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

DisEnchantment · Apr 24, 2020

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.

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

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

moinmoin · Feb 11, 2025

Gideon said:
More info on TSMC 2nm:

https://www.reddit.com/r/hardware/s/lW3rXU4NOq

Direct link to the article at SemiWiki:

IEDM 2025 – TSMC 2nm Process Disclosure – How Does it Measure Up? - Semiwiki

Initial thoughts At IEDM held in December 2024, TSMC presented: “2nm Platform Technology featuring Energy-efficient Nanosheet Transistors and Interconnects co-optimized with 3DIC for AI, HPC and Mobile SoC Applications,” the authors are: Geoffrey Yeap, S.S. Lin, H.L. Shang, H.C. Lin, Y.C. Peng...

semiwiki.com

Gideon · Feb 11, 2025

moinmoin said:
Direct link to the article at SemiWiki:

IEDM 2025 – TSMC 2nm Process Disclosure – How Does it Measure Up? - Semiwiki

Initial thoughts At IEDM held in December 2024, TSMC presented: “2nm Platform Technology featuring Energy-efficient Nanosheet Transistors and Interconnects co-optimized with 3DIC for AI, HPC and Mobile SoC Applications,” the authors are: Geoffrey Yeap, S.S. Lin, H.L. Shang, H.C. Lin, Y.C. Peng...

semiwiki.com

They have more data behind paywall (apparently with more details) but from the public bersion TL;DR basically is this:

Note: the 3nm paper reference suggests this is in comparison to N3E, not N3.

Power:

At 10nm TSMC provided a larger power reduction than Samsung and maintained that lead until 3nm where Samsung Gate All Around (GAA) provided a large enough improvement to mostly close the gap to TSMC’s 3nm FinFET process in power (GAA versus FinFET is expected to provide a greater power improvement). TSMC 2nm announced power improvement of 30% versus 3nm is greater than Samsung’s 25% improvement and TSMC once again maintains a lead.

...
Just the N3 to N2 bars on the graph show a 55% improvements versus the announced 30% improvement.

Performance:
Based on this analysis it is our belief that Intel 18A has the highest performance for a 2nm class process with TSMC in second place and Samsung in third place.

Area:

For high density logic cells TSMC is well ahead of Samsung and Intel on density, Intel is second, and Samsung is third.

As previously mentioned, the TSMC paper does not include SRAM cell sizes, however there is a graph of SRAM density versus node, see figure 3.

The rest we know. no Backside Power Delivery, and lots of rumors about yields and price

And here's another comment on reddit (why I linked to it in the first place):

Just to add, Scott has HD cell at 313 mm2 for N2. 238 mm2 for 18A, and 231 mm2 for SF2
No power figures for 18A, but N2 was 0.14, SF2 was 0.17
Performance he had 18A at 2.53, 2.27 for N2, 2.19 for SF2. He believes 18A will lead vs A16 and SF1.4 for performance
(No reason listed as how it's achieved)

Intel's 18A looks quite nice for HPC applications, but I find it doubtful it can compete in <25W mobile parts if the 30-55% power reduction from N3E (not N3 !!) is correct, particularily as the density also seems to be better

511 · Feb 12, 2025

cannedlake240 said:
Prediction for IFS node perf/w:
18A is N4P +-5%, N3 series advantage will be most pronounced at low voltage
18AP is then somewhere in N3B to N3P range
14A(official number +15% over 18A) ≈ N3P to N2 range

Won't rule out even lower numbers but It seems unlikely. Would be curious If 14A somehow regained some of the lost ground on 18A regression

Nice prediction 🙂 I would take TechInsights Prediction than this one though as for low power characteristics even Intel 4(RWC) matches N3B (LNC)

cannedlake240 · Feb 12, 2025

511 said:
Nice prediction 🙂 I would take TechInsights Prediction than this one though as for low power characteristics even Intel 4(RWC) matches N3B (LNC)
View attachment 116851

I'm aware of this and it does raise some questions. If this is really the case how is IFS not getting fab contracts worth 10s of billions already? PDK quality issues? Is there some industry wide conspiracy against Intel? Or maybe ParametricYield bros are right? Maybe 18A not yielding chips that meet PnP standards(like Icelake)

511 · Feb 12, 2025

cannedlake240 said:
I'm aware of this and it does raise some questions. If this is really the case how is IFS not getting fab contracts worth 10s of billions already? PDK quality issues? Is there some industry wide conspiracy against Intel? Or maybe ParametricYield bros are right? Maybe 18A not yielding chips that meet PnP standards(like Icelake)

If it had parametric yield issues we won't be having GNR/SRF/ARL-U/MTL on Intel 3/4 so no.

The only possible issue is Trust and risk the only big customers that doesn't have Trust issues going with Intel is Hyperscaler like Googl/MSFT/Amazon. Customes like AMD/Nvidia/Qualcomm just fear for design leaking to Intel Products

As for 18A PnP issues it's mostly FUD some can be genuine also Morris Chang is saying Intel to stop it's foundry efforts that only means one thing he is afraid of the progress.

cannedlake240 · Feb 12, 2025

511 said:
If it had parametric yield issues we won't be having GNR/SRF/ARL-U/MTL on Intel 3/4 so no.

The only possible issue is Trust and risk the only big customers that doesn't have Trust issues going with Intel is Hyperscaler like Googl/MSFT/Amazon. Customes like AMD/Nvidia/Qualcomm just fear for design leaking to Intel Products

As for 18A PnP issues it's mostly FUD some can be genuine also Morris Chang is saying Intel to stop it's foundry efforts that only means one thing he is afraid of the progress.

But i4/3 and i18A are completely different nodes... Just because the former is ok doesn't mean 18A has to be as well. 14nm had major issues, was delayed 1.5 years even though 22nm went relatively ok(to outsiders), and theyre both finfet nodes.

511 · Feb 12, 2025

cannedlake240 said:
But i4/3 and i18A are completely different nodes... Just because the former is ok doesn't mean 18A has to be as well. 14nm had major issues, was delayed 1.5 years even though 22nm went relatively ok(to outsiders), and theyre both finfet nodes.

22nm didn't had issues iirc but yes after 22nm they have been downhill 14nm was delayed and 10nm was delayed for like 4 or maybe 5 years.

7nm got delayed and when pat came only than it was fixed and it was not the original 7nm which was scrapped we got 7nm+(Intel 4) and 7nm++ (Intel 3) but if 18A had issues we should have known by now.

As for 18A Ian cutress Tech insights Scotten Johnes has info regarding Yield and node data

moinmoin · Feb 12, 2025

511 said:
Morris Chang is saying Intel to stop it's foundry efforts that only means one thing he is afraid of the progress.

Who cares what an old retired man is saying? He has been no CEO of TSMC for 20 years and no chairman for 7 years now. It's like quoting Pat now.

Hesperax · Feb 12, 2025

Article from Digitimes on TSMC and Intel:

TSMC US board secret talks: Trump floats three options to boost Intel

TSMC's first board of directors meeting in the US on February 12 did not reveal any new fab expansions or high-level personnel adjustments, as previously speculated. Nevertheless, the market is closely monitoring the ongoing confidential conversations between the US government and TSMC.

www.digitimes.com

Summary of the Article since it has a Paywall:

https://twitter.com/x/status/1889857817722626518

511 · Feb 13, 2025

Hesperax said:
Article from Digitimes on TSMC and Intel:

TSMC US board secret talks: Trump floats three options to boost Intel

TSMC's first board of directors meeting in the US on February 12 did not reveal any new fab expansions or high-level personnel adjustments, as previously speculated. Nevertheless, the market is closely monitoring the ongoing confidential conversations between the US government and TSMC.

www.digitimes.com

Summary of the Article since it has a Paywall:

https://twitter.com/x/status/1889857817722626518

LMFAO AI generated?

DavidC1 · Feb 13, 2025

511 said:
but if 18A had issues we should have known by now.

Going from +15% and +10% to just +15% could be called an issue.

511 said:
The only possible issue is Trust and risk

That's possible, but the other part is a) Arrowlake has uarch changes so it's not an ISO-comparison and b) Some of David Huang's and Geekerwan's results are questionable.

Ian Cuttress, David Huang, Chips & Cheese, and Geekerwan tests are a grade lower in quality than what Anandtech of old used to put out. The move to the simplified Smartphone era didn't just simply UI, but websites and even information as well, often to the reduction of detail.

Videos for example is like moving from L3 cache to system memory. Yea it's fast enough, but it's much much slower. Videos are NOT for quickly looking up information. Website is.

A picture may be worth a thousand words, but a video is maybe worth 50 words.

cannedlake240 said:
I'm aware of this and it does raise some questions. If this is really the case how is IFS not getting fab contracts worth 10s of billions already?

10-15% differences for a new foundry is nowhere near enough to be worth switching a supplier, tools, and relations, especially one that struggled immensely and treated customers like second class.

Think about why YOU buy stuff. For such small "performance" differences everything else matters.

OneEng2 said:
Not saying you are wrong here, but where did you get the 55mm2? It does wash on casual inspection though since current N3B Skymont is ~ 1.73mm2 with cache. Darkmont is likely to have more complexity and more cache.

1.73 x 24 cores on a similar density is about 42mm2. If you add required IO then 55mm2 makes perfect sense. I don't know why you think it would be lot bigger. And yes Darkmont has very minor changes. It probably won't increase it's size by more than 5%. All the L3 cache is all on the Base Foveros tiles too.

maddie · Feb 13, 2025

DavidC1 said:
Videos for example is like moving from L3 cache to system memory. Yea it's fast enough, but it's much much slower. Videos are NOT for quickly looking up information. Website is.

A picture may be worth a thousand words, but a video is maybe worth 50 words.

Spot on.

Youtube and similar might have destroyed decades of productivity improvement.

511 · Feb 13, 2025

maddie said:
Spot on.

Youtube and similar might have destroyed decades of productivity improvement.

Clicks are more important than Quality of content xDd

511 · Feb 13, 2025

DavidC1 said:
Going from +15% and +10% to just +15% could be called an issue.

Could be could be not we will have 18A presentation in like 5-6 days at ISSCC.

DavidC1 said:
That's possible, but the other part is a) Arrowlake has uarch changes so it's not an ISO-comparison and b) Some of David Huang's and Geekerwan's results are questionable.

Maybe but they are the only 3rd party data points that are reliable as for ISO-Comparison I agree but that's about as close we can get we won't have PDKs lying around to simulate the difference.We will have Intel 3 vs 4 data points soon with the same uArch cause I had a chat with David Huang and he said he is also interested in ARL-U RWC vs MTL-U RWC comparison so he will be buying the device and test it soon.

DavidC1 said:
Ian Cuttress, David Huang, Chips & Cheese, and Geekerwan tests are a grade lower in quality than what Anandtech of old used to put out. The move to the simplified Smartphone era didn't just simply UI, but websites and even information as well, often to the reduction of detail.

Videos for example is like moving from L3 cache to system memory. Yea it's fast enough, but it's much much slower. Videos are NOT for quickly looking up information. Website is.

A picture may be worth a thousand words, but a video is maybe worth 50 words.

Yes I agree with you here

DavidC1 said:
10-15% differences for a new foundry is nowhere near enough to be worth switching a supplier, tools, and relations, especially one that struggled immensely and treated customers like second class.

Sure but switching to a different foundry is a good way for negotiation. On Intel's struggle j would like to applaud Intel's Management they had the combined Moat of TSMC + Nvidia and they managed to shoot themselves in the foot that is something else

DavidC1 said:
Think about why YOU buy stuff. For such small "performance" differences everything else matters.

1.73 x 24 cores on a similar density is about 42mm2. If you add required IO then 55mm2 makes perfect sense. I don't know why you think it would be lot bigger. And yes Darkmont has very minor changes. It probably won't increase it's size by more than 5%. All the L3 cache is all on the Base Foveros tiles too.

True too bad they delayed Clearwater Forest DCAI is such a mess that is hilarious.

digitaldreamer · Feb 13, 2025

Quote from Digitimes on TSMC and Intel:
TSMC along with several major companies would invest in Intel’s standalone foundry business

LOL

DavidC1 · Feb 13, 2025

511 said:
We will have Intel 3 vs 4 data points soon with the same uArch cause I had a chat with David Huang and he said he is also interested in ARL-U RWC vs MTL-U RWC comparison so he will be buying the device and test it soon.

Intel 3 is a nice improvement over Intel 4. You can tell just by the spec sheets. The greatly increased base frequency is usually indicative of improved power profile. One example of that was from Arrandale to Sandy Bridge transition. The -U parts and the subportables were thought as almost Celerons because of how behind they were. Arrandale with Turbo closed some of the gap but it was Sandy Bridge that really made the idea of Ultrabooks viable.

Of course in this context that doesn't tell us anything since we're trying to see how it compares with N3.

511 said:
Sure but switching to a different foundry is a good way for negotiation. On Intel's struggle j would like to applaud Intel's Management they had the combined Moat of TSMC + Nvidia and they managed to shoot themselves in the foot that is something else

You can also abandon them shortly after the desired company does well again. The joke about Nvidia could be applied here as well.

"People want AMD to be competitive so they can buy cheaper Nviida cards"

to

"People want Intel to be competitive so they can buy better TSMC nodes"

lightisgood · Feb 17, 2025

This is simply good news...
The i18A customers are not only x86, but also GPU (probably include AI).

Intel's Celestial dGPUs Are Expected To Feature The Xe3P Architecture, Will Likely Use Intel Foundry Instead of TSMC

Intel's next-gen Celestial discrete GPUs are expected to feature the Xe3P architecture and could potentially be developed by the IFS.

wccftech.com

Geddagod · Feb 17, 2025

511 said:
Nice prediction 🙂 I would take TechInsights Prediction than this one though as for low power characteristics even Intel 4(RWC) matches N3B (LNC)
View attachment 116851

Tock cores decrease or keep perf/watt at the low end of the curve usually. Knowing Intel, perf/watt prob decreased with LNC even iso node at the low end of the curve lol. Plus all that core private cache has to be pretty expensive to power on and take up a disproportionate percent of total core power at that end of the curve...
Intel themselves claims that Intel 3 is ~ TSMC N3E?P? in perf/watt.

I dislike Scotten Jones' method of calculating perf/watt. He assumes that 10 ESF is ~N7 perf/watt, inflating the perf/watt calculations of Intel's future nodes.

511 said:
I had a chat with David Huang and he said he is also interested in ARL-U RWC vs MTL-U RWC comparison so he will be buying the device and test it soon.

I thought he said it would be a couple of months

511 · Feb 18, 2025

Geddagod said:
Tock cores decrease or keep perf/watt at the low end of the curve usually. Knowing Intel, perf/watt prob decreased with LNC even iso node at the low end of the curve lol. Plus all that core private cache has to be pretty expensive to power on and take up a disproportionate percent of total core power at that end of the curve...
Intel themselves claims that Intel 3 is ~ TSMC N3E?P? in perf/watt.
View attachment 117369
I dislike Scotten Jones' method of calculating perf/watt. He assumes that 10 ESF is ~N7 perf/watt, inflating the perf/watt calculations of Intel's future nodes.

Same someone asked Scot he said he doesn't have a good reference for Performance/Watt in the semiwiki comment section

Geddagod said:
I thought he said it would be a couple of months

Yes

adamge · Feb 18, 2025

lightisgood said:
This is simply good news...
The i18A customers are not only x86, but also GPU (probably include AI).

Intel's Celestial dGPUs Are Expected To Feature The Xe3P Architecture, Will Likely Use Intel Foundry Instead of TSMC

Intel's next-gen Celestial discrete GPUs are expected to feature the Xe3P architecture and could potentially be developed by the IFS.

wccftech.com

Celestial is a myth. The guy on youtube said ARC is cancelled. How could a cancelled business team produce new technology so many years later.

Edit: Wow, just wow at the comments section of that link.

511 · Feb 19, 2025

adamge said:
Celestial is a myth. The guy on youtube said ARC is cancelled. How could a cancelled business team produce new technology so many years later.

Raichu and couple of reliable leakers have said celestial is alive the only guy that said arc is dead is the clown MLID.

adamge said:
Edit: Wow, just wow at the comments section of that link.

Yeah the wccftech section is horrible

511 · Feb 19, 2025

https://twitter.com/x/status/1892246045385515266

Look at this despite different sram size the Density is same apparently there was no scaling for N2

Saylick · Feb 19, 2025

Ian is xeeting a bunch of slides. Going to be interesting looking through them all at the end of the day.

lightisgood · Feb 19, 2025

I think i18A is superb process.
"Make Intel Great Again", Intel is really doing it.

Jim Keller says a 'great Intel' is worth $1 trillion, company would be sold at fire sale pricing if sold now

It is better to take the company private and retain its IDM nature.

www.tomshardware.com

511 · Feb 19, 2025

Oh Boy here is the SRAM Shmoo Plot as well for both 18A and N2

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

Golden Member

Diamond Member

Golden Member

Golden Member

Senior member

Golden Member

Senior member

Golden Member

Diamond Member

Member

Golden Member

Golden Member

Diamond Member

Golden Member

Golden Member

Member

Golden Member

Senior member

Golden Member

Golden Member

Member

Golden Member

Golden Member

Diamond Member

Senior member

Golden Member