X-bit Labs: IBM Quietly Starts to Make Chips for AMD

[AtenRa]

IBM and GLOBALFOUNDRIES Begin First Production At New York’s Latest Semiconductor Fab

First products from Fab 8 developed and manufactured in New York’s ‘Tech Valley’

Saratoga County, N.Y. - 09 Jan 2012: GLOBALFOUNDRIES and IBM (NYSE: IBM) today announced an agreement to jointly manufacture advanced computer chips at the companies’ semiconductor fabs in New York’s “Tech Valley.” The chips are the first silicon produced at GLOBALFOUNDRIES’ newest and most advanced manufacturing facility, "Fab 8" in Saratoga County, and are planned to ramp to volume production in the second half of 2012. The new products recently began initial production at IBM’s 300mm fab in East Fishkill.

Heh, GloFo's FAB 8 in New York

 

[ViRGE]

Whatever AMD's initial ownership share of GF was, it has been whittled down consistently over the months and years since.

I think AMD are now down to about 12% and forecast to go to 0% within 2 years or less.

Officially it's down to 8.8% as of their Q4 2011 earnings report.

 

[Keysplayr]

Maybe AMD didn't have full faith in GloFlo to pull off satisfactory 32nm yields and used IBM as a bit of a cushion. Now they are locked in for how ever many wafers. I don't think this is necessarily a bad thing, all eggs in one basket and such. I remember Nvidia used IBM for some NV40 (68xx) as well as TSMC.

 

[ViRGE]

Maybe AMD didn't have full faith in GloFlo to pull off satisfactory 32nm yields and used IBM as a bit of a cushion. Now they are locked in for how ever many wafers. I don't think this is necessarily a bad thing, all eggs in one basket and such. I remember Nvidia used IBM for some NV40 (68xx) as well as TSMC.

Was it NV40? I thought it was NV35?

Anyhow, to tap IBM is a big deal. GPUs are largely synthesized, making it easier (note: not easy) to move those designs around. But APUs are still done by hand; to use a second fab means someone went through a lot of work to tweak the design for the fab.

 

[NTMBK]

AMD and IBM move into a closer partnership just as AMD announce that they are working on HSA to allow them to easily use different ISAs.

Which begs the question... why does everyone seem to think they're talking about ARM?

 

[Martimus]

Was it NV40? I thought it was NV35?

Anyhow, to tap IBM is a big deal. GPUs are largely synthesized, making it easier (note: not easy) to move those designs around. But APUs are still done by hand; to use a second fab means someone went through a lot of work to tweak the design for the fab.

While the link above suggests that IBM and GloFo are going to jointly use the new Fab in New York that GloFo built (meaning that it may just be semantics that IBM is manuafacturing some of the Piledriver chips), but IBM would probably be the easiest fab for AMD to adjust their design for. This is because GloFo and IBM use most of the same process technology.

 

[ultimatebob]

hm, i should ask my dad.

he works at ibm

Heh... unless he works at the fab they're producing these chips at, he probably doesn't have a clue about it. It's easy to forget how BIG of a company IBM is... they make everything from cash register software to mainframes.

 

[Dark Shroud]

Which begs the question... why does everyone seem to think they're talking about ARM?

Because right now ARM chips are a big deal and are the future of everything. At least that's what a lot of foolish people think.

 

[Meaker10]

Every FAB has a giant design document that tells you all the dimentions, quirks and limitations.

Some are hard fast rules, you can bend others.

Calling a fab "28nm" and "32nm" is MASSIVELY simplistic, it has it's place but you can't read too deeply into that number.

 

[Keysplayr]

Was it NV40? I thought it was NV35?

Anyhow, to tap IBM is a big deal. GPUs are largely synthesized, making it easier (note: not easy) to move those designs around. But APUs are still done by hand; to use a second fab means someone went through a lot of work to tweak the design for the fab.

Could be. Either way, one of those series was IBM fabbed.

 

[bononos]

Every FAB has a giant design document that tells you all the dimentions, quirks and limitations.

Some are hard fast rules, you can bend others.

Calling a fab "28nm" and "32nm" is MASSIVELY simplistic, it has it's place but you can't read too deeply into that number.

I was thinking of this earlier... When AMD spun off its foundry business to become more of a chip designer, was it a good idea in the long run?
Is there some synergy that comes when both design and production people can actually work together that AMD lost in the spinoff.

 

[nenforcer]

AMD and IBM have been manufacturing partners ever since AMD signed an agreement with IBM in 2001 to use Silicon on Insulator (SOI) techniques on their chips.

At that time AMD still manufactured the chips in their own foundries i.e. Dresden but used IBM's SOI manufacturing tech in the process.

This began with the Athlon XP chips and continues to this day through the Phenom 2 X6 Thuban's, if not AMD FX as well.

 

[formulav8]

SOI started with their 130nm process with Athlon 64 onward. There were no pre-AMD64 chips that used SOI. The 130nm Athlon XP's did Not use SOI.

 

[JustMe21]

Look at 3DFX. They were originally fabless and did well, then they bought STB and now they're gone.

 

[Idontcare]

Is there some synergy that comes when both design and production people can actually work together that AMD lost in the spinoff.

Yes. In foundries, the customization of process node specifics is much less flexible when it comes to design rules tradeoffs that can provide outsized benefits to the design team of a specific design project.

If you are vertically integrated (an IDM in semiconductor industry speak) with both the fabs and the design teams then process integration can make much more tailored design rule designs (and much later in the game) than can be accommodated in a foundry environment.

The other issue with fabless design houses is they are sharing profits. GloFo wants to make a profit on the wafers it sells AMD, and AMD in turn wants to make profit on reselling those chips.

When you are vertically integrated, like Intel, then as a business you can strategically align your cost structures solely with one end-goal in mind when it comes to profits, no extra middle-men in there trying to make profits at every step of the way.

It is really only an issue though when you have fabless design companies competing with IDM's. If you have fabless competing with fabless (AMD vs Nvidia) then the negatives of the foundry model are a wash as far as the net impact on competitiveness.

 

[nenforcer]

SOI started with their 130nm process with Athlon 64 onward. There were no pre-AMD64 chips that used SOI. The 130nm Athlon XP's did Not use SOI.

How about the 32nm Global Foundries produced AMD FX chips, are those still using SOI?

 

[notty22]

AMD Adds a Third Foundry for its Fusion APUs: Hello IBM

Reported by Theo Valich on Monday, February 6 2012 11:46 pm

At the recently held Financial Analyst Day held in Sunnyvale, California, AMD disclosed that the company added a third foundry in its manufacturing portfolio - IBM.



NEWS

After GlobalFoundries and TMSC, AMD executives admitted that the company will use IBM as a third foundry for its Fusion Accelerated Processing Units (APUs): TSMC manufactures 40nm E-Series, W-Series and Z-Series, while GlobalFoundries manufactures 32nm A-Series (codename: Llano). The manufacturing of standard CPUs (FX, Opteron) remains solely at GlobalFoundries domain.

​

Trinity, the successor of Llano will be manufactured using 32nm SOI process at both GlobalFoundries and IBM, presumably at IBM's East Fishkill facility in New York State. Two companies are not competitors though. IBM's East Fishkill is the location of former AMD CPU Silicon Design Team, which became GlobalFoundries CPU Silicon Design Team following the manufacturing spin-off in 2008. Following the issues with the manufacturing of Llano, it seems that AMD is moving forward to check GlobalFoundries processes with a different measurement.
This is not a surprising turn of events, since the manufacturing cluster in New York state is practically joined at hip: first revenue 32nm SOI chips manufactured at GlobalFoundries Fab 8 in Malta were for - International Business Machines, i.e. IBM.
The mantra that Rory P. Read, Chief Executive Officer of AMD was saying during his keynote speech was "execute". Remember, AMD lost a breakthrough contract with Apple in the first half of 2011 because the company could not execute and deliver a sufficient quantity of 32nm "Llano" chips. Same thing is with Lenovo, where Rory lead the company since its spun-off from IBM.
Can GlobalFoundries and IBM create enough Trinity chips to avoid the Llano debacle and put AMD inside the major players such as above mentioned Apple and Lenovo? Only time will tell.

​

Read more: http://vr-zone.com/articles/amd-add...usion-apus-hello-ibm/14759.html#ixzz1lk1QGFcr

 

[Phynaz]

16 week cycle time?

Jeez, no wonder Intel is kicking them every which way, it takes Intel a third of the time to produce a chip.

 

[Idontcare]

How about the 32nm Global Foundries produced AMD FX chips, are those still using SOI?

Yes.

AMD Adds a Third Foundry for its Fusion APUs: Hello IBM

Two companies are not competitors though.

Oh that's rich. I suspect their respective employees and shareholders would disagree

By this measure AMD and Intel aren't competitors either since they too cross-license all kinds of IP.

Hyperbole aside, what I suspect this means is that we are about 12-18 months away from an announcement of IBM selling its fishkill fab and CMOS development division to Global Foundries.

Its the only way to rationalize the continued R&D co-existence of GloFo and IBM in the fab ecosystem when they ARE foundry competitors in every sense of the word.

For the same reasons that TSMC is not part of the fab club, it makes no sense for Samsung, IBM, and GloFo to jointly develop their foundry process tech when doing so eliminates the very thing that differentiates foundries from one another in the first place.

But IBM could sell their fab while retaining foundry access and IP licensing revenue for their own hardware purposes. Then no conflict of interest, and GloFo can better align their process development timeline with the needs of their timeline sensitive customers (AMD at the moment).

 

[sm625]

There is an article floating around where "someone" at AMD is denying IBM is making chips for them.

 

[Meaker10]

I was thinking of this earlier... When AMD spun off its foundry business to become more of a chip designer, was it a good idea in the long run?
Is there some synergy that comes when both design and production people can actually work together that AMD lost in the spinoff.

Problem is, short term you sink. There was a good chance AMD would have gone under without spinning off global foundries.

Without the right investment you can't keep up with a company like Intel.

So the advantage of several companies fabbing in one place is that you share that cost.

 

[Meaker10]

Look at 3DFX. They were originally fabless and did well, then they bought STB and now they're gone.

They were always fabless, STB made the PCB and assembled the card, very different from fabbing the chip.

 

[jones377]

There is an article floating around where "someone" at AMD is denying IBM is making chips for them.

No surprise there since the original article was based on a false premise/misunderstanding on the part of the author.

 

[Borealis7]

just for the sake of completeness:

http://www.xbitlabs.com/news/other/...Denies_Building_Chips_at_IBM_s_Factories.html

IBM was just "consulting" according to AMD.

 

[bononos]

Problem is, short term you sink. There was a good chance AMD would have gone under without spinning off global foundries.

Without the right investment you can't keep up with a company like Intel.

So the advantage of several companies fabbing in one place is that you share that cost.

I don't know alot about how well AMD was managing on its fabs but I got the impression that AMD's cpu business was rather a small part of the entire enterprise probably generating less than 1/2 its revenue - 10yrs ago(?). So AMD could keep its old fabs chugging away making chips other than consumer cpus of its own design. I think sharing the cost goes even furthur than keeping the order pipeline filled. Now fabs are multi-billion investments and some fab companies are funded by sovereign wealth and have the support of govts. Even US-owned ones might possibly be beneficiaries of 'leaked' top notch research from the defense black budget (which includes thing like chip making R&D).