News IBM suing GlobalFoundries

[NTMBK]

IBM accused of chasing 'a quick payday' with $2.5B damages against AMD's OG chip maker

GlobalFoundries claims IBM is only now stalking it for damages over an alleged contract violation because it's going public.

www.pcgamer.com

One-time AMD fab limb, GlobalFoundries, is laying the litigation down on IBM to escape the $2.5 billion damages Big Blue is trying to squeeze out of it just ahead of its initial public offering (IPO). That, it states, is the prime driver behind IBM's recent demands with the company only looking for "a quick payday."

I can see IBM's side of this. They bet on GlobalFloundering for their next generation products, paid them a huge chunk of money... and then GloFo gave up on leading edge. I'd be pissed too.

 

[Jimzz]

Yea IBMs next gen design was based on GFs 7nm. When GF dropped their 7nm it put IBM at least a full year behind on its next power chip.

With that said this case will come down to what was in the contract exactly. Could go either way without knowing the exact language and what went on behind the scenes.

 

[NTMBK]

Yea IBMs next gen design was based on GFs 7nm. When GF dropped their 7nm it put IBM at least a full year behind on its next power chip.

With that said this case will come down to what was in the contract exactly. Could go either way without knowing the exact language and what went on behind the scenes.

I hope it goes to court- like the recent Epic/Apple fight, we might learn all sorts of fun behind the scenes detail.

 

[DrMrLordX]

IBM's position sort of makes sense, never mind that IBM did find a home for POWER10 |(Samsung). If GF didn't leave themselves an out clause in the contract, they may need to pay, though if penalties aren't laid out in the contract, it'll have to be based on actual damages to IBM's CPU business. It's not clear that IBM can show $2.5 billion in lost sales due to GF throwing in the towel on 7nm.

 

[NostaSeronx]

IBM partnered with Samsung in 2016 for POWER10 & POWER11. Effectively, ditching GlobalFoundries with AMD at the same time but to different fabs.

There was no trust that Fab8 could provide specialized FinFETs for both AMD & IBM. So, IBM & AMD broke off pretty early.

 

[moinmoin]

IBM partnered with Samsung in 2016 for POWER10 & POWER11. Effectively, ditching GlobalFoundries with AMD at the same time but to different fabs.

There was no trust that Fab8 could provide specialized FinFETs for both AMD & IBM. So, IBM & AMD broke off pretty early.

I wonder what the actual timeline was. In October 2014 IBM gave GloFo its fabs including its staff and even paid $1.5 billion on top as well as access to IP. Only in 2018, close to when 7nm was supposed to be launched, GloFo notified both AMD and IBM that it would no longer pursue 7nm at all. Both AMD and IBM obviously saw the writing on the wall early enough to prepare extensive contingency plans. But I always felt to both of them it must have been clear early on that GloFo not only won't be able to deliver as a second source for 7nm but even fail at that altogether.

 

[NostaSeronx]

I wonder what the actual timeline was. In October 2014 IBM gave GloFo its fabs including its staff and even paid $1.5 billion on top as well as access to IP. Only in 2018, close to when 7nm was supposed to be launched, GloFo notified both AMD and IBM that it would no longer pursue 7nm at all. Both AMD and IBM obviously saw the writing on the wall early enough to prepare extensive contingency plans. But I always felt to both of them it must have been clear early on that GloFo not only won't be able to deliver as a second source for 7nm but even fail at that altogether.

As detailed I can find.

10LP was completed by GloFo in 2015
7LP was completed by GloFo in 2016
test-site completed and verified performance/power/etc.

IBM only set POWER10 at 10LP at the time in 2014-2015.
However, they had a side project of porting POWER8/9 Scale-out from then GloFo-owned 22nm/14nm to Samsung's 10nm/7nm nodes as OpenPower SO by early 2016. With something escalating and IBM moving POWER10/POWER11 away from original plans at GloFo to new plans at Samsung, shortly after within the same year.

The decision to cancel to 7nm was actually bad faith from AMD/IBM not willing to fab at a second foundry. So, they in fact ditched early and only after GlobalFoundries killed 5LP and realized when it failed to bring them back, whelp they aren't coming back. Did they finally kill off 7LP/7LP+/3LP(also called NextGen).

I can't find the pdf with GlobalFoundries crossing out 5LP and having 3LP in a similar color next to it. Which was a threat to AMD/IBM which went ignored I guess. Since, AMD and IBM never bothered with any of the "Leading Performance" nodes.

 

[Schmide]

I'm really surprised IBM would put itself through this. It's one thing to be a customer and be promised something, then when a failure to deliver happens, an attempt is made to recoup losses. Another to be a former partner with access and pretend not to know the risks involved with such technology.

I would not be surprised if this disappears before discovery.

Value of the game 2.5billon vs airing ones dirty laundry.

 

[gdansk]

Especially odd when one considers the previous agreement IBM made with GF. I wonder what the details are and why GF had a bad falling out with IBM but not AMD.

 

[jpiniero]

Especially odd when one considers the agreement IBM made with GF. GF can easily argue that they took over the facility and found the work even more difficult and expensive than IBM had lead them to believe.

I think it's pretty clear that AMD weaseling their way out of being forced to use GloFo 7 is what led to them giving up.

 

[gdansk]

I think it's pretty clear that AMD weaseling their way out of being forced to use GloFo 7 is what led to them giving up.

I wouldn't be surprised but still that is not entirely GF's fault. If GF expected IBM and AMD as customers but ended up with only IBM it seems to favor cancellation. Presumably IBM would be entitled to some compensation for the portion they paid to development but I doubt it'd be $2.5 billion. I'm also curious why IBM waited so long to begin this dispute.

 

[moinmoin]

I wonder what the details are and why GF had a bad falling out with IBM but not AMD.

Was IBM a significant customer to GloFo already before the deal?
AMD was bound to the WSA anyway and until 2017 could been considered being closer to bankruptcy than becoming a major customer for 7nm.

I think it's pretty clear that AMD weaseling their way out of being forced to use GloFo 7 is what led to them giving up.

AMD had to comply with the WSA in any case. The only way AMD as the sole 7nm customer for GloFo could have saved 7nm would have been WSA terms even more in favor of GloFo, and even that would have been no actual guarantee due to AMD walking a fine line close to bankruptcy for several years.

Edit: Also note how wafer production is very cyclical to a point that some might miss that it's indeed slowly growing overall (this is only TSMC as the pure play foundry market leader, taken from wikichip):

GloFo probably figured that this more stagnating looking growth curve isn't worth the investment into leading nodes (and then after the last dip the curve changed into a linearly growing one...).

 

[NostaSeronx]

GlobalFoundries appears to be pushing for RFSOI/FDSOI at Malta.

Soitec and SunEdison Enter Into Patent License Agreement

/PRNewswire/ -- Soitec (Euronext) and SunEdison, Inc. (NYSE: SUNE) announced today that they have entered into a patent cross-license agreement relating to...

www.prnewswire.com

This carried over to GlobalWafers

GlobalFoundries and GlobalWafers Partnering to Expand Semiconductor Wafer Supply | GlobalFoundries

Long-term $800 million supply deal includes $210 million capital expansion, creation of more than 75 new jobs in Missouri and […]

globalfoundries.com

"...Today announced an $800 million agreement to add 300mm silicon-on-insulator (SOI) wafer manufacturing and expand existing 200mm SOI wafer production at GWC’s MEMC facility in O’Fallon, Missouri."

GloFo needed an uncertain amount of funds to Fab8 to support 15k wph of 7LP/7LP+. Which they didn't get nor would have gotten with IBM/AMD ditching.

However, for RFSOI and FDSOI they didn't need such funds.
Having no funding nor support from IBM/AMD, it makes sense for them to cancel further FinFET nodes. Since, combined the RFSOI+FDSOI makes more money faster overall.
45RF => >$1B (first customer 2018)
22FDX => >$4.5B (first customer 2018)
14LPP/12LP => >$4B (first customer 2015)

Even if IBM wins I do not think it will cause a dent to the SOIs, but might be the nail that kills off existing FinFETs at Fab8. There has been a couple times GlobalFoundries cut capacity for 14LPP/12LP already. $2.5B going away from Fab8 will just be them killing FinFETs. Since, customer drainage has been going on for a while. 14LPP/12LP/12LP+ can only go so far with 8LPP/8LPA, 7LPP/5LPP, 7FF/6FF/5FF, etc. I believe SMIC got some of GloFo's 14nm customers when GloFo killed Chinese access to 14LPP/12LP.

Far post-pivot:
Research and development of 45nm FD-SOI...
US,NY,Malta
Nov 2019+
Senior integration Engineer, 45nm MOL process integration

^-- I believe this node will replace 45RF, later on, which is a PDSOI node. While there appears to several specialty versions of 45FD; UHV, Photonics, RF, etc.

 

[DrMrLordX]

Was IBM a significant customer to GloFo already before the deal?

GF produced POWER8 and POWER9. At least.

 

[beginner99]

I wouldn't be surprised but still that is not entirely GF's fault. If GF expected IBM and AMD as customers but ended up with only IBM it seems to favor cancellation.

Which is ironic in hindsight because their 7nm factories would be running at full blast right now with GF being able to charge almost whatever they want. Eg. it would be a gold mine.

 

[moinmoin]

Which is ironic in hindsight because their 7nm factories would be running at full blast right now with GF being able to charge almost whatever they want. Eg. it would be a gold mine.

GloFo clearly bet on the wrong horse at the wrong time.

 

[NostaSeronx]

I do not think GlobalFoundries would be able to charge almost whatever they want. Nor, would they be turning out gold with their FinFETs. Waiting for customers to appear wouldn't look good.

Would you fab here? (equal or less than 14nm includes a fictional 7nm, btw)


or here?


If 7LP came, without the help of AMD/IBM being initial customers. Then, it would have been immediately depleted on a capacity front regardless. No customers = no capacity, which is the same policy UMC does.

SMIC bonus:


Not very stable, and they also have combined 28nm and 14nm. Which isn't usually a good sign.


I doubt GlobalFoundries would be making any dough from 7LP especially on the investment. Like what $40-$60B for 7LP/7LP+/5LP/3LP, and revenues of 14LP/12LP/12LP+ have yet to achieve a return on investment(>$10B spent/>$4B returned). So, financially it would have been a deathtrap, but at least someone else would have owned the fab.

On the prior post, there was 45nm FDSOI;
45SPCLO => MPW4C00 April 2021
45RFE => MPW4E00 October 2021

Community & Services | GLOBALFOUNDRIES

GF Partner Community & Services GF Partner Community Manufacturing Services

globalfoundries.com

"45SPCLO/45RFE (45nm FD-SOI)" compared to 45CLO/45RF which is PD-SOI.
Should have checked the MPW service before I made that post.

 

[eek2121]

GlobalFoundries appears to be pushing for RFSOI/FDSOI at Malta.

Soitec and SunEdison Enter Into Patent License Agreement

/PRNewswire/ -- Soitec (Euronext) and SunEdison, Inc. (NYSE: SUNE) announced today that they have entered into a patent cross-license agreement relating to...

www.prnewswire.com

This carried over to GlobalWafers

GlobalFoundries and GlobalWafers Partnering to Expand Semiconductor Wafer Supply | GlobalFoundries

Long-term $800 million supply deal includes $210 million capital expansion, creation of more than 75 new jobs in Missouri and […]

globalfoundries.com

"...Today announced an $800 million agreement to add 300mm silicon-on-insulator (SOI) wafer manufacturing and expand existing 200mm SOI wafer production at GWC’s MEMC facility in O’Fallon, Missouri."

GloFo needed an uncertain amount of funds to Fab8 to support 15k wph of 7LP/7LP+. Which they didn't get nor would have gotten with IBM/AMD ditching.

However, for RFSOI and FDSOI they didn't need such funds.
Having no funding nor support from IBM/AMD, it makes sense for them to cancel further FinFET nodes. Since, combined the RFSOI+FDSOI makes more money faster overall.
45RF => >$1B (first customer 2018)
22FDX => >$4.5B (first customer 2018)
14LPP/12LP => >$4B (first customer 2015)

Even if IBM wins I do not think it will cause a dent to the SOIs, but might be the nail that kills off existing FinFETs at Fab8. There has been a couple times GlobalFoundries cut capacity for 14LPP/12LP already. $2.5B going away from Fab8 will just be them killing FinFETs. Since, customer drainage has been going on for a while. 14LPP/12LP/12LP+ can only go so far with 8LPP/8LPA, 7LPP/5LPP, 7FF/6FF/5FF, etc. I believe SMIC got some of GloFo's 14nm customers when GloFo killed Chinese access to 14LPP/12LP.

Far post-pivot:
Research and development of 45nm FD-SOI...
US,NY,Malta
Nov 2019+
Senior integration Engineer, 45nm MOL process integration

^-- I believe this node will replace 45RF, later on, which is a PDSOI node. While there appears to several specialty versions of 45FD; UHV, Photonics, RF, etc.

If you don’t make the investment, you don’t get the business. They are in an industry where investment is a necessity. If they had a 7nm process today, I guarantee you they would have customers. They should strongly consider a 5nm process.

 

[NTMBK]

If you don’t make the investment, you don’t get the business. They are in an industry where investment is a necessity. If they had a 7nm process today, I guarantee you they would have customers. They should strongly consider a 5nm process.

They've fallen off the wagon- it would be very hard to catch up now.

 

[NostaSeronx]

If you don’t make the investment, you don’t get the business. They are in an industry where investment is a necessity. If they had a 7nm process today, I guarantee you they would have customers. They should strongly consider a 5nm process.

Probably has to wait for after 12FDX in 2023/2024. Since, GlobalFoundries is not a slave to device sizes.

I am also interested in what changes they put in 12FDX btw.
"We have 12FDX here and we know we can do the device – we are actually pretty close to the performance targets already.
We have been doing work on 12FDX here in NY for over a year. We have working devices and they’re not far off the performance target..."
- February 2018

"The 12FDX program is alive and well. ... we are about 60 percent done. We are optimizing the process."
- September 2020

Ocean12 ends December 31st 2021. So, the biggest research project for 12FDX is going to be done soon. I wonder how accurate the 2023/2024 timeline is... since 12FDX has features that 22FDX/22FDX+ can't do. Functionally, 12FDX replaces 7LP on the roadmap anyway. I have checked, Leti still has 12FDX using 10nm performance boosters.

How much performance is it going to have?

-Freq of SLVT@Vnom it is 6% faster than 7nm.
-Perf-same power it is 10% worse than 7nm.
-Power-same perf it is 7% worse than 7nm.
-# of good dies per wafer it is 18% worse than 7nm.
-Masks used it is just 53 which is 10 less than 14LPP and 35 less than 7LP.

---
10FDSOI boosters allowed 64CPP/48MPP FDSOI to be:
NMOS +110% mobility over 64CPP/48MPP FinFET
PMOS +90% mobility over same.
Similar static power at same delay.
20% lower dynamic power at same delay.

There is a bit of a massive boost in performance with any FDSOI at 64CPP. However, there is no guarantee for a 64CPP option on 12FDX.
~~

It is weird that it was basically 100% done just waiting to hit performance targets in 2017-2018. To being 60% done with no early production in sight. So, maybe they will make 12FDX a quasi-10nm node.

 

[darkswordsman17]

GloFo clearly bet on the wrong horse at the wrong time.

Which is why them arguing that IBM is only after a quick buck ironic. GF cancelled 7nm because they weren't going to get a quick enough ROI on it and literally said they were seeking more profits short term. If I'm IBM, I use GF's own words against it.

It was an exceptionally dumb shortsighted move that I still cannot figure out what they were thinking. The major development was basically done and they were ready to start producing soon, so they just flushed away all the return for all that money they'd spent. I could see them abandoning anything major beyond 7nm, but to abandon it when they'd already done most of the hard work was so senseless that the only thing I can think of is that it was an attempt to self-sabotage to seek a deal. Maybe they were hoping on of the other major players, Apple, Nvidia, or Intel would swoop in and offer to pony up to get it running in exchange for production? And maybe that's why IBM is suing them, that GF tried to leverage all the development money IBM had spent to get them to spend more so as to not lose anything? Perhaps they pulled that with both IBM and AMD and they both called their bluff and said "ok, fine, we'll just go elsewhere".

Which, sure hindsight is 20/20 but can you imagine how much GF is kicking themselves now, between the chip shortages, the tariffs, TSMC moving to the US (and getting deals to do so)?

 

[Hitman928]

Which is why them arguing that IBM is only after a quick buck ironic. GF cancelled 7nm because they weren't going to get a quick enough ROI on it and literally said they were seeking more profits short term. If I'm IBM, I use GF's own words against it.

It was an exceptionally dumb shortsighted move that I still cannot figure out what they were thinking. The major development was basically done and they were ready to start producing soon, so they just flushed away all the return for all that money they'd spent. I could see them abandoning anything major beyond 7nm, but to abandon it when they'd already done most of the hard work was so senseless that the only thing I can think of is that it was an attempt to self-sabotage to seek a deal. Maybe they were hoping on of the other major players, Apple, Nvidia, or Intel would swoop in and offer to pony up to get it running in exchange for production? And maybe that's why IBM is suing them, that GF tried to leverage all the development money IBM had spent to get them to spend more so as to not lose anything? Perhaps they pulled that with both IBM and AMD and they both called their bluff and said "ok, fine, we'll just go elsewhere".

Which, sure hindsight is 20/20 but can you imagine how much GF is kicking themselves now, between the chip shortages, the tariffs, TSMC moving to the US (and getting deals to do so)?

From another article I read, GF also sold and licensed a bunch of patents that IBM had given them in the deal as well as sold off a bunch of equipment before officially cancelling 7 nm. It made it seem like GF was acting in bad faith from the beginning of the deal, but obviously that's IBM's side of the story. I have to say though, having worked somewhat closely with GF in the past, they are by far my least favorite fab to deal with and I wouldn't be surprised if they did IBM dirty from the beginning of all this.

 

[NostaSeronx]

Mubadala was fine with dropping $10B here and there for awhile.

I think however as GlobalFoundries completely blundered 14LPP and killed off plans for Fab 1 Module 2 sharing bulk nodes with Fab 8. Made them hesitant to fund GlobalFoundries anymore.

So, $10B for 20LPM/14XM ramp at Fab8. Which GlobalFoundries took that and spun it for development of 10nm FinFETs. Then, they cancel the 10nm node, for a 7nm node and GlobalFoundries clearly alone... only spends >$2 billion for 7nm development.

To get Fab 8 Module 2 up and running they needed at least $10 billion at minimum, for 40-60 kwpm of 7LP/7LP+/5LP/3LP.
To expand capacity at Fab 8's existing module it would need $4B-$5B to even get 7LP/7LP+/5LP/3LP to 10-15 kwpm.

I'm pretty sure it wasn't so much GlobalFoundries killing the node. Rather Mubadala killing the stupidity with lack of investment. Basically, kicking the funds to USG instead, if you(USG) aren't going to fund, we(MIC) aren't going to fund.

Anyone expecting FinFETs at Fab8 to make more money than Fab1's FDSOI and Fab7's RFSOI needs the ole' lack of funds treatment.

---
There is also Fab 1G/Fab 7H plans going about in the rumor mills(investor forums/glofo-related forum boards). If they start developing those new fabs now or a little later from now(+4Q). They coincide with 12FDX volume production.

 

[SarahKerrigan]

GF produced POWER8 and POWER9. At least.

Not before the deal they didn't. Power8 was built on IBM's 22nm SOI process.

 

[DrMrLordX]

Not before the deal they didn't. Power8 was built on IBM's 22nm SOI process.

32HP (I guess that was POWER7+?)

While that is true, it doesn't invalidate my statement.