GlobalFoundries Puts its 7 nm Program on Hold Indefinitely

[FordGT90Concept]

They have no choice because their sugar daddy stopped giving them more sugar. They're hoping they can sell enough 14 nm chips to dig themselves out of the hole they are in. It's a gamble because if someone manages to move beyond 7-10 nm, 14 nm becomes far less attractive.

TL;DR: GloFo is no longer an innovation company, just a production company that will license other company's fab tech.

 

[Valantar]

They have no choice because their sugar daddy stopped giving them more sugar. They're hoping they can sell enough 14 nm chips to dig themselves out of the hole they are in. It's a gamble because if someone manages to move beyond 7-10 nm, 14 nm becomes far less attractive.

TL;DR: GloFo is no longer an innovation company, just a production company that will license other company's fab tech.

Yep, pretty much. No wonder they've stopped investing, though, given that they've already sunk ~20 billion into them and would need to double that to build a new fab - which would still place them at a disadvantage compared to bigger competitors with more fab capacity. And sure, 14nm will become less attractive in time, but at the same time, it'll by then be a well-established and likely very cheap process. If their proposed developments pan out, it'll still have its uses. Not to mention that I don't think we'll see anything noticeably better than TSMC 7nm for the next 5-ish years. EUV might make an entrance, but it's unlikely we'll see significant process improvements post-7nm for quite a while. And they'll likely be able to licence Samsung 7nm or something similar for relatively cheap by then.

 

[efikkan]

Huh? I thought things got cheaper on smaller nodes... at least, once you have one running successfully.

Price per transistor may go down if the yields are good and the production speed is high enough. But the cost savings usually comes into play towards the end of the node's lifecycle.

The problem with 7nm is not just the yields, but the yields in combination with the slow patterning on the DUV process. Even with more transistors per wafer, the production speed of first generation 7nm is so slow that it becomes too costly.

The big elephant in the room is EUV, this is supposed to be the key to 7nm's success. We don't know for sure if EUV will be ready anytime soon, so we might be looking at either a slower 7nm adoption phase, or limited quantities for a long time.

What does this mean for Zen 2 and Navi?

I assume it will mainly impact the mid- to lower end products. Even if GF didn't give up 7nm, TSMC would still be the only one shipping at 7nm through most of 2019. And as I've mentioned before, even if we're optimistic, the shipping volume of 7nm in the first half of 2019 will be very limited. GF was supposed to ship several quarters later than TSMC anyway, so it will most likely slow down the transition period rather than delay the first Zen 2 products.

There are a couple of things I would like to know, like how many production lines does TSMC have for 7nm compared to 16/12nm, and how much of this is reserved for Nvidia.

TL;DR: GloFo is no longer an innovation company, just a production company that will license other company's fab tech.

I thought they already were just licensing Samsung's technology (14nm FF).

 

[FordGT90Concept]

There are a couple of things I would like to know, like how many production lines does TSMC have for 7nm compared to 16/12nm, and how much of this is reserved for Nvidia.

AMD's orders will dwarf NVIDIA. GPUs are useless without CPUs.

I thought they already were just licensing Samsung's technology (14nm FF).

They are. That's why GloFo is on 14nm at all. Licensing is very expensive too. I'm under the impression that GloFo never actually reported a net profit.

 

[efikkan]

AMD's orders will dwarf NVIDIA. GPUs are useless without CPUs..

Most of the production capacity is booked years in advance, on TSMC "16/12nm" Nvidia ended up using all of it for Pascal and Volta/Turing, even using some from Samsung on 14nm as well. Nvidia will certainly have tried to book as much as possible of 7nm, and would need about twice as many production lines as on 16/12nm to produce the same volume of wafers. AMD who wants to make CPUs and GPU will require even more of that, if they can get it.

 

[FordGT90Concept]

AMD has no choice. AMD will get what TSMC can make.

 

[Valantar]

AMD has no choice. AMD will get what TSMC can make.

That's not true. AMD can go with Samsung if necessary.

 

[medi01]

Nvidia will certainly have tried to book as much as possible of 7nm

To produce what? A competitor to glorious sold out "shit can be even more expensive than you'd think" 2xxx series with that RTX stuff that cripples perf so much that even 1080p is not enjoyable?

I think you don't realize what the game here is.
AMD is an underdog with highly limited resources. As game theory goes, its only chances are GAMBLING.
Focus resources on a single thing (as you don't have money for more than one) and pray to Cthulhu for it to succeed.
Vega, HBM, HBM2, 7nm, you name it.

We were told they bet on 7nm, we were told GF sucks as much or even more than most suspected.

I wish I knew what release of 7nm Vega actually means in terms of fab process maturity.

 

[FordGT90Concept]

That's not true. AMD can go with Samsung if necessary.

Samsung announced they successfully prototyped it back in May. They're quite a ways behind TSMC.

 

[Valantar]

Don't think Samsung has achieved 7 nm for processors yet.

Key word there being "yet". TSMC will (likely) get them there first, but Samsung is still a fall-back option.

 

[FordGT90Concept]

Not when they're already working on everything for TSMC 7 nm. To switch to Samsung at this point would basically require starting over.

 

[Valantar]

Not when they're already working on everything for TSMC 7 nm. To switch to Samsung at this point would basically require starting over.

Of course this would incur very significant delays. All I'm saying is that TSMC still isn't their only option. Adding nuance, not contradicting Besides, AMD is likely to have known about this shift for at least a couple of quarters, and if necessary, could have started porting designs to Samsung 7nm (or at least looking into the possibility of this) a while ago.

 

[WikiFM]

After some hours readying about process nodes, I discovered that since 2012 each foundry creates its own process node. I concluded that TSMC is ready with its 7nm because its 7nm are simplier than the 10 nm of Intel. How did I get that idea? Well according to official guidelines about the physical properties of transistors of the ITRS, the specs hasn't been fullfilled by TSMC since its 16 nm, which is more similar to its 20 nm than the official 16/14nm spec. Even its 12nm is more similar to its 20nm than the official 16/14 nm spec.

But TSMC hasn't been the only one cheating, Samsung's 10 nm is actually 14 nm according to the official specs, and Samsung's 14 nm is actually more similar to its 20 nm than the official 16/14 nm spec too. I feel so stupid for not knowing all these before, I really believed that each process node was the same for every foundry.

Sources: https://en.wikichip.org/wiki/WikiChip Nodes 22 nm to 7 nm
https://www.semiconductors.org/clientuploads/Research_Technology/ITRS/2015/0_2015 ITRS 2.0 Executive Report (1).pdf pages 38, 48

 

[R0H1T]

Of course this would incur very significant delays. All I'm saying is that TSMC still isn't their only option. Adding nuance, not contradicting Besides, AMD is likely to have known about this shift for at least a couple of quarters, and if necessary, could have started porting designs to Samsung 7nm (or at least looking into the possibility of this) a while ago.

IIRC GF & Sammy were collaborating on 7nm, not 100% sure, like they did with 14nm? If so moving to Samsung won't be that hard or costly IMO.

 

[Valantar]

IIRC GF & Sammy were collaborating on 7nm, not 100% sure, like they did with 14nm? If so moving to Samsung won't be that hard or costly IMO.

I believe I've seen reporting saying the same quite a while ago, but my impression is that this was early speculation/based on inaccurate information, and that GF had since moved on to their own design. Given that Samsung hasn't been mentioned in any of the reporting around this cancellation, I still believe that's the case.

 

[nemesis.ie]

@WikiFM This has been known for a long time. Marketing is marketing.

The important factor is how much improvement we see from a given foundry's node to their next node, what they call it is less important.

TSMC's "7nm" will be a substantial improvement from their current "12nm". We will see how the end products perform and that will be the real metric.

 

[WikiFM]

TSMC's "7nm" will be a substantial improvement from their current "12nm". We will see how the end products perform and that will be the real metric.

You are forgetting TSMC's 10 nm which is the real newest node, this is the one to compare to 7 nm. 12 nm is just polishing of its 16 nm process.

 

[nemesis.ie]

Good point. Things are also made trickier by the fact that Zen and Vega are both on GF's 14/12nm(+) at the moment rather than TSMC's processes.

 

[WikiFM]

Vega at 12 nm? I havent read that anywhere, as far as I know Vega is still (for better or worse) in 14 nm

 

[nemesis.ie]

Yep, good catch. Edited.