) of GF's 22nm are no better than say those of Intel's 32nm node then does it really matter whether Globalfoundries calls their 22nm node the 22nm node or the 16nm node or the apple node?

So you are right, closing the gap in the timeline between releasing node labels does not require them to close the gap in the actual process technology capabilities. It remains to be seen what the Idsat, Ioff, Vdd, IDDQ, etc specs turn out to be for GF's 22nm.

If they merely close the gap to Intel by delivering even lower performing xtors (in the past AMD was both late and lower performing...12 months later than Intel to deliver 45nm and even then the Idsat, etc, were lower than Intel's) for the technology node than they might have otherwise delivered had their timeline included a 12 month gap then it equally fails to serve their customers and ultimately the consumers as the products generated will be all the less competitive in the marketplace when we go to Newegg to make our purchasing decisions.

Originally posted by: deputc26
Excellent explanations drizek and IDC, I was just wondering how AMD had managed to be competitive with Intel for so long with inferior process tech. I had shakily decided that their fabs must have been significantly cheaper to operate; architecture superiority makes much more sense.

Two key contributing reasons - first is gross margins and revenue, the second is APM.

There is a difference between spending money so liberally that your business is effectively a non-profit organization over the course of 3 decades versus restricting your spending and confining the budgets of your R&D and manufacturing groups such that they remain within a fiscal footprint as needed to enable the company as a whole to reach >50% gross margins and routinely deliver consistent dividend payouts to the shareholders for decades.

Likewise AMD did develop and implement one of the most sophisticated automated closed-loop feedback control systems (called APM for Automated Precision Manufacturing) that has a production fab environment has ever seen to my knowledge and in my firsthand experience. For more info on AMD's APM checkout this online video published by AMD/

Among enabling other things, a key advantage for AMD in having APM is that APM enables the effective implementation of their CTI (Continuous Transistor Improvement) and STT (Shared Transistor Technology). Without the rigorous in-line analysis provided by way of APM, introducing substantial changes to the process flow as needed to implement CTI milestones would be a serious reliability and QA qualification nightmare (which is why practically no one does it).

 

[tokie]

It will be interesting come 2011/2012 when production ramps up for GF. I wonder what will happen to UMC, who are already behind TSMC. Seems like TSMC and GF will both be on similar timelines?

 

[Idontcare]

Originally posted by: tokie
It will be interesting come 2011/2012 when production ramps up for GF. I wonder what will happen to UMC, who are already behind TSMC. Seems like TSMC and GF will both be on similar timelines?

There are many differences between GF and the rest of the traditional foundry players (TSMC/UMC/Chartered/etc) and that is in terms of revenue shipping from legacy nodes and existing customer base.

Checkout slide 7 of this most recently published TSMC financial report and not that <25% of their sales comes from 65nm and smaller nodes.

GF is unencumbered by the requirement of allocating resources to maintain a pre-existing legacy customer chain and node base. On the one hand it is desirable to have this "base" as it delivers the consistent sales volume year over year, but on the other hand it does reduce the degrees of freedom and nimbleness with which a fab can be operated and managed in the face of shifting customer orders and demands.

In Globalfoundries situation it makes them very dangerous at 28nm and 22nm as they are in a "do or die" type situation when it comes to snagging customers from UMC and TSMC as well as under-promising and over-delivering on both their timeline and breadth of their process technology offerings.

The folks working on TSMC's 28nm and 22nm will feel all the less heat and all the less sense of urgency when it comes to carrying out their daily tasks as the ramifications of their inactions (or slower actions relative to GF) will not be so immediately felt across the company nor with the same maginitude of repurcussions as would be the case within GF.

(consider the 40nm fiasco, TSMC will survive as >75% of their revenue is still 90nm and older tech, but GF would be fatally undermined if they had similar fiasco at 28nm or 22nm)

I dare say right now is the perfect time to be a GF process development engineer...you have management's full attention and resources have got to be at their highest levels ever. From here out it has to be a decline...either because 28nm/22nm wasn't a success and the company marches into deathspiral (and resources become scarce) or because 28nm/22nm becomes a smashing success and management decides the R&D team was/is over-resourced going forward as they try and cut costs so GF as a company begins to net profits. It's the standard life-cycle of american management. (naturally I am jaded to view it this way as this is exactly what TI management elected to do between 90nm to 65nm to 32nm, and moto management between 180nm to 130nm to 90nm)

If you are a GF process development engineer I say enjoy these next two years as they will likely be the best you will experience at GF.

 

[exar333]

Originally posted by: deputc26
Excellent explanations drizek and IDC, I was just wondering how AMD had managed to be competitive with Intel for so long with inferior process tech. I had shakily decided that their fabs must have been significantly cheaper to operate; architecture superiority makes much more sense.

They have'nt been competitive (in a business sense) with Intel, that is the whole issue. Intel makes money and AMD loses money, much of this has to do with what was previously posted about how Intel can make twice the chips for the same cost, and often sell them for more $$$.

 

[Atechie]

Originally posted by: ExarKun333

Originally posted by: deputc26
Excellent explanations drizek and IDC, I was just wondering how AMD had managed to be competitive with Intel for so long with inferior process tech. I had shakily decided that their fabs must have been significantly cheaper to operate; architecture superiority makes much more sense.

They have'nt been competitive (in a business sense) with Intel, that is the whole issue. Intel makes money and AMD loses money, much of this has to do with what was previously posted about how Intel can make twice the chips for the same cost, and often sell them for more $$$.

Just look at the die size of Phailure2...the same as the Core i7 die size.
Now look at the prices for the either a Phailure2 or a Core i7...and then compare the Die size of a Core2Quad with the Phailure2 and the price.

Don't need to be a rocket scientist in order to see why AMD is bleeding money.

 

[ArchAngel777]

Originally posted by: Atechie

Originally posted by: ExarKun333

Originally posted by: deputc26
Excellent explanations drizek and IDC, I was just wondering how AMD had managed to be competitive with Intel for so long with inferior process tech. I had shakily decided that their fabs must have been significantly cheaper to operate; architecture superiority makes much more sense.

They have'nt been competitive (in a business sense) with Intel, that is the whole issue. Intel makes money and AMD loses money, much of this has to do with what was previously posted about how Intel can make twice the chips for the same cost, and often sell them for more $$$.

Just look at the die size of Phailure2...the same as the Core i7 die size.
Now look at the prices for the either a Phailure2 or a Core i7...and then compare the Die size of a Core2Quad with the Phailure2 and the price.

Don't need to be a rocket scientist in order to see why AMD is bleeding money.

This is just flaming material. The Phenom II is a great product and moved AMD in the right direction.

 

[Martimus]

Originally posted by: Atechie
Just look at the die size of Phailure2...the same as the Core i7 die size.
Now look at the prices for the either a Phailure2 or a Core i7...and then compare the Die size of a Core2Quad with the Phailure2 and the price.

Don't need to be a rocket scientist in order to see why AMD is bleeding money.

The funny thing is that you might actually have some valid points, but no one will listen to them due to your presentation.

 

[heyheybooboo]

Originally posted by: Idontcare

edit #2: adding link to Anand's article on the AT frontpage regarding GF's roadmap and fab background:

Globalfoundries Starts on 22nm Fab & Announces First non-AMD Customer

Until this morning, Globalfoundries only had a single customer - AMD, but that just changed with the announcement that STMicro would be using their fabs.

http://www.anandtech.com/cpuch...ts/showdoc.aspx?i=3614

ATIC owns 65.8% of the new company and AMD owns the rest, although the two share voting rights.

When Anand has trouble sleeping he may want to plow through some of AMD's SEC filings. There are instances listed where action by the directors requires unanimous consent and (IIRC) AMD maintains veto power over appointments to the Board.

I don't know what to really think of all that - other than AMD has not relinquished control of much of anything and this would seem to conflict with the move in public companies for more independent directors not tied to company management (not Hector bashing here) ...

 

[drizek]

Reading the front page article makes all this look a lot better. I didn't realize they had that much cash.

Yesterday I was thinking about what it could mean for ARM when intel takes Atom to 32nm. I had no idea that the iPhone 3G was still pimping 90nm and the 3GS was only 65nm.

Even if GF is behind Intel, they will still be way ahead of their other competitors for things like SoCs and GPUs. I think it is fair to say that ARM is more energy efficient than x86, but Intel can catch up if they get far enough ahead in manufacturing. 40nm and 28nm from GF can help to really give these companies a leg up.

I think GF has a lot of potential. They will be manufacturing GPUs, CPUs, SOCs and (I'm assuming) SSD flash.

 

[Idontcare]

Originally posted by: heyheybooboo
I don't know what to really think of all that - other than AMD has not relinquished control of much of anything and this would seem to conflict with the move in public companies for more independent directors not tied to company management (not Hector bashing here) ...

I foresee this as being very transitory, AMD would have no problem at all relinquishing control of GF to ATIC but are treading a fine line over the whole x86 license situation with Intel (including the current round of lawsuits and countersuits).

In my version of reality I like to believe AMD intends to hold onto this position in GF until they can negotiate new terms to their x86 license with Intel such that it becomes worded to allow AMD to use any foundry of their choosing and that AMD intends to use the upcoming rounds of anti-trust actions in the US (plus the lawsuits) as leverage to generate such favorable rewording of the license.

Could all be malarkey, but you just know AMD could not have convinced ATIC to walk into this deal without there being some kind of pre-planned/stated/agreed upon exit strategy for AMD to get out of GF's BOD because until they do it will only serve as ammo for TSMC/UMC/etc to undermine GF's legitimacy as an independent foundry for any customer but AMD when it comes to propaganda and FUD with any alternative would-be GF customers (like Nvidia).