[TechEye] GlobalFoundries to buy IBM Semi

[NTMBK]

The Wall Street Journal has found that contract chipmaker GlobalFoundries (GloFo) is the leading candidate to buy IBM's semiconductor operations.
Several companies have been named as potential buyers including Intel and Taiwan Semiconductor Manufacturing.
However, the Wall Street Journal quoting unidentified sources as saying that GloFo was making Big Blue the best offer it could not refuse.

http://news.techeye.net/chips/globalfoundries-to-buy-ibm-semi

Should be a nice boost to GlobalFoundries' fab portfolio, and especially to their R&D.

 

[norseamd]

amd needs this

 

[Nothingness]

http://news.techeye.net/chips/globalfoundries-to-buy-ibm-semi

Should be a nice boost to GlobalFoundries' fab portfolio, and especially to their R&D.

Didn't they already share process R&D with IBM through Common Platform? I guess not all of IBM process R&D was shared, but I'm not sure what is being sold exactly.

Also I wonder what the impact will be on the third member of that alliance, Samsung.

 

[NostaSeronx]

IBM bought 20nm FDSOI from STMicro. The 20FD mentioned from GlobalFoundries is actually STMicros 14nm FDSOI process.

GlobalFoundries by buying IBM's foundries they get the unrestricted 20nm FDSOI.

Also I wonder what the impact will be on the third member of that alliance, Samsung.

Samsung and GlobalFoundries are more connected than IBM and Samsung. So, the partnership will most likely improve further since GlobalFoundries is required to backup Samsung.

 

[mrmt]

http://news.techeye.net/chips/globalfoundries-to-buy-ibm-semi

Should be a nice boost to GlobalFoundries' fab portfolio, and especially to their R&D.

The tech press is so incompetent that they can't sniff news even when it's just under their nose.

Globalfoundries proposal to IBM, according to the WSJ, was indeed the highest bidder, but there's a catch on it: GLF is trying to engage IBM in a kind of WSA, something that IBM won't accept without a multitude of covenants in the contract, if it accepts something at all. How much much of a handicap it is for your business to have a 3rd class unreliable foundry as your exclusive foundry partner? Big, way big, and a few hundred millions won't be enough to cover it. There are also some issues with IP from node development.

Another important bit: The deal isn't imminent, Intel is still negotiating with them.

 

[witeken]

So there isn't anything definitive to report?

 

[mrmt]

So there isn't anything definitive to report?

The only thing concrete is that TSMC is out of the race.

 

[NTMBK]

The tech press is so incompetent that they can't sniff news even when it's just under their nose.

Globalfoundries proposal to IBM, according to the WSJ, was indeed the highest bidder, but there's a catch on it: GLF is trying to engage IBM in a kind of WSA, something that IBM won't accept without a multitude of covenants in the contract, if it accepts something at all. How much much of a handicap it is for your business to have a 3rd class unreliable foundry as your exclusive foundry partner? Big, way big, and a few hundred millions won't be enough to cover it. There are also some issues with IP from node development.

Another important bit: The deal isn't imminent, Intel is still negotiating with them.

The WSA also binds both ways, remember- GloFo are required to provide a certain number of wafers on a suitable process, just as their customer is required to buy them. If IBM gets their predictions right on POWER sales, then a WSA shouldn't be a big issue- they get guaranteed wafers on the exact same process they would have been using if they hadn't spun off the fab. It's not like IBM's fab is going to suddenly regress to 28nm bulk just because they changed owner. And no other fab is offering a process suitable to IBM's needs- it's not like they could port POWER8 to TSMC's bulk 20nm SoC process.

The reason the WSA went so horrendously wrong for AMD is because they totally miscalculated demand for their x86 products. The Opteron business went into a terminal nosedive which it never recovered from, and the desktop business didn't do too much better. The only bright spark was Bobcat, which was on the wrong fab to help at all.

 

[NTMBK]

So there isn't anything definitive to report?

Pretty much. Sounds like negotiations are ongoing, and even if GloFo is in the lead now Intel could come back with a stronger bid. A more accurate headline would be "GlobalFoundries front runner to buy IBM Semi", but I'm just working with what TechEye give me.

I'm not really sure what Intel would do with it though- they can't fill the fabs that they've got. And Intel has in the past made it very clear that they won't fab for direct rivals, which IBM certainly is- Intel are targeting their POWER systems with high end Xeon E7 servers. Either Intel would have to reverse that policy and open up their fabs to IBM (at least the IBM originating fabs), or IBM would be without a suitable process for their high end POWER chips.

 

[jpiniero]

I'm guessing that's why Intel is interested - they would like to see POWER be discontinued.

 

[mrmt]

The WSA also binds both ways, remember- GloFo are required to provide a certain number of wafers on a suitable process, just as their customer is required to buy them.

That's not the kind of agreement AMD has with Globalfoundries, as you can verify to what happened with them in 2012: By that time they had Krishna and Wichita designs ready, and Globalfoundries didn't have their 28nm process ready. What happened with Globalfoundries? Nothing. AMD had to scrap the project and Globalfoundries didn't have to pay anything due to breaches.

Today GLF's 28nm process is still inferior to TSMC's 28nm process, and what's the result for AMD? Nothing, they still have to backport everything to GLF and stick with whatever GLF can provide for them. AMD's WSA binds the buyer to the foundry, but not the foundry to the buyer. The foundry can field whichever pie-in-the-sky process and AMD has to bite it.

I doubt that IBM would accept something even remotely along these lines, and this is what should have stalled the talks. IBM has no need to be tied to a 3rd rate foundry partner, especially when they have Intel on the other side of the table offering access to their process node, probably without strings attached.

If IBM gets their predictions right on POWER sales, then a WSA shouldn't be a big issue- they get guaranteed wafers on the exact same process they would have been using if they hadn't spun off the fab.

When you have a marketing falling 20% YoY, that is certainly an issue. In fact, one of the causes of IBM quitting this business is because sales are crashing. What's POWER sales in 5 years? Will we have another 20% drops in the next years or will the drain stop?

It's not like IBM's fab is going to suddenly regress to 28nm bulk just because they changed owner. And no other fab is offering a process suitable to IBM's needs- it's not like they could port POWER8 to TSMC's bulk 20nm SoC process.

TSMC is offering a suitable process. SUN T5 is manufactured with TSMC 28nm bulk and performs very well. Die size is withing acceptable limits (20% smaller than 16 core piledriver), clocks reasonably high (3.6GHz for the 16 core version) and it seems the customer is happy.

Intel is also open for business.

So I don't think Globalfoundries is the only game in town. In fact, it's the worst game in town.

The reason the WSA went so horrendously wrong for AMD is because they totally miscalculated demand for their x86 products.

Oh, so there's nothing wrong with one-sided take-or-pay charges, with the foundry delaying nodes and not being subject to sanctions of any kind and mandatory backportings?

PM me please, I have contract that might suit us both.

 

[NTMBK]

PM me please, I have contract that might suit us both.

Actually, there is a bridge I've had my eye on...

 

[mrmt]

I'm guessing that's why Intel is interested - they would like to see POWER be discontinued.

Intel buying IBM foundry certainly wouldn't kill POWER, it might ensure its survival by a bit longer, especially if IBM can access Intel bleeding edge process to build Power.

POWER is dying because of the economics of the business, that won't change much regardless of who buys IBM fabs.

 

[NostaSeronx]

mrmt.

The design flows for GlobalFoundries are shared with Samsung. Any problems GlobalFoundries would be having Samsung will have too. In any case that Samsung's foundry fails it is required that GlobalFoundries handles the orders. They must be synchronized in the event that it does happen so the process is equal.

 

[Jimzz]

Intel buying IBM foundry certainly wouldn't kill POWER, it might ensure its survival by a bit longer, especially if IBM can access Intel bleeding edge process to build Power.

POWER is dying because of the economics of the business, that won't change much regardless of who buys IBM fabs.

POWER is dying because IBM laid off all its good engineers and tried to use ones from India.

I know someone who works inside IBM and he has said that India is next to get laid off as they do next to nothing. Now that there is not enough good engineers left to pick up after them they are finally being caught as do nothings in india.

 

[mrmt]

The design flows for GlobalFoundries are shared with Samsung. Any problems GlobalFoundries would be having Samsung will have too. In any case Samsung's foundry fails it is required that GlobalFoundries handles the orders. They must be synchronized in the event that it does happen so the process is equal.

What did I say about Samsung?

 

[NostaSeronx]

What did I say about Samsung?

I am just pointing out that GlobalFoundries wasn't the one at fault. Since, Samsung S1 & S2 and GlobalFoundries Fab 1 & Fab 8 are virtually the same foundry. Any yield issue would be neglected by having four foundries rather than just one foundry.

The reality of the matter Wichita and Krishna was canned because the whole engineer team left for Samsung. AMD at the time was afraid of Samsung which controls the process that GlobalFoundries uses. It is not GlobalFoundries issue that AMD didn't want to participate in the GlobalFoundries and Samsung party.

 

[ShintaiDK]

Not a surprise since IBM is irrelevant and havent much to offer. I just hope for GloFo that they dont pursue the same failed goals IBM had.

And POWER was on its path to extinction long ago.

The future IBM only does in software/service.

 

[mrmt]

POWER is dying because IBM laid off all its good engineers and tried to use ones from India.

I know someone who works inside IBM and he has said that India is next to get laid off as they do next to nothing. Now that there is not enough good engineers left to pick up after them they are finally being caught as do nothings in india.

Are you implying that the latest POWER chips fell short of the targeted parameters? Because this is what the kind of resource management problems you are implying would warrant, but I never saw anything about that anywhere.

I saw some shift from mainframes to x86 servers that accelerated *a lot* after Westmere EX, and now with IVB-EX we're set for another round of share drops, but I saw this more of a consequence of the general business model than of engineering deficiencies of POWER. Are you able to expand a bit on the subject?

 

[mrmt]

The reality of the matter Wichita and Krishna was canned because the whole engineer team left for Samsung. AMD at the time was afraid of Samsung which controls the process that GlobalFoundries uses. It is not GlobalFoundries issue that AMD didn't want to participate in the GlobalFoundries and Samsung party.

Kaveri was also delayed and Globalfoundries had 0 28nm products to talk about before AMD chips, except for that rock-something which is nothing more than a small, insignificant production run. There were problems with their 28nm chips, that there were problems inside AMD is just another story.

 

[Jimzz]

Are you implying that the latest POWER chips fell short of the targeted parameters? Because this is what the kind of resource management problems you are implying would warrant, but I never saw anything about that anywhere.

I saw some shift from mainframes to x86 servers that accelerated *a lot* after Westmere EX, and now with IVB-EX we're set for another round of share drops, but I saw this more of a consequence of the general business model than of engineering deficiencies of POWER. Are you able to expand a bit on the subject?

I can't say what other stuff I know but power right now is at a crossroads. Either step up and hire some engineers so the work that is needed is done or its going to die.

POWER's newer cores have some other issues that have not come out yet, at least other than rumors... many true. That and next layoffs are supposed to be announced by the end of April. That makes me think IBM has enough bidders they don't need to keep working on design just finish up old contracts and let the new owners decide.
I wish I could say more but don't want to get my friend in trouble, at least until he gets his pink slip.

 

[NostaSeronx]

Kaveri was also delayed and Globalfoundries had 0 28nm products to talk about before AMD chips

All projects were halted when Dirk Meyer was booted out. It wasn't till Rory Read came into the picture when the projects were resumed or cancelled. Kaveri's tapeout on 28nm-SHP was delayed by one year. Krishna and Wichita had turbo core, Kabini and Temash did not. If you go further you would find out Krishna and Wichita were 14h processors. 16h is not the same architecture that was planned to be on 28nm SHP.

, except for that rock-something which is nothing more than a small, insignificant production run. There were problems with their 28nm chips, that there were problems inside AMD is just another story.

ARM companies do not normally use leading technologies.

 

[NTMBK]

All projects were halted when Dirk Meyer was booted out. It wasn't till Rory Read came into the picture when the projects were resumed or cancelled. Kaveri's tapeout on 28nm-SHP was delayed by one year. Krishna and Wichita had turbo core, Kabini and Temash did not.

Uh huh, sure, and Kaveri didn't clock faster than 2GHz on a 28nm SOI process. Provide some sources, or quit making stuff up.

 

[NostaSeronx]

Provide some sources, or quit making stuff up.

Most if not all my sources can be acquired by google search engine.

http://lmgtfy.com/?q=Kaveri's+tapeout+severely+delayed
http://lmgtfy.com/?q=Krishna+and+Wichta+turbo+core

 

[witeken]

Referring to Google isn't a source...