[Anandtech]: GlobalFoundries Stops All 7nm Development !!

[DrMrLordX]

And let's not forget, GF has had major problems with execution. So, great, their 22FDX is finally decent. Yay, I guess. In the meantime we've had Nosta telling us that it was going to come raging out and squash 16/14FF for years, and the industry is moving on to 7FF now. Its here now, not "on the roadmap".

What do you expect? Nothing @NostaSeronx predicts ever happens. Ever. So if you want to know what will NOT happen in the future (and what is not happening in the present), you know where to go . . .

 

[chrisjames61]

Samsung explains it a bit better:

Essentially, the FinFET back-end will be reused for the next FDSOI node. Basically, Samsung could do (mature back-end from):
28FDS -> 18FDS(14LPP) -> 15FDS(11LPP) -> 11FDS(7LPP) -> 8FDS(4GAA)

This is to make the FDSOI node cheaper than the FinFET node at initial consumption. As FDSOI will always be cheaper than FinFET @ same node.

In all the years you having making these predictions I have never, ever seen one pan out as you state.

 

[dark zero]

I think he meant all post-12nm chips.

Indeed. FD SOI is decent but it might be ideal for low cost mobile chips like the Mediatek Helio A or P series, but beyond that is hard to see something advanced.

And I hardly see AMD going TSMC considering Apple Qualcomm and NVIDIA using that process and maybe taking most of the production for them.

They need a backup and Samsung is the indicated for that job.

 

[Dayman1225]

And I hardly see AMD going TSMC

AMD literally said all their 7nm portfolio will be fabbed at TSMC.

 

[dark zero]

AMD literally said all their 7nm portfolio will be fabbed at TSMC.

Indeed, but if the competition pays more for the priority of their products?

They will be royally screwed up

 

[french toast]

Indeed. FD SOI is decent but it might be ideal for low cost mobile chips like the Mediatek Helio A or P series, but beyond that is hard to see something advanced.

And I hardly see AMD going TSMC considering Apple Qualcomm and NVIDIA using that process and maybe taking most of the production for them.

They need a backup and Samsung is the indicated for that job.

^^This.

 

[moinmoin]

Samsung due to its insistence to only want to start 7nm with EUV is a long time off still. If you want to go 7nm TSMC is your only choice for now. And AMD is among the bigger customers for TSMC already as well as a relatively early adopter of TSMC's 7nm (possibly their first on 7nm HP).

 

[french toast]

Samsung due to its insistence to only want to start 7nm with EUV is a long time off still. If you want to go 7nm TSMC is your only choice for now. And AMD is among the bigger customers for TSMC already as well as a relatively early adopter of TSMC's 7nm (possibly their first on 7nm HP).

But AMD cancelled a product before Q4 tapeout, if TSMC doesn't have the capacity then what's stopping Samsung taking it?
Exynos is going to be on a leading edge node, I don't know about 7nm, but certainly 8nm LPP...10nm LPU.

 

[moinmoin]

But AMD cancelled a product before Q4 tapeout, if TSMC doesn't have the capacity then what's stopping Samsung taking it?
Exynos is going to be on a leading edge node, I don't know about 7nm, but certainly 8nm LPP...10nm LPU.

We don't know whether TSMC doesn't have sufficient capacity. We don't know what AMD intended to tape out at GloFo, it may have been the same as Rome/"Zeppelin 2" at TSMC in which case there actually is no cancelled product, just one instead two versions of the same die. The two versions were likely not a necessity due to capacity but due to the WSA. Samsung may be nice to have in due time, but preparing a tapeout takes time as well so AMD may as well wait for EUV at both foundries (which I assume to match the planned launch time for Zen 3 based chips).

 

[TurtleCrusher]

This is horrible for everyone who works in the industry. Cutting edge US fabs are disappearing.

Welp, no New York trips anymore!

 

[dark zero]

Samsung due to its insistence to only want to start 7nm with EUV is a long time off still. If you want to go 7nm TSMC is your only choice for now. And AMD is among the bigger customers for TSMC already as well as a relatively early adopter of TSMC's 7nm (possibly their first on 7nm HP).

Indeed, but I know that AMD is not idiot and they know that going on one fab is suicidal... they paid hard with Bulldozer torture....

So, maybe Samsung will help AMD in some other designs (Raven Ridge succesor or a Core m competitor?)

 

[PeterScott]

Indeed, but I know that AMD is not idiot and they know that going on one fab is suicidal... they paid hard with Bulldozer torture....

So, maybe Samsung will help AMD in some other designs (Raven Ridge succesor or a Core m competitor?)

Yeah, being essentially TSMC exclusive, is what destroyed NVidia. Oh, wait...

 

[dark zero]

Yeah, being essentially TSMC exclusive, is what destroyed NVidia. Oh, wait...

Not really, but nVIDIA is GPU only... on CPU they are pretty much alive only thanks to Nintendo.

Apple is on their own way... so is a wild card.

Qualcomm is TSMC only it seems. And if something happens on the fabs, they might be royally screwed.

Mediatek is another Wild Card... what they are doing with Mediatek with their 22nm process???

 

[PeterScott]

Not really, but nVIDIA is GPU only... on CPU they are pretty much alive only thanks to Nintendo.

So being single fab isn't a problem for GPU makers, just CPU makers?

 

[dark zero]

So being single fab isn't a problem for GPU makers, just CPU makers?

Nope. Is a problem for a company that goes on both bussinesses.
Obviously on TSMC they might develop their most important products, but on the ones which are the lower tiers, but that sells more and requires high volumes, is risky to go on just one.

The only that can do that feat is Intel.

 

[PeterScott]

Nope. Is a problem for a company that goes on both bussinesses.
Obviously on TSMC they might develop their most important products, but on the ones which are the lower tiers, but that sells more and requires high volumes, is risky to go on just one.

The only that can do that feat is Intel.

Why. I don't see any logic to the argument. They simply want to be on the best fab. It doesn't matter if you build SoCs, CPUs and/or GPUs.

NVidia has a huge thriving business selling GPUs from the highest end to the lowest end, all on TSMC for many years.

AMD already has the their low end stuff at GF for the foreseeable future, TSMC for the leading edge. So they already have to deal with two fabs. Most other companies stick with one.

I see just ZERO chance Samsung will enter the picture in the foreseeable future.

 

[itsmydamnation]

I think the only way you will see amd @ Samsung is if they have a big lead in terms of euv.

But amd is fine, IBM on the other hand, massive Ed ram caches really high metal stacks, paying GF billions and doing personal transfers, I wouldn't be surprised to see a law suite .....

 

[NostaSeronx]

But amd is fine, IBM on the other hand, massive Ed ram caches really high metal stacks, paying GF billions and doing personal transfers, I wouldn't be surprised to see a law suite .....

IBM will only need to wait a bit after 12FDX.

"While ETSOI technology is initially offered as a performance booster for low-power applications at 28-nm, it is very likely that it can cover many of the high-performance applications in 20-nm and beyond."
-> Extremely thin SOI for system-on-chip applications - 9-12 Sept. 2012
It's very important that the above 20-nm and beyond is taken with this:
-> Strain engineered extremely thin SOI (ETSOI) for high-performance CMOS - 12-14 June 2012
Which details what is needed: "High-performance strain-engineered ETSOI devices are reported. Three methods to boost the performance, namely contact strain, strained SOI(SSDOI) for NFET, and SiGe-on-insulator (SGOI) for PFET are examined. "

December 2015 - March 2016 via Ultra-thin body & buried oxide SOI substrate development and qualification for Fully Depleted SOI device with back bias capability
which shows this: https://i.imgur.com/fuNQyf4.jpg

<14nm UTBOX15s (strained) for just SSDOI

Then the date is inside: FD-SOI Technology by Bich-Yen Nguyen, Soitec
https://i.imgur.com/ibzT0OU.png

Beyond 12FD: Both SSDOI and SGOI
The date is pretty important as it is after: https://www.globalfoundries.com/new...nter-long-term-supply-agreement-fd-soi-wafers
Meaning the above roadmap is the WSA GlobalFoundries-SOITEC roadmap.

GlobalFoundries post-12FDX whether it be 12FDX+ or 7FDX. It will be High Performance in the eyes of IBM.

 

[iBoMbY]

Why. I don't see any logic to the argument. They simply want to be on the best fab. It doesn't matter if you build SoCs, CPUs and/or GPUs.

The main problem is the capacity. Now everyone is fighting for TSMC's 7nm capacity: Apple, Huawei (Kirin 980 running right now), AMD, NVidia (for the next gen), Tachyum Prodigy (okay, probably never getting that far), IBM (probably), and probably a dozen others.

Edit: Broadcom is one of the others.

Edit2: Also Qualcomm plans on using TSMC 7mn.

 

[PeterScott]

The main problem is the capacity. Now everyone is fighting for TSMC's 7nm capacity: Apple, Huawei (Kirin 980 running right now), AMD, NVidia (for the next gen), Tachyum Prodigy (okay, probably never getting that far), IBM (probably), and probably a dozen others.

Edit: Broadcom is one of the others.

Edit2: Also Qualcomm plans on using TSMC 7mn.

With AMD now "all in" on 7nm and all of it with TSMC, AMD will now be one of the bigger TSMC customers, and they seem to have their design in pretty early on 7nm as well. It's really the smaller players that might find themselves squeezed and risk going to Samsung.

Every Fab you work with takes dedicated resources; more staff and money, and AMD is already committed to two Fabs, they won't be adding a third anytime soon.

 

[Gideon]

New Gary Patton (CTO of GF) interview about the 7nm abandonment:
https://spectrum.ieee.org/nanoclast...n-why-the-company-abandoned-the-bleeding-edge

A Special highlight regarding the "wunderWaffe" 12FX, certain people are pushing non-stop as the "new 7nm":

IEEE Spectrum: Your next step in fully-depleted silicon-on-insulator technology is a 12-nm technology called 12FX. At one point it seemed like it would perform as a competitor to 7-nm technology. Is that still the case?

Gary Patton: I don’t think it really competes. We got a little confused ourselves last year, to be honest regarding competition with 7 nm. Really it’s a different market place. It’s really about customers who decided to go the FDSOI path who are looking for that blend of power, performance, and cost with integration of RF capability, for example. They need to take those products that they’re taping out with us today at 22-nm FDSOI. And at the right time, probably three years from now be able to tape out a product at 12-nm.

Oh oh, and this non-answer IMO tells a whole lot more by what it doesn't say (e.g. clients like AMD should not be be interested in that node, at least not with their usual products):

IEEE Spectrum: Will customers who aren’t already at 22-nm FDSOI have reason to jump in?

Gary Patton: Everybody gets excited about leading edge. But if you look at the statistics and even out at 2022 about 75 percent of market is at nodes of 12 nm and over. And two-thirds of the market hasn’t made the jump into [either SOI or FinFET] and are operating on older technology nodes. So we see tremendous opportunity for growth in our FD platform.

 

[NTMBK]

Why. I don't see any logic to the argument. They simply want to be on the best fab. It doesn't matter if you build SoCs, CPUs and/or GPUs.

NVidia has a huge thriving business selling GPUs from the highest end to the lowest end, all on TSMC for many years.

AMD already has the their low end stuff at GF for the foreseeable future, TSMC for the leading edge. So they already have to deal with two fabs. Most other companies stick with one.

I see just ZERO chance Samsung will enter the picture in the foreseeable future.

Nvidia already uses Samsung for their low end GPUs (1050 and below).

 

[PeterScott]

Nvidia already uses Samsung for their low end GPUs (1050 and below).

I know, which is why, earlier I wrote:

Yeah, being essentially TSMC exclusive, is what destroyed NVidia. Oh, wait...

I just got lazy on the back and forth in keeping a qualifier. NVidia occasionally does a low end part elsewhere, but they are mainly and essentially a TSMC shop.

Do you actually think AMD is going to add Samsung anytime soon?

 

[NTMBK]

I know, which is why, earlier I wrote:


I just got lazy on the back and forth in keeping a qualifier. NVidia occasionally does a low end part elsewhere, but they are mainly and essentially a TSMC shop.

Do you actually think AMD is going to add Samsung anytime soon?

Ah, I missed that earlier quote 🙂 and no, AMD has been very explicit about going all in on TSMC for 7nm.

 

[PeterScott]

Ah, I missed that earlier quote 🙂 and no, AMD has been very explicit about going all in on TSMC for 7nm.

I am just surprised that people are suddenly excited about Samsung. They are fairly behind at 7nm from everything I have read, and AMD is already committed to GF and TSMC, it really looks like no chance that of AMD going to Samsung in the near future.

As far as volume constraints, I bet TSMC will give AMD more volume than GF could ever supply.