Is there any talk about them getting back into more leading edge processes? I have to assume that with the (US) government taking a real interest they are looking at them as a local resource for new cool shit for the technology sector of the economy not just satisfying old crusty DoD contracts.
GF is highly unlikely to produce 10nm and smaller - the required investment would be too large. Only TSMC, Samsung and Intel are large enough to be able to afford 10nm and smaller. (A single modern FAB can cost over three times GF annual revenue (not just profit).) GF will stick to the easier processes.
I think GF is already producing 12nm wafers, and has decided to stop working on its 7nm process sometime last year. The 12nm wafers it produces are a small improvement over its 14nm process.
"GF is highly unlikely to produce 10nm and smaller - the required investment would be too large. Only TSMC, Samsung and Intel are large enough to be able to afford 10nm and smaller. (A single modern FAB can cost over three times GF annual revenue (not just profit).) GF will stick to the easier processes."
They already invested in 7nm and 7nm EUV process, they also have at least two EUV machine AFAIK. If they want to stay relevant, they have to move to 7nm anyway, soon or later, but they simply have to. Of course the others will be by then on 5/3nm technology
no, they didn´t cancelled anything: "Despite having installed at least one EUV machine at its Fab 8 facility in Malta, N.Y., all those plans are now on indefinite hold, the company announced Monday."
Not much high end. Reporderly 3nm is barely an half node over 5nm (there is an intersting article here on Anandtech). This the reason Intel will stay on 7nm for a lot, but Intel will produce low end i3 cpus on TSMC 3nm. This is a clear indication that 3nm is only 40% in average more compact versus 5nm and Intel 7nm is 20% more compact of TSMC 5nm. There is an advantage over compactness, likely Intel 7nm is capable of far higher clock speeds versus 3nm with only a small disasvantage in silicon area. The power consumption will be fixed with better materials. We all know 10nm++ have 15% less power consumtion over 10nm+. We barely will be able to distinguish a Genoa from a Whitley Lake from a power consumption point of view. In short words we can not give much credit to 3nm, it was thinked mainly for Phone SOCs and performance like second chance. It have sense for Intel produce low margins and low clocked SKUs on TSMC new half node. IMO Tsmc 5nm will be the last chance for AMD to push on process, sub 5nm is only a tradeoff, the reason?? 3nm require EUV double patterning, that is an seppuku on Foundry margins. Very happy Intel to avoid this silicon suicide. LIkely TSMC will go in red on 3nm Fabs. The real 3nm will born several years from now with the new EUV scanners capable of that geometries with a single exposure. Unfortunately these machines are VERY late.
Global Foundries is the ITANIUM of the chip manufacturers. Brillaint power points, flush with SWF money, buying/rolling over competitors and still managed to become worse and worse.
They literally had 12nm FD-SOI gifted from IBM when they acquired IBM micro and are unable to productize/commercialize it.
Now that the Gulf money had stopped, they are now leeching on subsidies by the US and EU governments. LOL!
If SMSNG opens its latest fab in TX (3nm and below) and TSMC's latest (3nm below)rumored in AZ, why will US Gov pay GF for 45nm production??
Because the use cases don't require leading edge tech. DoD projects have timelines measured in decades, and once a product is certified they don't just swap in the latest chips every year. If the purchase order was for 10million 45nm chips over the next five years, that's what GF will produce. If a processor is going into a fighter jet, you'll be making that same chip for the next 15-20 years; once the design and product is certified it does not change unless there's a major overhaul.
There are several issues with leading edge chips. A new architecture could possibly have undiscovered security flaws. They are expensive. They don't have decade+ track records. Plus for some applications smaller architectures themselves are a problem. Radiation hardened chips are made on "old" 65nm and 45nm processes because they want the transistors physically spread out. When you cram the transistors together in a 14nm or 10nm design they are far more susceptible to developing (multiple) errors if high energy particles hit them. Just like regular silicon becomes flooded with current leakage below 5nm, for military/space applications you cannot risk the processors flipping bits if an energetic particle hits them.
And again, they don't need or want some 125W 5ghz processor that costs an arm and a leg when a trusty 10w 3ghz processor for 1/10th the cost gets the job done.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
21 Comments
Back to Article
Operandi - Thursday, March 4, 2021 - link
Is there any talk about them getting back into more leading edge processes? I have to assume that with the (US) government taking a real interest they are looking at them as a local resource for new cool shit for the technology sector of the economy not just satisfying old crusty DoD contracts.Duncan Macdonald - Thursday, March 4, 2021 - link
GF is highly unlikely to produce 10nm and smaller - the required investment would be too large. Only TSMC, Samsung and Intel are large enough to be able to afford 10nm and smaller. (A single modern FAB can cost over three times GF annual revenue (not just profit).) GF will stick to the easier processes.Calin - Thursday, March 4, 2021 - link
I think GF is already producing 12nm wafers, and has decided to stop working on its 7nm process sometime last year.The 12nm wafers it produces are a small improvement over its 14nm process.
Mobile-Dom - Thursday, March 4, 2021 - link
its 12nm FF wasnt much of an increase, the 12nm FD-SOI (12FDX) was pretty impressive if what I read was correctksec - Thursday, March 4, 2021 - link
I think GF focus on FD-SOI is the right path. It is hard to compete against Intel ( funded by x86 CPU ), Samsung ( funded by NAND and DRAM ) and TSMC.del42sa - Friday, March 5, 2021 - link
12FDX is halted and postponed to 2024. It was originally scheduled to begin in 2020del42sa - Friday, March 5, 2021 - link
"GF is highly unlikely to produce 10nm and smaller - the required investment would be too large. Only TSMC, Samsung and Intel are large enough to be able to afford 10nm and smaller. (A single modern FAB can cost over three times GF annual revenue (not just profit).) GF will stick to the easier processes."They already invested in 7nm and 7nm EUV process, they also have at least two EUV machine AFAIK. If they want to stay relevant, they have to move to 7nm anyway, soon or later, but they simply have to. Of course the others will be by then on 5/3nm technology
scineram - Friday, March 5, 2021 - link
No.del42sa - Saturday, March 6, 2021 - link
yesSpunjji - Monday, March 8, 2021 - link
That would be a no - they cancelled their 7nm development more than two years ago:https://www.anandtech.com/show/13277/globalfoundri...
Spunjji - Monday, March 8, 2021 - link
No. GF cancelled 7nm development more than 2 years ago:https://www.anandtech.com/show/13277/globalfoundri...
del42sa - Tuesday, March 9, 2021 - link
no, they didn´t cancelled anything:"Despite having installed at least one EUV machine at its Fab 8 facility in Malta, N.Y., all those plans are now on indefinite hold, the company announced Monday."
https://spectrum.ieee.org/nanoclast/semiconductors...
scineram - Tuesday, October 5, 2021 - link
Cope.haukionkannel - Friday, March 5, 2021 - link
Exactly! GF don`t have money for highend nodes!Gondalf - Sunday, March 7, 2021 - link
Not much high end. Reporderly 3nm is barely an half node over 5nm (there is an intersting article here on Anandtech).This the reason Intel will stay on 7nm for a lot, but Intel will produce low end i3 cpus on TSMC 3nm.
This is a clear indication that 3nm is only 40% in average more compact versus 5nm and Intel 7nm is 20% more compact of TSMC 5nm.
There is an advantage over compactness, likely Intel 7nm is capable of far higher clock speeds versus 3nm with only a small disasvantage in silicon area. The power consumption will be fixed with better materials. We all know 10nm++ have 15% less power consumtion over 10nm+.
We barely will be able to distinguish a Genoa from a Whitley Lake from a power consumption point of view.
In short words we can not give much credit to 3nm, it was thinked mainly for Phone SOCs and performance like second chance. It have sense for Intel produce low margins and low clocked SKUs on TSMC new half node.
IMO Tsmc 5nm will be the last chance for AMD to push on process, sub 5nm is only a tradeoff,
the reason?? 3nm require EUV double patterning, that is an seppuku on Foundry margins.
Very happy Intel to avoid this silicon suicide. LIkely TSMC will go in red on 3nm Fabs.
The real 3nm will born several years from now with the new EUV scanners capable of that geometries with a single exposure. Unfortunately these machines are VERY late.
HyperText - Friday, March 5, 2021 - link
Good news for ASML :)rocketbuddha - Friday, March 5, 2021 - link
Global Foundries is the ITANIUM of the chip manufacturers. Brillaint power points, flush with SWF money, buying/rolling over competitors and still managed to become worse and worse.They literally had 12nm FD-SOI gifted from IBM when they acquired IBM micro and are unable to productize/commercialize it.
Now that the Gulf money had stopped, they are now leeching on subsidies by the US and EU governments. LOL!
If SMSNG opens its latest fab in TX (3nm and below) and TSMC's latest (3nm below)rumored in AZ, why will US Gov pay GF for 45nm production??
Smell This - Friday, March 5, 2021 - link
Y A W N
Kakti - Saturday, March 6, 2021 - link
Because the use cases don't require leading edge tech. DoD projects have timelines measured in decades, and once a product is certified they don't just swap in the latest chips every year. If the purchase order was for 10million 45nm chips over the next five years, that's what GF will produce. If a processor is going into a fighter jet, you'll be making that same chip for the next 15-20 years; once the design and product is certified it does not change unless there's a major overhaul.There are several issues with leading edge chips. A new architecture could possibly have undiscovered security flaws. They are expensive. They don't have decade+ track records. Plus for some applications smaller architectures themselves are a problem. Radiation hardened chips are made on "old" 65nm and 45nm processes because they want the transistors physically spread out. When you cram the transistors together in a 14nm or 10nm design they are far more susceptible to developing (multiple) errors if high energy particles hit them. Just like regular silicon becomes flooded with current leakage below 5nm, for military/space applications you cannot risk the processors flipping bits if an energetic particle hits them.
And again, they don't need or want some 125W 5ghz processor that costs an arm and a leg when a trusty 10w 3ghz processor for 1/10th the cost gets the job done.
Spunjji - Monday, March 8, 2021 - link
Hell, for space applications they tend to run in the Mhz range.tygrus - Sunday, March 7, 2021 - link
GF would need to license sub-12nm from existing top3 or offer facility as joint venture with one of those top3 (TSMC, Samsung; unlikely to be Intel).