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

 

[DZero]

Now I am optimistic about 2nm from Samsung. 4nm got a good evolution, specially the Exynos 2400e and 2400, and 2nm should do the trick.

Meanwhile the poor 3nm failed hard.

 

[Geddagod]

Now I am optimistic about 2nm from Samsung. 4nm got a good evolution, specially the Exynos 2400e and 2400, and 2nm should do the trick.

Meanwhile the poor 3nm failed hard.

samsung 2nm is just renamed samsung 3nm.

According to Samsung Electronics' third-quarter report on the 17th, the company announced that "the 2nm first-generation gate-all-around (GAA) process has improved performance by 5%, power efficiency by 8%, and area by 5% compared to the 3nm second-generation process."

 

[511]

At the same time they made that announcement they also said they have one product planned on that node internally. So I'm expecting 14A to be very low volume, and all future node development to be canned atp.

Now he changed the tone at CES about 14A looks like his 4D Chess move worked 🤣🤣

 

[regen1]

Considering LBT said that 14A would not go ahead without any big orders... maybe LBT did cook some orders?

There's the other angle of him never saying 14A possibly being cancelled ever after(till now) meeting with US Gov. Probably they forced his hand even if he might have had real intention to exit the Fab side(some think it as some sort of great gamble ?). So who knows?
But 14A on paper should've better stuff for external on paper but let's see. It is still too far away.

 

[dangerman1337]

At the same time they made that announcement they also said they have one product planned on that node internally. So I'm expecting 14A to be very low volume, and all future node development to be canned atp.

But LBT is emphasizing a huge push on 14A, how can it be low volume and if high volume with High NA EUV Machines they'd probably develop future nodes? Because it'd be dumb not to go into the CFET (A7 nodes per IMEC) era which High NA EU enables with huge density increases. Honestly I feel the Nanosheet era of nodes is more orientated towards Smartphones overall with CFET swinging it back towards more High Performance computing.

At the same time they made that announcement they also said they have one product planned on that node internally. So I'm expecting 14A to be very low volume, and all future node development to be canned atp.

One of these SF2 nodes can actually be a proper node density shrink. Similar thing happened with Samsung 3nm, OG samsung 3nm wasn't a "proper" shrink but 3GAP was.
Problem is that I think even a full shrink will make this at best between TSMC N3 and N2.

I mean N3P or better with SF2X while being considerable cheaper is great enough, I'd be bewildered if Nvidia goes "nah" on SF2X on RTX 60 if it is at minimum on par with N3P, like why?

 

[poke01]

Apple-TSMC: The Partnership That Built Modern Semiconductors

Wafer demand model, node economics, and the shifting power dynamics as AI reshapes the foundry landscape

newsletter.semianalysis.com

 

[511]

Apple-TSMC: The Partnership That Built Modern Semiconductors

Wafer demand model, node economics, and the shifting power dynamics as AI reshapes the foundry landscape

newsletter.semianalysis.com

Apple Funding TSMC is a big reason for TSMC success especially after the fiasco with Samsung Intel denying Apple made sense at that point of time based on some posts I have seen from ex Intel employee.

 

[Doug S]

Because it'd be dumb not to go into the CFET (A7 nodes per IMEC) era which High NA EU enables with huge density increases.

CFET is enabled by massively more complex depo/etch cycles, I don't think it will require finer lithography than what we have currently (at least not to make the CFETs themselves)

I wouldn't get too excited about "huge" density increases from CFET. That's sort of the culmination of stacking transistor components vertically to reduce their footprint. We're talking what, one (traditional) Moore's Law cycle of doubling density, or in that ballpark? And who knows what that added complexity is going to do to wafer pricing..

 

[Geddagod]

Now he changed the tone at CES about 14A looks like his 4D Chess move worked

His tone about 14A was always "we are all in" and from the very q2 2025 earnings call they announced their possibility of canning 14A, "laser focused" on the node. He kinda has to say that, otherwise no customers would even want to entertain the thoughts of using the node.

The difference is whether he is "all in" on the R&D of 14A, or the ramp of volume for 14A.

But LBT is emphasizing a huge push on 14A, how can it be low volume and if high volume with High NA EUV Machines they'd probably develop future nodes?

I think you place far too much important on the importance of high na euv machinery.

I mean N3P or better with SF2X while being considerable cheaper is great enough, I'd be bewildered if Nvidia goes "nah" on SF2X on RTX 60 if it is at minimum on par with N3P, like why?

Realistically it just has to be better than N4P, given how AMD seems to not care at all anymore and Intel is pretty much a non factor.
I think minimum on par with N3P is sugar coating it. I think minimum on par with N3E is prob safer. Considering SF2 is very likely going to be worse.
The problem too is if N2X or N2P is an actual shrink, it's going to place serious doubts on the yields of Samsung's leading edge nodes agai.

 

[fastandfurious6]

Intel Foundry hasn't produced any flagship high-end CPU since 2022 Intel 7 (rebrand 10nm).

The last one was Leaky Raptor Lake.
Arrow Lake is TSMC N3.

I think I asked before, but what exactly is the thing that makes newer Intel processes unable to produce high-end chips?

It's a pattern:
Intel 3 meteor lake = low/mid-end mobile
intel 18A phantom lake = low/mid-end mobile
no Intel 3 flagship, no intel 18A flagship

 

[poke01]

Intel Foundry hasn't produced any flagship high-end CPU since 2022 Intel 7 (rebrand 10nm).

The last one was Leaky Raptor Lake.
Arrow Lake is TSMC N3.

I think I asked before, but what exactly is the thing that makes newer Intel processes unable to produce high-end chips?

It's a pattern:
Intel 3 meteor lake = low/mid-end mobile
intel 18A phantom lake = low/mid-end mobile
no Intel 3 flagship, no intel 18A flagship

They can’t do high frequency. Only TSMC can

 

[511]

Intel Foundry hasn't produced any flagship high-end CPU since 2022 Intel 7 (rebrand 10nm).

The last one was Leaky Raptor Lake.
Arrow Lake is TSMC N3.

I think I asked before, but what exactly is the thing that makes newer Intel processes unable to produce high-end chips?

It's a pattern:
Intel 3 meteor lake = low/mid-end mobile
intel 18A phantom lake = low/mid-end mobile
no Intel 3 flagship, no intel 18A flagship

Granite Rapids Clear Water Forest Forest Diamond Rapids are High end server CPU on 18A/Intel 3

 

[Geddagod]

They can’t do high frequency. Only TSMC can

It's a bit insane since I doubt Intel's current node lineup isn't still very, very HPC focused.
Maybe 14A changes that though.

Granite Rapids Clear Water Forest Forest Diamond Rapids are High end server CPU on 18A/Intel 3

Ig they are high end chips, but I think the reasons the nodes were chosen despite not being the best ones available comes to a trade off between "is this product going to be competitive even if we go external".

In that sense I think all those server products fail that test. I think LNL obviously is a decent enough design, and benefits from being external. ARL is harder, but I don't think Intel expected them to faceplant the uncore as hard as they did. NVL follows this rule too.

 

[511]

It's a bit insane since I doubt Intel's current node lineup isn't still very, very HPC focused.

Intel 7 is a unique node that just focuses on xtor performance

Maybe 14A changes that though.

It's going to be perf/watt and density focused nodes if we go buy external focus of node

Ig they are high end chips, but I think the reasons the nodes were chosen despite not being the best ones available comes to a trade off between "is this product going to be competitive even if we go external".

Exactly this Product teams evaluate if we go external for X/Y/Z what are the benefits if there are genuine benefits they choose external else sticks to IFS

In that sense I think all those server products fail that test. I think LNL obviously is a decent enough design, and benefits from being external. ARL is harder, but I don't think Intel expected them to faceplant the uncore as hard as they did. NVL follows this rule too.

Even TSMC can't save mesh/L3 screw-up we have seen that with ARL. NVL get's a different uncore/fabric

 

[fastandfurious6]

Granite Rapids Clear Water Forest Forest Diamond Rapids are High end server CPU on 18A/Intel 3

server = pack as many cores as possible at ~2.5ghz

not 6ghz high-end

they all also still worse than equivalent AMD EPYC series


i.e. EPYC 16core >>>>>> Intel 16core

 

[511]

server = pack as many cores as possible at ~2.5ghz

You said high end .....

not 6ghz high-end

As if AMD has any core that can do 6Ghz lmao

they all also still worse than equivalent AMD EPYC series


i.e. EPYC 16core >>>>>> Intel 16core

Powered by Intel software?

 

[jpiniero]

[News] TSMC Reportedly Plans Mature-Node Tool Shift to Singapore, Accelerates Exit as Arizona Expands

As TSMC steps up its Arizona push with a US$197 million land purchase for a future gigafab, the foundry is also reshaping capacity at home — trimming ...

www.trendforce.com

TSMC possibly moving some of their old nodes to Singapore to make room for new nodes.

 

[fastandfurious6]

Yeah TSMC 2nm and upcoming 2nm+ backside power will be immensely popular for a long while

maxed out production for #1 process in world they'll need all the space they can find

 

[marees]

so it seems that (the much acclaimed) panther lake gpu uses TSMC & not intel

that means the TSMC monopoly on GPU continues — unless you want to settle for old hardware like the switch

neither intel 18a nor samsung 2nm is upto snuff for GPUs — yet.

 

[511]

l

so it seems that (the much acclaimed) panther lake gpu uses TSMC & not intel

Intel clearly labeled external in PTL Slide

that means the TSMC monopoly on GPU continues — unless you want to settle for old hardware like the switch

neither intel 18a nor samsung 2nm is upto snuff for GPUs — yet.

It will be foreseeable future since NVL iGPU got moved to N2 as well

 

[ashFTW]

so it seems that (the much acclaimed) panther lake gpu uses TSMC & not intel

that means the TSMC monopoly on GPU continues — unless you want to settle for old hardware like the switch

neither intel 18a nor samsung 2nm is upto snuff for GPUs — yet.

There are two GPU tiles - 4Xe3, and 12Xe3. The 4Xe3 is made on Intel 3.

 

[DrMrLordX]

so it seems that (the much acclaimed) panther lake gpu uses TSMC & not intel

What was the last iGPU made on Intel silicon?

 

[Geddagod]

Yeah TSMC 2nm and upcoming 2nm+ backside power will be immensely popular for a long while

Honestly A16 doesn't seem popular at all. Only Nvidia is rumored to be using it like 2 years after HVM readiness.

 

[jdubs03]

Honestly A16 doesn't seem popular at all. Only Nvidia is rumored to be using it like 2 years after HVM readiness.

Probably a bit early to know. I think by the end of this year we will have a better idea of its popularity.

 

[adroc_thurston]

I think by the end of this year we will have a better idea of its popularity.

It's basically an orphan node.
Everyone is skipping straight to A14 in H2'28.