Discussion Leading Edge Foundry Node advances (TSMC, Samsung Foundry, Intel)

[DisEnchantment]

TSMC's N7 EUV is now in its second year of production and N5 is contributing to revenue for TSMC this quarter. N3 is scheduled for 2022 and I believe they have a good chance to reach that target.


N7 performance is more or less understood.


This year and next year TSMC is mainly increasing capacity to meet demands.

TSMC Radically Boosts CapEx to Expand Production Capacities, To Reach $14B For 2019

www.anandtech.com

For Samsung the nodes are basically the same from 7LPP to 4 LPE, they just add incremental scaling boosters while the bulk of the tech is the same.

Samsung is already shipping 7LPP and will ship 6LPP in H2. Hopefully they fix any issues if at all.
They have two more intermediate nodes in between before going to 3GAE, most likely 5LPE will ship next year but for 4LPE it will probably be back to back with 3GAA since 3GAA is a parallel development with 7LPP enhancements.

Samsung’s Aggressive EUV Plans: 6nm Production in H2, 5nm & 4nm On Track

www.anandtech.com

Samsung's 3GAA will go for HVM in 2022 most likely, similar timeframe to TSMC's N3.
There are major differences in how the transistor will be fabricated due to the GAA but density for sure Samsung will be behind N3.
But there might be advantages for Samsung with regards to power and performance, so it may be better suited for some applications.
But for now we don't know how much of this is true and we can only rely on the marketing material.

This year there should be a lot more available wafers due to lack of demand from Smartphone vendors and increased capacity from TSMC and Samsung.
Lots of SoCs which dont need to be top end will be fabbed with N7 or 7LPP/6LPP instead of N5, so there will be lots of wafers around.

Most of the current 7nm designs are far from the advertized density from TSMC and Samsung. There is still potential for density increase compared to currently shipping products.
N5 is going to be the leading foundry node for the next couple of years.

For a lot of fabless companies out there, the processes and capacity available are quite good.

 

[Ajay]

Hmm, posted in the wrong thread. My apologies.

 

[biostud]

I don't care where they build, so long as its not communist or dictator ruled. Diversification will give EVERYONE more security and capacity.

Except for Taiwan, I don't know if they like the idea of TSMC moving production to other parts of the world. :/

 

[DrMrLordX]

Just posting an update on the hold up

It's funny, because Intel is already receiving subsidies from the EU. What kind of a threat is that?

Given Intel's process struggles, it's lamentable that anyone feels compelled to subsidize them. Why aren't foundries like Globalfoundries being considered? If all governments care about is automotive and military applications, FDX nodes should fit the bill in the majority of cases. And at least GF is (comparatively) honest about what they can and can't accomplish with their fab tech.

Except for Taiwan, I don't know if they like the idea of TSMC moving production to other parts of the world. :/

It will earn money for the RoC and reduce any country's incentive to attack their island. What's not to like?

 

[Ajay]

It's funny, because Intel is already receiving subsidies from the EU. What kind of a threat is that?

Given Intel's process struggles, it's lamentable that anyone feels compelled to subsidize them. Why aren't foundries like Globalfoundries being considered? If all governments care about is automotive and military applications, FDX nodes should fit the bill in the majority of cases. And at least GF is (comparatively) honest about what they can and can't accomplish with their fab tech.

I think they are still owned by the UAE.

 

[NostaSeronx]

Why aren't foundries like Globalfoundries being considered?

They are considered, though...
- Singapore expansion -> New Fab for 2023
- US expansion -> Building capability guaranteed for Fab 8.2
- EU expansion -> Co-fab with STM in France.


Some of the stuff is held off till announcement of small step roadmap of 12/11/10FDX.
12 => 84Cx/56Mx
11 => 64Cx/56Mx
10 => 64Cx/48Mx

And new undoped 12Fin(84Cx/64Mx) and 11/10Nex(same as FDX).

 

[Roland00Address]

Given Intel's process struggles, it's lamentable that anyone feels compelled to subsidize them. Why aren't foundries like Globalfoundries being considered? If all governments care about is automotive and military applications, FDX nodes should fit the bill in the majority of cases. And at least GF is (comparatively) honest about what they can and can't accomplish with their fab tech.

Some of the military and automotive things will be covered by FDX and any bulk 14nm or older. But not everything.

Likewise what is the "boundaries" of profane vs excellent use of national security is up for debate. It is what it is. Some people see this as perversity others do not, and no amount of shaming by one group is going to change the minds of the other group for they see each other as too separate that they do not care and shaming does not work.

 

[DrMrLordX]

I think they are still owned by the UAE.

Mubadala specifically. Though GF has (and has long had) foundries in the United States and Germany (and Singapore). They have no foundries in China. Intel, on the other hand . . .

 

[DrMrLordX]

Likewise what is the "boundaries" of profane vs excellent use of national security is up for debate. It is what it is. Some people see this as perversity others do not, and no amount of shaming by one group is going to change the minds of the other group for they see each other as too separate that they do not care and shaming does not work.

That ignores the practical reality that Intel is presenting itself as a leading-edge foundry while proving that it is no longer a leading-edge foundry.

 

[Ajay]

That ignore the practical reality that Intel is presenting itself as a leading-edge foundry while proving that it is no longer a leading-edge foundry.

Fake it to you make it?

 

[Roland00Address]

That ignore the practical reality that Intel is presenting itself as a leading-edge foundry while proving that it is no longer a leading-edge foundry.

The military* insists on multiple suppliers as a matter of normalcy. TSMC has US fabs in development, like Intel is number #2 with leading edge. *shrug*

*(there are multiple parts of the government each with their own interests and value processes with how to do things.)

How subsidies occur, with the details, is more important than subsidies yes or no. Guaranteeing demand but allowing private sector to get first dibs if demand is high is a reasonable way of structuring a subsidy. This means factories get built with the goal of some flexibility of how much “extra” projected use in supply.

 

[pakotlar]

That ignore the practical reality that Intel is presenting itself as a leading-edge foundry while proving that it is no longer a leading-edge foundry.

Leading edge seems to refer to multi-mask DUV and single mask EUV, and more generally the pursuit of bleeding edge design. Intel isn’t claiming that they have the leading process.

 

[Thibsie]

Leading edge seems to refer to multi-mask DUV and single mask EUV, and more generally the pursuit of bleeding edge design. Intel isn’t claiming that they have the leading process.

No just claiming they will in very short order lol

 

[Doug S]

The military* insists on multiple suppliers as a matter of normalcy. TSMC has US fabs in development, like Intel is number #2 with leading edge. *shrug*

The US fabs, like all TSMC fabs outside of Taiwan, are not leading edge. This is an N5 fab scheduled to begin mass production in 2024 (assuming the pandemic and CHIPS Act delays haven't pushed that back) when N3E will be in mass production in Taiwan and N2 will be around the corner.

Though as far as the US military and traditional manufacturing (autos etc.) needs that's fine, they don't need bleeding edge. At least not until self driving cars become feasible.

 

[DrMrLordX]

The military* insists on multiple suppliers as a matter of normalcy.

All the more reason for the gub'ment to cut in Globalfoundries.

Intel is number #2 with leading edge. *shrug*

I would argue that Samsung is pulling ahead of them. 3GAE isn't all that great, and the yields may currently be bad, but at least it exists. It certainly seems to be further along than Intel 4. By the time 3GAP enters risk production, Samsung will likely have the lead.

Intel isn’t claiming that they have the leading process.

Sure seems like they are. All that noise about Intel 4, 3, and 20a . . .

 

[pakotlar]

All the more reason for the gub'ment to cut in Globalfoundries.



I would argue that Samsung is pulling ahead of them. 3GAE isn't all that great, and the yields may currently be bad, but at least it exists. It certainly seems to be further along than Intel 4. By the time 3GAP enters risk production, Samsung will likely have the lead.



Sure seems like they are. All that noise about Intel 4, 3, and 20a . . .

They’re claiming that they’ll hit “node supremacy” by 2025 (previously 2024-2025), with 18A, but aren’t saying they’re there yet.

 

[Roland00Address]

All the more reason for the gub'ment to cut in Globalfoundries.

agreed

 

[JasonLD]

All the more reason for the gub'ment to cut in Globalfoundries.

Unless they are interested in doing sub-7nm process again, Intel is pretty much the only US company with a chance to compete against TSMC and Samsung.

 

[Ajay]

Mubadala specifically. Though GF has (and has long had) foundries in the United States and Germany (and Singapore). They have no foundries in China. Intel, on the other hand . . .

Good point. One would think, with GF hack g s plant in the Senate Majority Leader’s state, that GF would be on the list of companies to receive funds. I do think that much of the funding is for new plants, equipment and the on-shoring of support businesses for a more robust supply chain. Guess I need to find the bullet points again.

 

[Exist50]

GlobalFoundries is interesting from the point of subsidies. On one hand, their technology stack, despite being unsuited for modern application processors, seems like a good match for the automotive market and others hit most strongly by the chip shortage. But on the other hand, they have distinctly lower growth prospects, and can't advertise "job creation" as strongly as Intel can. Given how short political memory is, that's probably going to increase in relative importance for the sake of funding. Especially with already promised developments.

Also, Intel's clearly failed hard over the last decade... but has GloFo been much better? They've been widely panned for missed promised pretty much since they were spun off (32nm was supposed to hit what? 6GHz?), had to license 14nm from Samsung to get a working FinFET node, and then completely dropped their attempts at 7nm. Perhaps they do deserve a part of the pie, but they don't have a terribly strong argument for "deserving" it either.

 

[NostaSeronx]

... but has GloFo been much better? They've been widely panned for missed promised pretty much since they were spun off (32nm was supposed to hit what? 6GHz?), had to license 14nm from Samsung to get a working FinFET node, and then completely dropped their attempts at 7nm. Perhaps they do deserve a part of the pie, but they don't have a terribly strong argument for "deserving" it either.

32/28:
32SHP (Q1 2011): 50% higher frequency vs 45nm (IBM's 2009-32nm was 35% higher frequency)
28HP (Q3 2011): 40% higher frequency

FinFET:
It was actually reversed GlobalFoundries 14XM was working since Q1 2013. They then switched to Samsung to get a working node in Q4 2014.
The collaboration meant GlobalFoundries had to operate with Samsung's schedule. It also killed off 20LPM: Qualcomm's 20LPQ and AMD's 20AN, there was also three other ARM-specific companies that were on 20LPM/14XM at the time. AMD became the only customer at GlobalFoundries for 14LPP. Till 14LPP-RF popped up and Qualcomm finally returned. Most companies running 14LPP(/12LP/12LP+)@GF are built off crypto-bro ventures so that market is surely crashing now. They also don't have a high volume market at most 1 million units before they go to Samsung 8nm or TSMC 7nm.

GloFo had a fast-track for:
28SLP -> 20LPM -> 14XM
Each node had a custom variant specific to AMD/Qualcomm.
Had plans to run 20LPM/14XM at Fab 1 Module 2.
2013 Fab Booster: 540,000 wafer per year (2011) to 960,000 wafers per year (2013) included Fab 1 Module 2 moving towards 20nm/14nm. With the Annex(pseudo-Module 3) taking up older nodes.
AMD's wafer supply agreement for Dresden was 25,000 wafers per month. So, the switch to 14LPP killed huge amounts of revenue growth for GlobalFoundries.

GloFo didn't have a fast-track for:
28SLP -> 14LPP
Given hindsight, GlobalFoundries had to build a custom 14nm variant for AMD.
Zen Summit Ridge/Raven Ridge => 14LPPv2 (Custom FinFET with HPC overdrive) which was refined and resold as 12LP for Zen+.
Only a single foundry supported because Samsung didn't support AMD's custom variant. At the time AMD was exclusively TSMC+GF.
20LPM => 20SoC, 20AN => 20SoC+ && 14XM => 16FF, 14XA => 16FF+

GF28A = TSMC-esque 28nm node
GF20AN = TSMC-esque 20nm node
GF14XA = TSMC-esque FinFET node
AMD's main/highest volume product-line taped out first at TSMC, then taped out at GlobalFoundries. Before the fab boosts, AMD was guaranteed 50% wafer capacity for both Fab1/Fab8. In the run before Samsung collaboration, GlobalFoundries could have had high revenue at both fabs. Rather than running Fab 8 at a loss and Fab 1 not getting a new node till 2017.

7nm:
7nm is generally the fiasco of Malta being a money sink with no positive revenue. Mubadala wasn't going to what they did in 2014: $10B to speed up FinFETs only to get no return. Expected return by ATIC/Mubadala technology to Mubadala was 7 to 10 years for the 14LPP investment to turn a profit. From what info I can get for 7LP was that it was even worse like 20 to 30 years.

Money:
Only reason to get a huge chunk is to get Fab 8.2/8.3 up. Other than that it is just capacity security for a trusted foundry.

If I remember correctly the Trusted Foundry advisory has been dropped for leading edge. Intel 14GP > GF 14LPP(12LP/12LP+) is the preference in trusted microelectronics. The money going to GlobalFoundries will clearly not be used for FinFETs. Since, Intel is serviceable for 14nm and beyond with top-grade qualification when Trusted-group gets to tape-out.

My guess:
SiPh, Qubits, FDSOI, PDSOI, RF, eMemory, 3DIC, etc.

Given the revenue so far, GlobalFoundries has a faster return for the SOI-related nodes. So any money would have to go to expanding SOI production rather than erroneously restarting 7nm.

 

[DrMrLordX]

Unless they are interested in doing sub-7nm process again, Intel is pretty much the only US company with a chance to compete against TSMC and Samsung.

See below.

GlobalFoundries is interesting from the point of subsidies. On one hand, their technology stack, despite being unsuited for modern application processors, seems like a good match for the automotive market and others hit most strongly by the chip shortage. But on the other hand, they have distinctly lower growth prospects, and can't advertise "job creation" as strongly as Intel can. Given how short political memory is, that's probably going to increase in relative importance for the sake of funding. Especially with already promised developments.

Agreed on the automotive/military angle. GF doesn't have to be leading edge. 12FDX can and will last for a long time, and if GF actually gets subsidies, they can slowly piece together new nodes over time, either through cross-licensing with Samsung or IBM or through novel R&D. Or they can scrape up expired patents.

Also, Intel's clearly failed hard over the last decade... but has GloFo been much better?

Not really. But their failures and successes are known quantities, and they've stabilized in a foundry role that Intel still hasn't managed despite their much-ballyhooed Foundry 2.0 model. Today, with GF, you know exactly what you're going to get: 12nm+, 22FDX, and uh I guess 12FDX, though I don't know if they're shipping any 12FDX wafers yet (honestly not paying much attention to that rn). With Intel strutting around acting like they're going to be the premier foundry by 2025, you don't really know what you're going to get out of them if you subsidize them. Are you going to get 18a on schedule, or will you just get a lot of 10nm? Or 14nm and 22ffl? It's hard to tell.

If all the gub'ment really cares about is automotive, military, and commodity chips (like stuff that goes into monitors or whatever) then GF is a known quantity that can get the job done, presumably for less than $10 billion. Intel is busy funneling a lot of money into R&D as well as buying a ton of wafers from TSMC.

 

[JasonLD]

f all the gub'ment really cares about is automotive, military, and commodity chips (like stuff that goes into monitors or whatever) then GF is a known quantity that can get the job done, presumably for less than $10 billion. Intel is busy funneling a lot of money into R&D as well as buying a ton of wafers from TSMC.

Except it is more than just having those chips being developed on US soil, it is about having latest “leading edge” chips fabricated in US. Intel is pretty much the only US company capable of doing that.

 

[NostaSeronx]

Except it is more than just having those chips being developed on US soil, it is about having latest “leading edge” chips fabricated in US. Intel is pretty much the only US company capable of doing that.

Samsung America is technically a US company, even if it is a subsidiary of the Korean business.

Samsung's 2024 plan had that 3nm/2nm fab in Taylor, Texas.
"The company will start work on the site in the first half of 2022 with the target of having the facility operational in the second half of 2024. This implies starting with a 3nm capability with a view to 2nm production. This would be consistent with the $17bn investment in the building, infrastructure and equipment, but the company has not said whether there will be multiple fabs on the site."

Fab S2 Expansion(Austin new module) for 3nm (2021 build -> 2023 production)
Fab S4?/V2?(Taylor new fab) for 3nm and 2nm (2022 build -> 2024 production)

IBM is collaborating with both Intel and Samsung. It is very likely that 3nm/2nm && 20A/18A will be trusted certified by ~2025.
Intel - Oregon + Arizona + Ohio(unconfirmed to be canned) + New Mexico expansion(hasn't been announced yet)
Samsung - Austin + Taylor + two other fabs in the East U.S.(unannounced yet but set for 2026/2027 respectively)

Gate-all-around/surround-gate has cheap radiation-hardened efforts from avionics to aerospace.

though I don't know if they're shipping any 12FDX wafers yet (honestly not paying much attention to that rn).

12FDX in its first iteration was shipping to top-tier customers in late 2017 throughout 2018.
12FDX in its second iteration is supposed to be shipping to top-tier customers right now in this half of 2022.

12FDXv1(20nm gate length) = Same wafer and gate-stack as 22FDX, 14FD-derived. However, it had 10FD's Dual-BB, etc.
12FDXv2(14nm gate length) = New wafer(sSOI) and new gate-stack, 7FD-derived. The new roadmap past 12FDX(84Cx/56Mx) is for 11FDX(64Cx/56Mx)/10FDX(64Cx/48Mx). Which will be part of GlobalFoundries small steps roadmap, same tools for several node generations.

General reminder (same device to same device):
22FDXv1 (2016-2017) = Faster than 14LPP.
22FDXv2 (2019+ :: reduced channel height, more custom performance options, etc) = Faster than 14LPP.
22FDX+ (2021+) = Faster than 12LP+.
12FDXv1 (2019+ if it launched) = Faster than 12LP+ and slightly slower than 7LP.
12FDXv2 (2023+ :: tSOI with cSiGe, smaller gate, etc) = Faster than 7LP as fast as N5P.

However, most press will be for the older stuff:

GLOBALFOUNDRIES and SkyWater Technology Sign MOU for Technology Development to Strengthen Domestic Supply Assurance for U.S. Government - Skywater Technology

Collaboration will extend the domestic manufacturing supply capabilities of the U.S. and enable long-term access to cutting-edge, secure semiconductor solutions Santa Clara, Calif., and BLOOMINGTON, Minn. – June 18, 2020 – GLOBALFOUNDRIES® (GF®), the world’s leading specialty foundry and a...

www.skywatertechnology.com

US Department of Defense Awards SkyWater a $27M Option to Facilitate American-Made Semiconductors Critical to National Security and Domestic Supply Chains - Skywater Technology

Government funding enables SkyWater to further develop IP for its 90 nm rad-hard FDSOI platform to be used in strategic defense and space applications BLOOMINGTON, Minn. – July 12, 2022 – SkyWater Technology (NASDAQ: SKYT), the trusted technology realization partner, today announced the...

www.skywatertechnology.com

"Be fully synthesizable in at least two technology nodes, examples include Intel 22FFL, GlobalFoundries 12LP, and Skywater 90RH"
Intel 22FFL/14GP >== is about 60%
GF 45nm >== is 30%
SkyWater 90nm >= 10%
even though GF 12LP is listed on several websites, it is not utilized compared to 45nm. 2H2022 is supposed to bring up 22FDX and 12FDX production at Malta. Which might allow GF to steal 22FFL/14GP before they go officially in production in next-gen machines.

 

[DrMrLordX]

Intel is pretty much the only US company capable of doing that.

Heh! Not really. GF certainly can't, but Intel is really falling behind as well.

 

[JasonLD]

Heh! Not really. GF certainly can't, but Intel is really falling behind as well.

Capacities aside, Intel is the only other company out there doing sub 7nm besides TSMC and Samsung.