Discussion Leading Edge Foundry Node advances (TSMC, Samsung Foundry, Intel) - [2020 - 2025]

[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.

AnandTech Forums: Technology, Hardware, Software, and Deals

Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

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.

AnandTech Forums: Technology, Hardware, Software, and Deals

Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

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.

---------------------------------------------------------------------------------------------------------------------------------------------------


FEEL FREE TO CREATE A NEW THREAD FOR 2025+ OUTLOOK, I WILL LINK IT HERE

 

[Thunder 57]

Oh I agree and I've said before the long term effect will be that China is the first to make FEL lithography work which will be significantly better and cheaper than ASML's EUV roadmap. Yes it looks like the west is finally investing a bit in FEL (xLight) but unless it gets some real money behind it we'll be hopelessly behind China and chip fabrication will be another industry they win at.

All that the embargoes and limitations on the tech allowed to be transferred to China has done is made them realize they need to be self sufficient in every relevant technology. And with a command economy they can actually make that happen, while in the west we have big holes in our tech portfolio where China is much cheaper due to either labor cost or economies of scale, so no one is willing to invest without government subsidy (i.e. the "capitalist" version of a command economy)

China has succeeded because of IP theft and a large population that is declining (less than the West though). I'll put it another way, I'll believe it when I see it. No way the West gives up it's lead in fab without a fight.

 

[Hitman928]

Make of it what you will:

Nvidia recently tested out whether it would manufacture its chips using Intel’s production process known as 18A but stopped moving forward, two people familiar with the matter said. Nvidia did not answer a request for comment.

https://www.reuters.com/world/us/how-silicon-valley-dealmaker-charmed-trump-gave-intel-lifeline-2025-12-24/

 

[jdubs03]

Make of it what you will:

https://www.reuters.com/world/us/how-silicon-valley-dealmaker-charmed-trump-gave-intel-lifeline-2025-12-24/

CNBC folks were talking about it. Their conclusion was slight negative in the near-term.

I was reading the transcript from the Barclays Technology Conference from two-weeks ago; and the main investor liaison for Intel basically said 18AP and 14A were the nodes to look out for regarding external customers.

IMO, that’s a bit of goal post moving. But interestingly one of the analysts in that CNBC segment references 14A as the one of extreme importance.

 

[Geddagod]

Make of it what you will:

https://www.reuters.com/world/us/how-silicon-valley-dealmaker-charmed-trump-gave-intel-lifeline-2025-12-24/

More of what we already know. No one wants to use 18A. Onto 14A hopium.

 

[fastandfurious6]

No one wants to use 18A. Onto 14A hopium.

man... how can it be the leading edge number but they screw up so bad nobody wants it?

even much older processes have tons of customers, how come 18A can't find any?

 

[aigomorla]

If i read one more line about Taiwan that doesn't relate directly to TSMC i'll be burying you so deep that you will see CHINA.

No more political garbage in this thread PERIOD.
Next person will get double the points and hence will be face planted so deep you will see China.

Moderator Aigomorla

 

[jpiniero]

If i read one more line about Taiwan that doesn't relate directly to TSMC

What do you think the T in TSMC stands for?

 

[regen1]

Won't read too much into "Nvidia testing 18A and stopped moving forward".
It's not at if Rubin wasn't long ago committed to N3P(Nvidia 3N?), also not to forget Nvidia's dependency on both TSMC's packaging for its accelerators.

 

[fastandfurious6]

If i read one more line about Taiwan that doesn't relate directly to TSMC i'll be burying you so deep that you will see CHINA.

No more political garbage in this thread PERIOD.
Next person will get double the points and hence will be face planted so deep you will see China.

Moderator Aigomorla

With all due respect, and feel free to delete but not punish/ban as these come from very genuine place:

What is the real rationale for this warning - and post deletions - based on this forum's rules? Just a blanket "no politics"?

What is "political garbage" here? Why so strong/angry language? And why is specifically Taiwan matters prohibited while very frequently US political matters are discussed without any warnings/interventions?


All in all: cutting edge lithography globally has massive political stakes right now, increasing. Any proper discussion about it is inherently political...




You cannot comment publicly on a moderator's actions. We have told you this two times before.
read the rules.

AnandTech Forum Guidelines

AnandTech Forum Guidelines Introduction In an effort to make our management of the forums as transparent as possible, we have refined our member and posting guidelines and we have created a set of moderator guidelines. It is our hope that we can better satisfy members by being upfront and...

forums.anandtech.com

esquared
Anandtech Forum Director

 

[jpiniero]

Won't read too much into "Nvidia testing 18A and stopped moving forward".

Because nVidia was never serious about it. It was mainly to just scam "Investors"

I think there's a chance nVidia will fab some products at Samsung but that's just speculation.

 

[aigomorla]

What do you think the T in TSMC stands for?

And i dare you to cross that line... Believe me it wont end up good.


Again... leave politics out of this, as it has no purpose in TECH period.
If you want to rant about politics do it in politics and news. Tech should be Tech, and not a long list on who or what Taiwan is, unless as you stated, its talking about TSMC... which i know stands for Taiwan Semiconductor Manufacturing Company, hence why i said thats the only time i want to hear about Taiwan.

 

[regen1]

Because nVidia was never serious about it. It was mainly to just scam "Investors"

I think there's a chance nVidia will fab some products at Samsung but that's just speculation.

Currently from Nvidia's PoV, unless they are capacity limited why use Intel, even if they have decent node? (Cons: Jeopardize TSMC relationship? help a potential competitor ? ) (Pros: Diversified supply chain ? may be they don't care as much about this for now). Wafer pricing is least of their worries for now.

Samsung is more "proven" than Intel when it comes to foundry model. Intel hasn't manufactured much for outside players relatively and is a bit far away in PDK and IP game.

Packaging is a different issue, Nvidia will likely use Intel's packaging at least for some consumer products like Serpent Lake.

Spoiler

Nvidia is a part of RAMP-C, they got to test 18A wafers even outside of say one to one fab-customer engagement.
If one has to do dGPUs at Intel, it has to be 18A-P and not vanilla 18A that's assuming 18A-P has all the goods.

 

[511]

Packaging is a different issue, Nvidia will likely use Intel's packaging at least for some consumer products like Serpent Lake.

Serpent lake is a Intel product with Nvidia doing the GPU Tile so ofc it would be packed at Intel lmao

 

[regen1]

Serpent lake is a Intel product with Nvidia doing the GPU Tile so ofc it would be packed at Intel lmao

Yeah, it is Intel product(hint: Lake), Nvidia kinda would use Intel to do OEM work rather than how they were doing a lot of it with N1X/GB10. They should have let Nvidia have it as their product and do the dirty work. (Hope Intel can play these partnership deals with Nvidia well enough but many doubt the ability of present Board including the CEO).
Even then they are more likely to use Intel packaging for their products first than their nodes.

 

[511]

Yeah, it is Intel product(hint: Lake), Nvidia kinda would use Intel to do OEM work rather than how they were doing a lot of it with N1X/GB10. They should have let Nvidia have it as their product and do the dirty work. (Hope Intel can play these partnership deals with Nvidia well enough but many doubt the ability of present Board including the CEO).
Even then they are more likely to use Intel packaging for their products first than their nodes.

With the timing of the product it may be Intel fabbed and could serve as Nvidia Test Chip on IFS

 

[regen1]

With the timing of the product it may be Intel fabbed and could serve as Nvidia Test Chip on IFS

That's if the GPU tile is manufactured on an Intel node and not TSMC's (RGT rumor suggests TSMC node). We'll have to wait for confirmation.

 

[511]

That's if the GPU tile is manufactured on an Intel node and not TSMC's (RGT rumor suggests TSMC node). We'll have to wait for confirmation.

this has the most chance of being on IFS than anything else imo

 

[marees]

Just a thought:

Could Samsung prioritize memory allocation to the customers of their own fabs first ?

is that at all possible (to bundle apu fabbing with memory allocation)

https://twitter.com/x/status/2004137121171624359

Trillion dollar companies, the world's 3rd & 4th largest — google & Microsoft — fail in getting memory

https://twitter.com/x/status/2004120319456694625

 

[511]

Just a thought:

Could Samsung prioritize memory allocation to the customers of their own fabs first ?

is that at all possible (to bundle apu fabbing with memory allocation)

They memory company has pricing power over everyone now including Nvidia/MSFT/Google they can't change old contract prices though

 

[Win2012R2]

they can't change old contract prices though

Old contracts expire and a memory company can suggest to terminate it early by mutual agreement OR else (such as not getting any mem allocation in 2026)

 

[511]

Old contracts expire and a memory company can suggest to terminate it early by mutual agreement OR else (such as not getting any mem allocation in 2026)

At this point of time no one will agree to this

 

[Doug S]

Old contracts expire and a memory company can suggest to terminate it early by mutual agreement OR else (such as not getting any mem allocation in 2026)

They "expire" yes but rarely will you have a large contract with a supplier that all expires at once, unless you want flexibility to discontinue with that supplier (i.e. use as leverage when it comes time to renegotiate)

Whether you are contracting with a supplier to insure supply or to maintain price stability, you'll generally have multiple smaller rolling contracts. So maybe every 6 months one of your four 2 year supply contract expires and if isn't renewed (for whatever reason) you still have three other 2 year supply contracts still running.

That way if there's a surprise on their end and they refuse to agree to a new deal or you find their terms unreasonable, it doesn't have as big of an impact either cost basis wise (if you have to pay more with a new supplier, or prices are just going up in general) or supply wise.

The supplier gets benefit from splitting contracts like this too. That way if the customer is going to go another way they are losing their business over time instead of all at once, giving them more time to secure replacement customers or adjust their production plans.

 

[Joe NYC]

Here, from SemiWiki (via Twitter screen grab), Daniel Nenni hints that it is highly unlikely that either Jense Huang or Lisa Su will break away from TSMC and go to Intel.

https://twitter.com/x/status/2004911052115640669

 

[jdubs03]

Here, from SemiWiki (via Twitter screen grab), Daniel Nenni hints that it is highly unlikely that either Jense Huang or Lisa Su will break away from TSMC and go to Intel.

https://twitter.com/x/status/2004911052115640669

Seems odd that in his post he said the external customers with Intel 18A will be HVM in 2028-2030. I’m pretty sure that should be 14A/-E.

 

[Joe NYC]

Seems odd that in his post he said the external customers with Intel 18A will be HVM in 2028-2030. I’m pretty sure that should be 14A/-E.

Maybe 18A-P node, that is supposed to be more foundry friendly.

Maybe it is 2028-2030 because these products haven't taped out yet, and we can already consider to be in 2026, since nothing is getting done in next 4 days.