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

[PotatoWithEarsOnSide]

He also posts as if he were a bot.

 

[chrisjames61]

The only roadmap affected is the FinFET roadmap from: GlobalFoundries Press Release, Aug 27th 2018.
"GF is realigning its leading-edge FinFET roadmap to serve the next wave of clients that will adopt the technology in the coming years. The company will shift development resources to make its 14/12nm FinFET platform more relevant to these clients, delivering a range of innovative IP and features including RF, embedded memory, low power and more. To support this transition, GF is putting its 7nm FinFET program on hold indefinitely and restructuring its research and development teams to support its enhanced portfolio initiatives. This will require a workforce reduction, however a significant number of top technologists will be redeployed on 14/12nm FinFET derivatives and other differentiated offerings."

"GF is intensifying investment in areas where it has clear differentiation and adds true value for clients, with an emphasis on delivering feature-rich offerings across its portfolio. This includes continued focus on its FDXTM platform, leading RF offerings (including RF SOI and high-performance SiGe), analog/mixed signal, and other technologies designed for a growing number of applications that require low power, real-time connectivity, and on-board intelligence."

From the Report to EU Parliament, mid-2017:
"The results for this project are a technical success for supporting Soitec FDSOI business. In this project the SOI substrate technology blocks for 22FDX technology and 22FD+ were developed."
"A number of activities (outside the project) will transfer the pilot line to full size production facility: with 1st industrialization phase started in July 2017 in Soitec to catch 22FDX ramp. Due to the successes in this project significant investment was done in 2017 in the 300mm Bernin FDSOI fabrication line. This first investment phase is will be followed in 2018 with a second phase."

22FD+ became 22FDX+ as in this image: https://i.imgur.com/KOYu5AD.png
and I have previously talked about the Bernin II phases before in #283.

For 12FDX it was a spiral...
2016-
"Customer product tape-outs are expected to begin in the first half of 2019."
2017-
"The foundry is scheduled to move the 12FDX to risk production in the second half of 2018 followed by volume production in first-half 2019."
Which is verified in September 2017: https://forums.anandtech.com/thread...nce-applications.2519335/page-4#post-39089092
0.5PDK is risk production for GlobalFoundries.
1Q 2018: "We expect to be taking risk production on the parts early next year (2019), so we are pretty far along with the technology." - https://www.anandtech.com/show/1243...ew-with-dr-gary-patton-cto-of-globalfoundries
2Q 2018: "We expect tape outs on 12FDX in 2020 with deliveries in 2021." - http://www.eenewsanalog.com/news/globalfoundries-add-another-nvm-fdsoi/page/0/1
2H2018 - NostaQuess:
12FDX is being pushed further along to support 22FDX+. Which hopefully it will get inserted into the roadmap eventually.

22FDX (Power-optimized)
22FDX+ (Performance-optimized)
12FDX (Perf or Power-optimized separated in PDK extensions, not in node).

Also, Europe and China are throwing billions at GlobalFoundries. United States is giving zip, Saxony is thousands-millions, not even close to billions. Dresden/Chengdu are both declared foundries for FDSOI. So, Malta barely makes revenue targets and is a constant loss. Malta is a declared FinFET foundry, even though they do lab-work for 22FDX/22FDX+/12FDX. So, if FDSOI comes out specifically a net profit transistor type. Then, all the expenditures will revolve around the leading edge in FDSOI.

The only people who can steal customers from GlobalFoundries are: HLMC, Samsung, STMicroelectronics.
Samsung with two low-capacity FDSOI fabs => S1(Korea) & S2(America)
HLMC with one low-capacity FDSOI fab => H6(China)
STMicroelectronics with one low-capacity FDSOI fab but are now licensing from GlobalFoundries => France and Italy fabs are not competing.
GlobalFoundries will have two fabs: Fab 11(China-MegaFab) and Fab 1(Europe-GigaFab).

While, for FinFETs it will always be go TSMC. TSMC will probably provide better differentiated options for 16FF/12FF/10FF over GloFo's 14FF/12FF.

Hypothetically: If Apple lets say created an B-series on FDSOI, which is would be a budget form of the A-series. Only GlobalFoundries would be able to satisfy their order capacity of FDSOI.

Do you think anyone ever reads all that nonsense you always copy and paste?

 

[NostaSeronx]

Do you have a link on that. This is a huge deal if true, even if the timeline is pushed out.

No link.

Stacked Nanosheets have continuous widths, rather than quantized widths. GlobalFoundries wants to get off FinFETs sooner than later because of it. So, instead of waiting for 1000W EUVs for 3LP. That transistor type is being pulled ahead to the new 7LP. The freedom of design is similar enough to UTBB FDSOI. Single stack, dual stack, power, performance, etc.

This however requires GlobalFoundries to trash most of the 7-5-3 roadmap. With a nanosheet focus now rather down the road. The hold is to R&D to see if the new 7nm node is viable which it has been tested and verified. However, there is new BEOL/MOL/FEOL in the works to make it even better than what was before it. That is why the timeline has been pushed ahead, other than fiscal reasons.

FinFET 7LP -> FinFET 5LP -> Stacked Nanosheets 3LP => Similiar to Samsung
New Roadmap:
Stacked Nanosheets 7LP -> so on. (Stressors, multiple-height(1 planar, 2 planar, 3 planar, 4 planar, etc), CoolCube/Complementary Stacked Nanosheets; 3 planar NFET on one substrate interconnected to 3 planar PFET, etc enhancements) + what they need for 7LP for consumers. A lot of freedom to be explored and validated and tested.

- Rebudget
- Renew R&D partnerships with new contracts.
- Etc.

 

[Arachnotronic]

Do you have a link on that. This is a huge deal if true, even if the timeline is pushed out.

It's complete nonsense.

 

[DrMrLordX]

Nosta isn't wrong about anything: his quantum signature is just different than ours.

Somewhere, Keanu Reaves is saying "whoa" without even knowing why.

He also posts as if he were a bot.

He is AU - Artificial Unintelligence.

No link.

Next!

 

[Abwx]

Do you think anyone ever reads all that nonsense you always copy and paste?

GF 7nm was ready for ramping, it s just that this would had required 2-4bn to do the ramp and AMD was the only significant customer.
This wasnt enough to amortize the ramping since AMD account for only 1.5-1.8bn of GF yearly revenue.

It was obvious for GF management that AMD Epyc/Ryzen sales didnt take off as expected, and that they would be bankrupt if they started ramping 7nm without AMD sales increasing hugely in the meantime, and if we are to look at AMD s Q3 18 results they were perfectly on point.

 

[beginner99]

GF 7nm was ready for ramping, it s just that this would had required 2-4bn to do the ramp and AMD was the only significant customer.
This wasnt enough to amortize the ramping since AMD account for only 1.5-1.8bn of GF yearly revenue.

It was obvious for GF management that AMD Epyc/Ryzen sales didnt take off as expected, and that they would be bankrupt if they started ramping 7nm without AMD sales increasing hugely in the meantime, and if we are to look at AMD s Q3 18 results they were perfectly on point.

Chicken and egg problem. AMD would have needed to secure tsmc 7nm early access a long time ago, at least over 1 year. Hence their decision to go tsmc was made a long time ago. So i rather think GF had no customer at all. I think AMD finally made a power move when it had power. Remove all production from GF and if they are their only major customer even wsa payments won't cover cost of creating the process hence not worth it for gf hence no wsa for 7 nm. IMHO clever power move by amd.

 

[Tuna-Fish]

Recently, the story that Epyc 2 will contain 8 small zen chiplets and a large (older-process) IO die has made the rounds. I am still a bit skeptical, but for the purpose of this post, let's assume that they are completely correct.

The chiplets will obviously be made at TSMC 7nm. This has been all but completely confirmed. However, the IO die can be made on any of a number of processes, the most likely being the GF 14nm (because there will be free capacity, and AMD still has a WSA), the TSMC 16nm (the other nearly recent process), and the dark horse: GF 22FDX.

22FDX is not great for high-speed logic, but FD-SOI processes are better than finfets for analog components, as needed by high-speed IO.

Because the chip needs to fit a lot of IO, and that requires a lot of circumference, it will need to be much larger than the space taken by the IO. The logical thing to fill that space with is a gigantic, memory-side cache. The SRAM on 22FDX is nice and low-power, but takes a lot of area. However, GloFo has, for a long time now, been researching MRAM on 22FDX, and is now expecting to "ship products before end of 2018". The SRAM-like MRAM macro on 22nm would be very small, maybe even smaller than a typical 8T SRAM cell on 7nm. If they used that for the cache, they could ship a massive, fast memory-side L4 that can retain it's contents when sleeping, allowing very low idle power.

This is of course unlikely as the first production model, simply because it would make AMD's most important release depend on many bleeding-edge technologies. But maybe as a point upgrade on EPYC, using the same CPU chiplets but with a new IO die?

 

[Despoiler]

It looks like GF 7nm is back in play? They just spun up a new company called Avera Semi who is going to offer custom ASICs. 7nm is mentioned a few times in the release. I don't know what to make of this. We know GF cancelled 7nm because they lost AMD. Was GF 7nm not going to hit the release target or perf target AMD set which is why they went with TSMC? GF now has another 7nm partner. Is 7nm availability a later date?

https://www.globalfoundries.com/new...es-avera-semi-wholly-owned-subsidiary-deliver

 

[jpiniero]

It looks like GF 7nm is back in play?

Sounds like they might fab these custom ASICS at TSMC or SS later.

 

[NTMBK]

It looks like GF 7nm is back in play? They just spun up a new company called Avera Semi who is going to offer custom ASICs. 7nm is mentioned a few times in the release. I don't know what to make of this. We know GF cancelled 7nm because they lost AMD. Was GF 7nm not going to hit the release target or perf target AMD set which is why they went with TSMC? GF now has another 7nm partner. Is 7nm availability a later date?

https://www.globalfoundries.com/new...es-avera-semi-wholly-owned-subsidiary-deliver

"access to alternate foundry processes at 7nm and beyond". They're setting their ASIC business free so that it can use TSMC and Samsung at 7nm, and is no longer tied to using GloFo. Without this move they would lose any customers who wanted a <14nm ASIC.

 

[Abwx]

Chicken and egg problem. AMD would have needed to secure tsmc 7nm early access a long time ago, at least over 1 year. Hence their decision to go tsmc was made a long time ago. So i rather think GF had no customer at all. I think AMD finally made a power move when it had power. Remove all production from GF and if they are their only major customer even wsa payments won't cover cost of creating the process hence not worth it for gf hence no wsa for 7 nm. IMHO clever power move by amd.

Clever or not they had actually no choice, and it s hard to put the blame on GF since the volumes were nowhere in sight to accomodate billions of lost revenie due to ramping 7nm.

That being said the process is here should AMD get more revenue, GF is overall a better partnership than TSMC.

22FDX is not great for high-speed logic, but FD-SOI processes are better than finfets for analog components, as needed by high-speed IO.

There s a 12FDX if ever they want to use this process for some circuitries, obviously that s a relatively low power process.

https://www.globalfoundries.com/technology-solutions/cmos/fdx/12fdx

 

[DrMrLordX]

There s a 12FDX if ever they want to use this process for some circuitries, obviously that s a relatively low power process.

https://www.globalfoundries.com/technology-solutions/cmos/fdx/12fdx

Hey, it's someone that is not NostaSeronx talking about GF nodes. Yay! So what is the projected availability of 12FDX?

 

[Abwx]

H1 2019, so quite late for integration in a first gen chiplet if such a solution make sense, otherwise they have GF s 12nm at disposal, so far it wasnt entirely exploited by AMD for Zen+ and it should be adequate for whatever is not higher frequencies dependant for perfs.

https://www.globalfoundries.com/new...-extends-fdxtm-roadmap-12nm-fd-soi-technology

 

[Dayman1225]

H1 2019, so quite late for integration in a first gen chiplet if such a solution make sense, otherwise they have GF s 12nm at disposal, so far it wasnt entirely exploited by AMD for Zen+ and it should be adequate for whatever is not higher frequencies dependant for perfs.

https://www.globalfoundries.com/new...-extends-fdxtm-roadmap-12nm-fd-soi-technology

12 FDX was delayed into 2020 iirc.

edit:

A follow-on 12FDX process was originally due in 2019 but GF is holding off introduction of the process because customers are just now designing and ramping up products on 22FDX. Development of 12FDX is proceeding well and it will be introduced when needed, we estimate this will be around 2020. 12FDX will offer <0.4 volt operation and a >20% performance improvement over 22FDX.

 

[NostaSeronx]

12FDX might be delayed further to 2022. GlobalFoundries' 12FD node is getting an overhaul.

What might launch between however is a 22FD+ with a shrunk BEOL as 14FDX: 104 Cx & 64 Mx.
Which is to match better with Samsung nodes:
28FDS(28FD w/ 28nm BEOL) -> 22FDX (22FD w/ 22-28nm BEOL)
18FDS(28FD+ w/ 14nm BEOL) -> 14FDX (22FD+ w/ 14nm BEOL)

22FDX => 2018 => 2015 TD
14FDX => 2020
12FDX => 2022 => 2019 TD