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

 

[oak8292]

From Intel's 10Q, under Risk Factors:

"If we are unable to secure a significant external foundry customer for Intel 14A, our next generation semiconductor manufacturing process technology, we may pause or discontinue our pursuit of next generation leading-edge process technologies, which may have significant strategic business, financial, operational and reputational risks and repercussions."

https://www.intc.com/filings-reports/all-sec-filings/content/0000050863-25-000109/0000050863-25-000109.pdf

Reality hits hard. Moore’s Law is economic and it is dead. GF, after spinning off from AMD didn’t have the volume for EUV development even as a foundry. Now Intel admits it can’t afford process development and fabs alone. Will they try for a generic foundry customer and an Intel only HP library on the 14A node?

 

[DavidC1]

They could have done both tbh packing for Nvidia plus selling theirs what's the point of having both design and foundry under one roof if you miss both opportunity also Intel has been terrible due to their mindset they never designed their process for GPUs and they kept canning projects one after another for their GPU.
Xe GPU IP will keep improving though PTL's highlight is it's GPU.

Here's the problem.

Whatever you call it, innovator's dilemma or something else, they were reactive rather than being proactive to fixing the solution. Let me tell you why 18A is not good for mobile and GPU.

Because they tried that on 22nm process and lost a lot because of it.

You see, we were expecting a lot on 22nm but it disappointed us, because we're PC'ers and they said themselves it was only optimized for the Silvermont Atom, their main focus back then. So their bread and butter product was being sacrificed to prop up their late entry to the market. And 14nm and 10nm problems and delays are due to this reason as well.

Traditionally Intel was behind in density, so they wanted to have their cake and eat it to on 14nm and 10nm, thus did too much, which caused an implosion on their most important skill - process technology. So sometime in the 10nm to Intel 4 era, Intel decided killing their cash-cow for a non-existent market was a bad idea, hence future processes were desktop optimized again.

Let me summarize:
-22nm/14nm: Too much focus on mobile, causing losses on PC
-10nm/Intel 4: Back to focus on desktop
-18A: Now they need foundry customers, but they need a switch back to mobile focus

Intel is a reactionary company run by accounting guys, that happens to hire engineers.

Celeron: a reaction to Via's Cyrix(designed by the lean and mean Centaur team)
Pentium M: a reaction to Transmeta
Core 2: a reaction to AMD

The problem with being reactive is that eventually the new market will kill you. It's better you kill your own product rather than someone else do it for you. All these products, that they were forced to pivot because of surprising market reaction were something Intel should have naturally made and then customers would have come.

 

[511]

Take it with a grain of salt.

 

[Win2012R2]

Low end Nvidia GPU on brand new 14A that needs High NA (=expensive)?

Total fantasy land, but they will of course take PDKs, make token effort and secure further X% discount at TSMC.

We'll know it soon because without firm commit (money on the barrel) Intel won't move on 14A anytime soon.

 

[511]

Low end Nvidia GPU on brand new 14A that needs High NA (=expensive)?

Total fantasy land, but they will of course take PDKs, make token effort and secure further X% discount at TSMC.

We'll know it soon because without firm commit (money on the barrel) Intel won't move on 14A anytime soon.

Yeah lol but if they can get area down to make economic sense they will do it also 14A is both high-na and non high-na so there is that.

 

[mikk]

Low end Nvidia GPU on brand new 14A that needs High NA (=expensive)?

Total fantasy land, but they will of course take PDKs, make token effort and secure further X% discount at TSMC.

We'll know it soon because without firm commit (money on the barrel) Intel won't move on 14A anytime soon.

High-NA is optional on Intel 14A. They can use Low-NA EUV if something goes wrong with High-NA.

 

[Joe NYC]

View attachment 127752
Take it with a grain of salt.

I am totally taking it with a grain of salt - that Intel would fab their iGPUs and dGPUs at TSMC while NVidia would fab their at Intel fabs

 

[Win2012R2]

High-NA is optional on Intel 14A. They can use Low-NA EUV if something goes wrong with High-NA.

It might be optional, but as Intel say themselves they greatly reduce number of steps with High-NA, so yields will go down then.

It's not happening anyway - they won't get big external customer who will bankroll it upfront in the next 12 months.

It's over.

 

[The Hardcard]

View attachment 127752
Take it with a grain of salt.

Don’t rule out the Trump factor. I don’t think it would make a company use a significantly inferior process, but if Intel is close behind I can see it nudging the decision. Three years is a long time to get politically and socially hammered.

 

[511]

I am totally taking it with a grain of salt - that Intel would fab their iGPUs and dGPUs at TSMC while NVidia would fab their at Intel fabs

Nope Xe3P is 18AP based they are taking their GPUs home next gen

It might be optional, but as Intel say themselves they greatly reduce number of steps with High-NA, so yields will go down then.

It's not happening anyway - they won't get big external customer who will bankroll it upfront in the next 12 months.

It's over.

And the customers will regret when TSMC will increase the price again with 0 bargaining power.

 

[gdansk]

It may be dead already and they simply are not committing yet. But I can see some big companies hedging their bets for the 14A timeframe. And if the rumors about Samsung are accurate then it seems plausible Intel would be plan B, not Samsung.

 

[Win2012R2]

Nope Xe3P is 18AP based they are taking their GPUs home next gen

Well it's too late to change that now, but they need to think about their next 5 years very carefully or they can be the last one.

And the customers will regret when TSMC will increase the price again with 0 bargaining power.

Yeah, but that will be day after tomorrow, everyone will be in the same boat, so just inflate prices and add up extra markup to improve margins, perfect opportunity - as long as new node is greatest and latest blah blah.

Plus who in their right mind will risk whole business just to keep TSMC at bay? Well, Altera actually did it.

But I can see some big companies hedging their bets for the 14A timeframe.

The only way it could work if Fabs spinned off (maybe just new ones), and those who pre-pay (and they will have to) will get a BIG chunk of shares in Fabs they are paying to actually make. Still, too big risk, why upset the gravy train when things go so well?

 

[511]

Well it's too late to change that now, but they need to think about th

It's not too late it will launching in H1 27 most likely just like most of the next gen cards.

Yeah, but that will be day after tomorrow, everyone will be in the same boat, so just inflate prices and add up extra markup to improve margins, perfect opportunity - as long as new node is greatest and latest blah blah.

Plus who in their right mind will risk whole business just to keep TSMC at bay? Well, Altera actually did it.

Oh ofc TSM. Can 1.5X wafer Price they can't do sh** just seeing it.

The only way it could work if Fabs spinned off (maybe just new ones), and those who pre-pay (and they will have to) will get a BIG chunk of shares in Fabs they are paying to actually make. Still, too big risk, why upset the gravy train when things go so well?

They will kill 14A and Will not spin off IFS and will milk 18A they have said it in the Call.

 

[LightningZ71]

From 3nm class and beyond, currently, there are only two players in town: TSMC and Intel. Intel3 is a solid node for Intel's internal usage. 18a-ap will suffice for the next few years. It's far enough along to conclude that it's better than anything Samsung is likely to ever produce in any sort of economical volume.

The only two wild cards are Rapidus, which I think will be volume limited no matter what for a very long time, and China's various efforts. They WILL eventually be competitive, the question is time.

 

[511]

From 3nm class and beyond, currently, there are only two players in town: TSMC and Intel. Intel3 is a solid node for Intel's internal usage. 18a-ap will suffice for the next few years. It's far enough along to conclude that it's better than anything Samsung is likely to ever produce in any sort of economical volume.

My biggest complain with Intel 3 is only 3 products on it and it's not as dense as N3 though it will change in future but I feel like it should have had more products.

 

[Joe NYC]

Nope Xe3P is 18AP based they are taking their GPUs home next gen

Is that going to be on Panther Lake, Nova Lake or later?

 

[511]

Is that going to be on Panther Lake, Nova Lake or later?

Nova Lake iGPU tile halo as well and Xe3P dGPU(it's alive rn )

 

[oak8292]

My biggest complain with Intel 3 is only 3 products on it and it's not as dense as N3 though it will change in future but I feel like it should have had more products.

With 5 nodes in 4 years the intermediate nodes were going to be small and essentially uneconomic. Ramping to yield is expensive and wanting most of your products on your ‘best’ nodes means ‘betting the farm’ on 18A. Ramping 10 products on 3 nm and then replacing them with 18A products soon after is too expensive. A Xeon on 3 nm is going to have a long life after qualification. A consumer product will disappear as soon as a new node is better.

I think there was a hope that a 4 nm or 20 nm tile would displace Raptor Lake as a lead product.

 

[Doug S]

Don’t rule out the Trump factor. I don’t think it would make a company use a significantly inferior process, but if Intel is close behind I can see it nudging the decision. Three years is a long time to get politically and socially hammered.

Unless he goes full dictator and is still in office in February 2029 he'll be gone before Apple could ship any products with 14A, so Apple could pay lip service to that (and who knows maybe that's the reason for the "Apple is looking at 14A" rumors) but wouldn't have to follow through.

 

[Joe NYC]

Unless he goes full dictator and is still in office in February 2029 he'll be gone before Apple could ship any products with 14A, so Apple could pay lip service to that (and who knows maybe that's the reason for the "Apple is looking at 14A" rumors) but wouldn't have to follow through.

I think what is notable is that TSMC is no longer Morris Cheng company but it is C.C. Wei company. And C.C. Wei likes the US expansion.

And it seems like Trump is satisfied with that, and got along with C.C. Wei just fine.

IMO, Intel will just continue to fade into the background, will not be a topic of conversation, and TSMC will continue to further accelerate US build out.

Just in the last CC (conference call), CC (Wei) mentioned further acceleration in Arizona. And elsewhere, TSMC indicated slow down of investment to Japan and diverting the resources to the US.

 

[Geddagod]

From 3nm class and beyond, currently, there are only two players in town: TSMC and Intel. Intel3 is a solid node for Intel's internal usage. 18a-ap will suffice for the next few years. It's far enough along to conclude that it's better than anything Samsung is likely to ever produce in any sort of economical volume.

The only two wild cards are Rapidus, which I think will be volume limited no matter what for a very long time, and China's various efforts. They WILL eventually be competitive, the question is time.

Samsung 3GAP Exynos chips are out already. I don't think Intel 3 is in any better of a position than 3GAP tbf.
Samsung 2nm is rumored to be used for the Exynos 2600, which we just got ES samples leaking for, for early 2026.
Samsung as a whole is in a very similar position as Intel IMO. Even if Intel's nodes are more competitive in PPA, Samsung has the advantage of actually having been used for external customers in good volumes before.

 

[511]

Even if Intel's nodes are more competitive in PPA, Samsung has the advantage of actually having been used for external customers in good volumes before.

Samsung hasn't had a good yielding node since 8nm

 

[adroc_thurston]

Samsung hasn't had a good yielding node since 8nm

Oh no, their later EUV finfets (SF4 stuff) yield very alright.
But they're unimpressive PPA-wise.

 

[511]

Oh no, their later EUV finfets (SF4 stuff) yield very alright.
But they're unimpressive PPA-wise.

I haven't seen a single chungus die tbh on SF4 shipping in Volume it what's the PPA like though somewhere between N6/N5?

 

[Doug S]

And it seems like Trump is satisfied with that, and got along with C.C. Wei just fine.

So long as no one tells him that little of Apple's stuff will be made in the US, because Arizona only gets a new node when the even newer node enters mass production in Taiwan.