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

[DrMrLordX]

Leading Edge is not over at GlobalFoundries.

GF just said that's it's over. Sorry, not buying it.

 

[Dayman1225]

3nm by 2020.

*Risk production in 2020

 

[itsmydamnation]

GF just said that's it's over. Sorry, not buying it.

You can tell he is try to convince himself by the massive increase in his posting, before he could continue on in the delusion, but now with irrefutable evidence he needs to convince himself it never happened.

 

[NostaSeronx]

GF just said that's it's over. Sorry, not buying it.

Leading-Edge involves
1. FDSOI <== the future of this was accelerated.
2. FinFETs <== the future of this was canned.

“We are seeing strong demand for both our mainstream and advanced technologies, from our world-class RF-SOI platform for connected devices to our FD-SOI and FinFET roadmap at the leading edge.”
- Sanjay Jha / Feb 09, 2017

"We have found a way to separate ourselves from the pack by emphasizing our differentiated FDSOI roadmap and client-focused offerings that are poised to enable connected intelligence. We will continue to build on our momentum and look for ways to expand our reach to address the evolving needs of the industry,"
- Tom Caulfield / July 9th, 2018

*Risk production in 2020

Which is PDK version 0.5 if they follow GlobalFoundries' risk production measures.

 

[zinfamous]

Leading Edge is not over at GlobalFoundries.
Only FinFETs, IBM is aware of this decision and so was AMD. This list includes Qualcomm, MediaTek, HiSilicon, and Unisoc.

Everyone knew about GlobalFoundries cancelling FinFETs to improve upon FDSOI. The reasoning was simple, it had a higher return of investment for everyone. All of them wanted 12FDX over 7LP.

Once the restructuring from FinFETs and FDSOI to just FDSOI. GlobalFoundries will do some interesting things in Q1 2019.
---
"The only meaningful benefit of SOI for 14nm FF was to add eDRAM. Ironically that node is the first and last FF on SOI node of entire world. For the next nodes there is no choice for them other than using a foundry process (say 7nm at SS or TSMC) and drop eDRAM."
- Ali Khakifirooz, NVM engineer @ Intel, and one of the creators of UTBB FDSOI

and

3nm by 2020.

Do you think GlobalFoundries could have done this? (imho, no.)

IBM POWER roadmap doesn't not declare POWER10's node either.
https://pbs.twimg.com/media/DlKexFpX4AALa6B.jpg:large

eDRAM always fell way to eMRAM anyway:
https://pc.watch.impress.co.jp/img/pcw/docs/471/308/photo027.jpg

You're linking slides and images without attribution. What are the dates on these presentations? I'm rather curious if this information predates GloFo's announcement last week. I mean, look at that one up there, it lists 7nm FinFET for 2018. That was known as early as last year to be a tough call.

It's going to be hard to pay any attention to your posts unless you place time stamps on all of this evidence that you have accumulated.

 

[Gideon]

Leading-Edge involves
1. FDSOI <== the future of this was accelerated.
2. FinFETs <== the future of this was canned.

Quit warping reality to suit your preferences of a brighter FDSOI & CMT future. I'm sorry, but it ain't gonna happen. At least not until GF finds a new investor.

While there is a small chance, that GF will continue investing in smaller FDSOI nodes, there is zero proof for that for now, until there is a corresponding statement from some higher-up after 27. Aug 2018.

Meanwhile there are plenty of reasons to believe otherwise:
IMO the most telling one is that in their 27. August press release (the one canning 7nm) they only mentioned RF and RF SOI @ 22FDX, not continued (or even sped-up) investment into smaller FD-nodes, on schedule.

If I were a PR guy trying to salvage such devastating news, that would be the very first thing I would highlight. It requires a particularly distorted mind to think, that they plan to do just that, but will do so in absolute silence (what would that accomplish?)

All your ancient pretty slides are useless until you find some that were put up after the 7nm decision was made public (which I'd be eager to see btw).

At least Gary doesn't consider FD bleeding edge:

Now that we’re relieved from burden of having to fund this bleeding edge stuff, we’re able to redirect dollars and resources toward these other areas….

But what does Tom Caulfield say:

Lifting the burden of investing at the leading edge will allow GF to make more targeted investments in technologies that really matter to the majority of chip designers in fast-growing markets such as RF, IoT, 5G, industrial and automotive

Nope, doesn't consider this leading edge either ...

Bottom line:
I would have absolutely nothing against GF speeding up FD nodes. I just find the lack of evidence of such very telling, that unless they find buyers or new investors, they lack the funds to do so, so they don't bring that up.

 

[french toast]

Global foundries will turn into a mid sized fab like UMC, cheap and cheerful last gen processes for the markets that don't want or need expensive leading edge nodes.
Plenty of moderate profit to be made in those markets catering for those customers.
Mediatech?, STmicro?, broadcomm?

 

[jpiniero]

MediaTek is also using TSMC 7 nm as well, although not until next year.

 

[NostaSeronx]

You're linking slides and images without attribution. What are the dates on these presentations? I'm rather curious if this information predates GloFo's announcement last week. I mean, look at that one up there, it lists 7nm FinFET for 2018. That was known as early as last year to be a tough call.

It's going to be hard to pay any attention to your posts unless you place time stamps on all of this evidence that you have accumulated.

#248 is 2018(repeated Semicon West 2017), Semicon Europa 2016, CEA-Leti 2018(regurgitating GlobalFoundries).

#250 SFF 2018 JAPAN, September 4, 2018 // Samsung Foundry Forum
https://www.semiwiki.com/forum/f293/samsung-electronics-investors-forum-june-4-2018-a-10566.html
"7LPP risk produce second half of this year." from June 4, 2018 Forum. So, the SFF Japan conference just repeated what SIF said.
// "28LPP move to 28FDS and 18FDS (FDSOI). 28FDS now and 18FDS next year. NXP first adopter, but also popular in Japan."
IBM power roadmap is Hotchips 2018.
eMRAM is from defunct MagSil in 2011. eMRAM has been taken to better places with other companies. Especially, with Foundation IP(by Everspin) versions of eMRAM-F, eMRAM-S, and eMRAM-D; https://www.everspin.com/everspin-embedded-mram

On a totally unrelated note, decided to look at UMC:
1. http://www.umc.com/english/news/2013/20130723.asp
2. http://www.eenewseurope.com/news/what-happened-suvolta-0
2a. http://www.fujitsu.com/jp/group/mifs/en/resources/news/press-releases/2015/0408.html
3. http://www.umc.com/english/news/2018/20180629-1.asp

If UMC does a 14nm DDC, then the market is set. Other than FinFET is in competition.
https://m.eet.com/content/images/edn/UMC_14_1501146764.png
UMC issues might also be GlobalFoundries issues.

Also, UMC also has MRAM:
https://www.mram-info.com/avalanche-sign-agreement-umc-28nm-embedded-stt-mram-technology
Avalanche also has something interesting:
http://www.avalanche-technology.com/technology/embedded-mram-technology/
http://www.avalanche-technology.com/technology/3d-cross-point-mram-technology/

UMC2014:
28HPM will migrate to 28HPC
There is 14FF+ and 14FFC. Apparently, IBM was(is still) doing 10FF co-design, the same way CEA-Leti and STMicro did for GlobalFoundries for 22FDX/12FDX and 10FD/7FD. UMC still has a chance for 10nm SOI FinFETs for IBM, if that IBM assignee is still there.
https://m.eet.com/media/1175870/20121212pcfdsoinode898.jpg
10FD and 10FFSOI back in the day was super linked together.
https://www-03.ibm.com/press/us/en/pressrelease/41294.wss
"A 10nm Platform Technology for Low Power and High Performance Application Featuring FINFET Devices with Multi Workfunction Gate Stack on Bulk and SOI" <-- UMC appears in this paper.

so, HPC^u and HPC^u+ and 22nm are all DDC.
Also, wanted to point out HPC^u+ has 6 VTs!! => https://i.imgur.com/lnx6Yjg.png
Here is the 22nm => https://i.imgur.com/t6FO3SQ.png
Original 14nm benchmark was against 28HPM. So, the aggregate power efficiency benefit puts DDC as the LP winner. The secret winner only found at one semiconductor fab. DDC however has a mask count issue compared to FDSOI at GloFo/Samsung. 28/22 DDC @ UMC is 55 masks 1P11M and GloFo is 36 masks with 1P8M for 22FDX and 53 masks with 1P9M for 12FDX.

Not FinFET postion after 28nm:
Globalfoundries 1st (12FDX)
Samsung 2nd (18FDS)
UMC 3rd (No roadmap after 22nm DDC)
HLMC dark horse (By the will of China, all of China shall license our 22FDSOI node!) // http://www.hlmc.cn/en/News.aspx?id=91 // IMEC FDSOI *bleh*

 

[krumme]

GF is going to dissapear slowly. 10 years and they are gone or completely irrelevant. Not only for highend but all over. Bought out for nothing doing fab work for others.

There i no such thing as a niche market for them with their us based cost structure doing cheap processes. They will learn that. History shows niche is a thing of the 80ties.
Their RF competence is surely miniscule and easily out competed. Its way way to late to start scaling that now. If they started scaling 6 years ago and skipped all usual highend process dev they might have had a chance.

They have nothing to bring to the market others cant do cheaper or better. 14nm is dirt cheap and fully depreciated, and we are well into 7nm, before they get anything 22xxx whatever to the market. Its a dead end. They cant recover their fixed cost.
Its just imagination with no technical or organizational backbone.

They can just purshade Mubadala to continue and give them oil money because the alternative is using money on western whores and more racing cars for the political elite.
They think gf is a better investment. Or they try to delay the obvious because of sunk cost fallacy. Especially dangerous in a political controlled organization.

Obviously Mubadala is wrong and should take the entertainment.

 

[DrMrLordX]

GF is going to dissapear slowly. 10 years and they are gone or completely irrelevant. Not only for highend but all over. Bought out for nothing doing fab work for others.

There i no such thing as a niche market for them with their us based cost structure doing cheap processes. They will learn that. History shows niche is a thing of the 80ties.
Their RF competence is surely miniscule and easily out competed. Its way way to late to start scaling that now. If they started scaling 6 years ago and skipped all usual highend process dev they might have had a chance.

They have nothing to bring to the market others cant do cheaper or better. 14nm is dirt cheap and fully depreciated, and we are well into 7nm, before they get anything 22xxx whatever to the market. Its a dead end. They cant recover their fixed cost.
Its just imagination with no technical or organizational backbone.

They can just purshade Mubadala to continue and give them oil money because the alternative is using money on western whores and more racing cars for the political elite.
They think gf is a better investment. Or they try to delay the obvious because of sunk cost fallacy. Especially dangerous in a political controlled organization.

Obviously Mubadala is wrong and should take the entertainment.

Ouch! Maybe they should invest in blockchain instead? Next big thing!!!11!!11! (and I say this as a blockchain advocate)

Seriously though, I think they may do okay as a second-source if/when TSMC and Samsung run out of capacity. Their 14LPP/12nm nodes ARE competent. The IBM guys showed they could refine nodes for better performance over time, even under less-than-attractive circumstances. Fabs are expensive to set up. Customers can probably cut some nice deals on older processes from them, and by "older" I mean "current as of 2017".

Compare them to a bunch of also-ran fabs and you'll see they'll stack up pretty well in that department.

That being said, they may have to downsize. Considerably.

 

[PeterScott]

Wow, the previous two

GF is going to dissapear slowly. 10 years and they are gone or completely irrelevant. Not only for highend but all over. Bought out for nothing doing fab work for others.

There i no such thing as a niche market for them with their us based cost structure doing cheap processes. They will learn that. History shows niche is a thing of the 80ties.
Their RF competence is surely miniscule and easily out competed. Its way way to late to start scaling that now. If they started scaling 6 years ago and skipped all usual highend process dev they might have had a chance.

They have nothing to bring to the market others cant do cheaper or better. 14nm is dirt cheap and fully depreciated, and we are well into 7nm, before they get anything 22xxx whatever to the market. Its a dead end. They cant recover their fixed cost.
Its just imagination with no technical or organizational backbone.

They can just purshade Mubadala to continue and give them oil money because the alternative is using money on western whores and more racing cars for the political elite.
They think gf is a better investment. Or they try to delay the obvious because of sunk cost fallacy. Especially dangerous in a political controlled organization.

Obviously Mubadala is wrong and should take the entertainment.

Maybe you and Nosta could just cancel each other out, with your polar opposite viewpoints.

This is not doom and gloom for GF. Leadership made a rational decision here, they now have a chance to make money,since they aren't sinking so much into leading/bleeding edge process development.

GF will exit the spotlight of PC enthusiasts, as new leading edge CPU/GPU/SoC parts will be using other fab options, but there are plenty of industry uses for fabs outside of leading edge GPU/CPU/SoC parts.

 

[NTMBK]

I suspect that GloFo will find a second life producing specialised parts for the US military. Lots of work making radar and communications chips, in a secure fab based in the US.

 

[NostaSeronx]

I suspect that GloFo will find a second life producing specialised parts for the US military. Lots of work making radar and communications chips, in a secure fab based in the US.

Yep.

https://www.timesunion.com/tuplus-b...s-wins-chip-contract-with-Defense-7966717.php
"GlobalFoundries has won a contract with the U.S. government to provide sophisticated custom-built computer chips to the Department of Defense and its suppliers likely valued at more than $400 million.

The seven-year deal, first reported by The Wall Street Journal, essentially makes GlobalFoundries the sole supplier of high-end chips used in fighter planes, weapons systems and satellites used by the U.S. military and its spy agencies."
https://www.militaryaerospace.com/articles/2017/03/chip-fabrication-military-electronics.html
etc.

In "A DARPA Approach to Trusted Microelectronics":
< Although many DoD systems rely on older 90-nanometer (nm) or 130-nm technologies, leading-edge microelectronics offer specific, military-relevant advantages to DoD and its foreign competitors.

<The Americas comprised more than half of the $3.1 billion market for military andaerospace electronic devices in 2011. Within DoD, these devices support nearly all critical capabilities, among them the global positioning system, radar, command and control, and communications. Demand for defense-related electronics is global, however; foreign governments are bolstering their access to leading-edge microelectronics and leveraging readily available commercial devices for military use.

By Trusted Microelectronics Joint Working Group : New Methods to Instill Trust in Commercial Semiconductor Fabrication;
< Unfortunately, the DMEA Trust methodology for category 1A manufacturing is not extensible to most foreign fabs. During the first ten (10) years of the Trusted Foundry, only IBM Microelectronics offered access to complementary metal–oxide–semiconductor (CMOS) processes below 90nm lithography. In 2014 that capability was imperiled by the IBM Microelectronics Division’s sale to GLOBALFOUNDRIES (GF). In 2015, DoD successfully mitigated the risk of access loss by obtaining GF’s agreement to operate the former IBM Trusted Foundry through a U.S. subsidiary (GFUS2). As a result, the most advanced node in the Trusted Foundry program is GF’s 32nm SOI; even with GF in the fold, DoD has no Trusted Foundry at today’s advanced nodes such as 22nm FD-SOI, 16/14nm FinFET. Outside GF, the most advanced CMOS node in the Trusted Foundry program is 90nm CMOS, which was introduced by Intel and others more than ten (10) years ago.
=> Trust accreditation for category 1A essentially requires facility clearance for collateral secret work that eliminates most foreign fabs from consideration. Various requirements associated with the Trusted Foundry program have also limited its adoption by domestic fabs.

===
https://www.globalfoundries.com/market-solutions/aerospace-and-defense

A First Look at 22 nm FDSOI SRAM Single-Event Test Results:
- NASA Goddard Space Flight Center
- ASRC Federal Space and Defense, Inc. (AS&D, Inc.)
- Boeing
- Defense Microelectronics Activity

22FDX(2019) and 12FDX(2021) + 45RF(2019) and 22RF(2021), seems to be the winners for next-gen.

Boeing and Airbus both appear to have chosen FDX.
STMicro is already doing things: https://www.satellitetoday.com/inno...announce-soc-interactive-satellite-terminals/

http://dahlia-h2020.eu and http://dahlia-h2020.eu/about-project/
"DAHLIA has been developing a Very High Performance microprocessor System on Chip (SoC) based on STMicroelectonics European 28nm FDSOI technology with multi-core ARM processors for real-time applications, eFPGA for flexibility and key European IPs, enabling faster and cost-efficient development of products for multiple space application domains.

The performance is expected to be 20 to 40 times the performance of the existing SoC for space. This performance level, combined with a large set of integrated peripherals including dedicated on-chip functions for GNSS, TM and TC support, will enable key space applications to be executed within the same microprocessor significantly reducing cost and mass and boosting competitiveness of future European space equipments.

With the highest cumulative number of European satellites and electronics equipments successfully operating in orbit, Airbus Defense and Space and Thales Alenia Space represent together the indisputable best positioned companies in Europe to ensure the maximum relevance of the DAHLIA SoC for its future use by the whole European Space community.
Beyond Space applications, the adoption of the ARM processor will enable the convergence with terrestrial applications benefiting from the strong ARM ecosystem while the new SoC will ensure European strategic non dependence for the most critical component on board"

 

[krumme]

They are eating into prior years capital investment from mubadala.
Its not going to last.
How on earth is 400M usd going to fund new nodes? Its a drop in the ocean in fab world. I guess it just barely covers variable cost. So its good business now but its lipstick on a pig. Fab technology and -knowledge- wears down quickly.

 

[krumme]

Wow, the previous two


Maybe you and Nosta could just cancel each other out, with your polar opposite viewpoints.

This is not doom and gloom for GF. Leadership made a rational decision here, they now have a chance to make money,since they aren't sinking so much into leading/bleeding edge process development.

GF will exit the spotlight of PC enthusiasts, as new leading edge CPU/GPU/SoC parts will be using other fab options, but there are plenty of industry uses for fabs outside of leading edge GPU/CPU/SoC parts.

I am right. So no middle way.
I am not sure the decision was at GF. I think the call was from bank of Mubadala.
No more money.
And this decision was probably made several months back so strategies for gf and amd could be adapted before it went public.
Shareholdervalue.
While tsmc is on track for 5nm GF havnt made a node since 28nm.
Ofcource with all that 14nm stuff running now you can find customers for it in some cheaper variants for select niches.
But the backbone is still technology that is paid directly, and indirectly via amd, from Mubadala and was brought in from outside.

And btw its not gloom and doom. Its just slow shutdown until rest is sold for peanuts. Fair. Nothing wrong with that.

 

[DrMrLordX]

How on earth is 400M usd going to fund new nodes?

It won't. GF isn't going to develop new nodes. Or, at least, that's what all their announcements indicate.

 

[Dayman1225]

It won't. GF isn't going to develop new nodes. Or, at least, that's what all their announcements indicate.

Whens the last time you saw a fab announce that they are putting future nodes on hold indefinitely and halting R&D and then start it all back up again?

 

[DrMrLordX]

Whens the last time you saw a fab announce that they are putting future nodes on hold indefinitely and halting R&D and then start it all back up again?

Never.

 

[Dayman1225]

Never.

Exactly.

 

[Ajay]

They are eating into prior years capital investment from mubadala.
Its not going to last.
How on earth is 400M usd going to fund new nodes? Its a drop in the ocean in fab world. I guess it just barely covers variable cost. So its good business now but its lipstick on a pig. Fab technology and -knowledge- wears down quickly.

The only way I can see GF survive is by buying 3 year old process tech from someone like Samsung and implementing it. It's the only way they can survive as a long term (n-1, n-2) fab. Otherwise, they will end as you have said.

 

[Abwx]

I am right. So no middle way.
.

Well, a sentence of yours in a previous post :

Their RF competence is surely miniscule and easily out competed. Its way way to late to start scaling that now.

Even if FDSOI is the best process for RF applications..?.

Actually they should be quite competitive in this department, firms that adopted FDSOI did so for the RF performance, basically lower input capacitances wich allow for higher frequencies when devices are used within their linear region.

 

[Hitman928]

Well, a sentence of yours in a previous post :



Even if FDSOI is the best process for RF applications..?.

Actually they should be quite competitive in this department, firms that adopted FDSOI did so for the RF performance, basically lower input capacitances wich allow for higher frequencies when devices are used within their linear region.

The RF processes they inherited from IBM are fantastic. Not yet convinced on the ones they've developed on their own. The company I work for is actually considering getting the 22FDX PDK for evaluation so I might get some opportunity to delve in a little deeper.

 

[krumme]

Well, a sentence of yours in a previous post :



Even if FDSOI is the best process for RF applications..?.

Actually they should be quite competitive in this department, firms that adopted FDSOI did so for the RF performance, basically lower input capacitances wich allow for higher frequencies when devices are used within their linear region.

I read soironx technical explanation - it was actually very good imo - and yeaa on smaller dies it makes much technical sense to use it. But the basic business model must work before that and it just doesnt. Here is why:

The way i see it is that tsmc samsung and intel have a huge steady flow of old equipment needing work. 7nm will get very old in 8 years.
They of anybody will be interested in getting that to work. And its not small matters. Its 3 digit b of capex that needs work and a good depreciation profile
If that reqires adapting it to eg fdsoi then thats what they will do. Fdsoi isnt excactly rocket science anymore for companies that size. It pales in comparison to the capital investments. Why should they leave the market to gf?

If its to little market its not profitable. But then it isnt either for gf.

 

[NostaSeronx]

Current Rumors:
x Mubadala Technology is seeking a major stake or outright ownership of United Microelectronics Corporation
1a. They do not plan to fuse UMC and GlobalFoundries. This is them having two foundries invested rather than one.
1b. There will be a multi-owned Fab facility in Abu Dhabi;
- UMC will be Fab 1 starting at 65nm DDC.
- GlobalFoundries will be Fab 2 starting at 45RF and plans for Fab 2B with 22FDX/12FDX.
-- It is currently planned to follow the Chengdu model. No date when it will be built.



x Mubadala Technology is orientating for cryo-hardend CMOS that can implement Qubits cheaper.
FDSOI based on: http://inac.cea.fr/en/Phocea/Vie_des_labos/Ast/ast.php?t=fait_marquant&id_ast=1213
http://www.3dic.org/News/2016/11/A_CMOS_silicon_spin_qubit
https://www.usherbrooke.ca/iq/en/re...d-the-next-breakthrough-in-qubit-fabrication/
If FDSOI can do it, so can DDC.
https://ieeexplore.ieee.org/document/8066592/

"The market for quantum computing is projected to be around $5 billion to $10 billion a year by Morgan Stanley in the next 10 years, while a report by Homeland Security Research estimates that the global market for quantum computing and technologies will grow at a CAGR of 24.6% throughout 2018-2024.

By 2024, products and services market would be $8.45 billion while government-funded research and development would be $2.25 billion. According to CIR, revenue from quantum computing could reach $8 billion by 2027." - https://www.nasdaq.com/article/quantum-computing-what-it-is-and-who-the-major-players-are-cm939998

===
There is also double/triple dipping with:
http://www.cognit.com/
https://www.mubadala.com/en/news/co...urces-authority-introduce-technology-based-hr
https://www.mubadala.com/en/news/yahsat-completes-thuraya-acquisition-and-appoints-new-ceo