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.

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

 

[DrMrLordX]

Granite Rapids is out next month for DC and it should beat Genoa more often than not, while being perfectly competitive with Turin if SRF is anything to go by.

Granite Rapids will not be competitive with Turin. At least not in workloads that concern potential DCG clients. Also it's dependent on an advanced node that isn't producing in the same volume as Intel 7. Intel will not be able to produce as much Granite Rapids as they can Sapphire or Emerald Rapids.

(And SRF is VERY good at what it's supposed to do)

Ehhhhh agree to disagree, this is a foundry thread so I'll leave it at that. Just take a look at the availability of Sierra Forest and that should tell you how "good" it is and how capable Intel is of supplying the market (to the extent that the market wants it at all).

, Clearwater Forest is gonna be even better next year and Diamond Rapids will beat Turin and at least be competitive against Venice.
It's been a bumpy ride to say the least but Intel can see the light at the end of the tunnel as far as the server (And workstation/HEDT) market is concerned.

Yeah sure. Intel is gonna have to prove that with hard numbers. Anything else is just hype and noise at this point. Their foundry business is not positioned to replace the enormous volume of Skylake-SP/Cascade Lake-SP that they pushed onto the market years ago, and they've even struggled to reach those productions volumes with 10nm/Intel 7 (IceLake-SP, Sapphire Rapids-SP, Emerald Rapids). Now that Intel 7 datacentre product volume is maybe stabilizing, the market has already moved onwards. Nobody in their right mind expects Intel to ramp up volume on 20a/18a products AND to leapfrog Turin or Turin Dense. If it's possible, Intel has to prove it.

 

[igor_kavinski]

Intel's life (and AMD's too possibly) would be made extremely difficult whenever Apple/Qualcomm and some of the bigger RISC-V players enter the server market. The data center power and cooling costs differential if sufficiently large, is going to result in huge contracts being bestowed on the ARM players and AMD/Intel will be left twiddling their thumbs. I think it's @dullard who is always saying that AMD/Intel need to join forces against ARM rather than fighting each other.

 

[poke01]

I think it's @dullard who is always saying that AMD/Intel need to join forces against ARM rather than fighting each other.

The problem is the very closed nature of x86. If x86 vendors licensed x86 10-15 years ago just like ARM did this wouldn't have been a problem. AMD/Intel is too small against Nvidia, Apple, Amazon, Google and Microsoft. The biggest mistake that Intel made was to not make x86 designs for the iPhone. If Intel had agreed to Jobs then x86 would have been irreplaceable.

 

[DavidC1]

The problem is the very closed nature of x86. If x86 vendors licensed x86 10-15 years ago just like ARM did this wouldn't have been a problem. AMD/Intel is too small against Nvidia, Apple, Amazon, Google and Microsoft. The biggest mistake that Intel made was to not make x86 designs for the iPhone. If Intel had agreed to Jobs then x86 would have been irreplaceable.

Very slowly, the thing that Intel did to RISC vendors is playing out again. Volume is king. Where volume is, money and people are. So all the top people are more likely to gravitate towards the ARM ecosystem, so they have more chances of success.

Even if Intel did agree to make x86 iPhone, with the fundamental mindset of the company protecting their cash cow, they'd have got kicked out eventually.

 

[jdubs03]

Read this just now. Seems like China is getting closer. I think they’re going to hit a wall though if they can’t get the latest ASML lithography machines. I don’t see how how they can get down to lower sizes without it becoming prohibitively expensive.

China's chip capabilities just 3 years behind TSMC, teardown shows

Analysis reveals limited impact of U.S. tech curbs on consumer products

asia.nikkei.com

What y’all think?

 

[maddie]

Read this just now. Seems like China is getting closer. I think they’re going to hit a wall though if they can’t get the latest ASML lithography machines. I don’t see how how they can get down to lower sizes without it becoming prohibitively expensive.

China's chip capabilities just 3 years behind TSMC, teardown shows

Analysis reveals limited impact of U.S. tech curbs on consumer products

asia.nikkei.com

What y’all think?

Not knowing any details, I'm fairly certain they have a plan to further advance their indigenous litho, unless we think they're just going to stop all development. We have several threads with Intel's woes. One common feature is their hubris made them feel invincible.

 

[jdubs03]

Not knowing any details, I'm fairly certain they have a plan to further advance their indigenous litho, unless we think they're just going to stop all development. We have several threads with Intel's woes. One common feature is their hubris made them feel invincible.

Afaik ASML pretty much has a monopoly on state of the art lithography machinery. I find it very hard to believe that even a highly subsidized entity could miraculously make up for years and years worth of R&D that ASML has undertaken to get where they are today. They already produced the machinery to get down to below 2nm. Right now SMIC is at 7nm. They won’t have access to EUV and that is required for 3nm. Seems like there’s a big obstacle in their way to bridge that three year difference.

 

[maddie]

Afaik ASML pretty much has a monopoly on state of the art lithography machinery. I find it very hard to believe that even a highly subsidized entity could miraculously make up for years and years worth of R&D that ASML has undertaken to get where they are today. They already produced the machinery to get down to below 2nm. Right now SMIC is at 7nm. They won’t have access to EUV and that is required for 3nm. Seems like there’s a big obstacle in their way to bridge that three year difference.

If we can agree, that for China, this will justify a "Manhattan Project" level of R&D and that the ASML way is one, not the only way, of generating, collecting and focusing EUV wavelengths used in wafer processing, then I, for one, would not be complacent. Just my belief.

 

[jdubs03]

If we can agree, that for China, this will justify a "Manhattan Project" level of R&D and that the ASML way is one, not the only way, of generating, collecting and focusing EUV wavelengths used in wafer processing, then I, for one, would not be complacent. Just my belief.

Even if that is the case they have to start from pretty much scratch. They were using ASML machinery to get to where they are now. That’s been taken away/not allowed the EUV machinery to get to 3nm (sorry don’t really want to repeat that but it’s still important). In order to even maintain the 3 year deficit they’d have to have their own homegrown lithography machinery at the level of their current situation. And that won’t be the case they have to build up the infrastructure and that’s gonna take years.

Also per your analogy, the Manhattan Project took four years to come to fruition.

The only way that they could quickly get some volume at sub 5nm nodes with their current available technology is lots of multi patterning, and that’s very expensive.

 

[maddie]

Even if that is the case they have to start from pretty much scratch. They were using ASML machinery to get to where they are now. That’s been taken away/not allowed the EUV machinery to get to 3nm (sorry don’t really want to repeat that but it’s still important). In order to even maintain the 3 year deficit they’d have to have their own homegrown lithography machinery at the level of their current situation. And that won’t be the case they have to build up the infrastructure and that’s gonna take years.

Also per your analogy, the Manhattan Project took four years to come to fruition.

The only way that they could quickly get some volume at sub 5nm nodes with their current available technology is lots of multi patterning, and that’s very expensive.

You really think they're now starting? Sanctions has been a thing for years and obviously did and will increase going forward. If they succeed, and it's not a sure thing, then the foundry world will be shaken.

 

[jdubs03]

You really think they're now starting? Sanctions has been a thing for years and obviously did and will increase going forward. If they succeed, and it's not a sure thing, then the foundry world will be shaken.

Tbh I don’t know. If they were forward thinking they’d have started already but my question to you is then: do you really think they’re at the current level of the ASML solutions that which they’re current using?

 

[LightningZ71]

No, they can't replicate what ASML makes at the leading edge of the market... TODAY. They got some of the best, latest generation DUV machines that were produced. They are willing to accept the throughput and financial hut of heavy multipatterning to produce "5nm class" products.

They most certainly have 3nm class EUV machines in research labs as it's something that they have been working towards for years. Having absolutely zero concern for existing IP, an an extensive track record of obtaining technical information through questionable means, they undoubtedly have all the research information that they need to make EUV machines. What they lack is likely the tooling and production base needed to produce the machibes with the crazy amounts of precision that are required.

THAT is where they are going to have to take their time. It WILL happen. GAA and BPSD are things that are known in the industry, and they will likely employ them quickly.

I expect "3nm class" limited production by the end of 2025, volume in 2H26, and "2nm class" products on shelves by 2H27. Even with sanctions in place, the size of their domestic market will allow them to scale enoigh to make it economical quickly.

 

[DavidC1]

The hubris currently is that ASML's way is the only way to go and they can never be breached. It will be as history shows. When majority says "IMPOSSIBLE" it happens.

Also Moore's Law favors lower cost, and lower power. That means as time passes, the gap between the peak and the bottom gets lower and lower. It may not matter that much in the end.

N3 has near zero SRAM scaling and N2 has almost no logic scaling. Intel's 18A has mere 30% density improvement, and 14A, which is 2 years away from 14A has mere 15% perf/watt and 20% density improvement.

What will future hold? 5A is 10% perf/watt and 10% density over 7A?

 

[Doug S]

Even if Intel did agree to make x86 iPhone, with the fundamental mindset of the company protecting their cash cow, they'd have got kicked out eventually.

Intel was totally down with making x86 SoCs for the iPhone. It was Jobs who told them he only wanted ARM SoCs and would not accept x86, and Intel's "NIH" caused them to pass on the opportunity of the century.

 

[DavidC1]

Intel was totally down with making x86 SoCs for the iPhone. It was Jobs who told them he only wanted ARM SoCs and would not accept x86, and Intel's "NIH" caused them to pass on the opportunity of the century.

Oh really?

Intel could've powered the original iPhone, but decided against it, says ex-CEO Otellini

Intel's Paul Otellini has admitted, on his last day as Intel's CEO, that the original iPhone could've been powered by an Intel CPU -- but Intel decided against it.

www.extremetech.com

We ended up not winning it or passing on it, depending on how you want to view it. And the world would have been a lot different if we'd done it," Otellini told me in a two-hour conversation during his last month at Intel. "The thing you have to remember is that this was before the iPhone was introduced and no one knew what the iPhone would do... At the end of the day, there was a chip that they were interested in that they wanted to pay a certain price for and not a nickel more and that price was below our forecasted cost. I couldn't see it. It wasn't one of these things you can make up on volume. And in hindsight, the forecasted cost was wrong and the volume was 100x what anyone thought."

Had nothing to do with NIH. They lost because they believed iPhone would have barely sold and wasn't worth sacrificing their cash cow. Otellini did not have the vision for it. This is how the mindset of finance works. Pure numbers.

 

[jdubs03]

Oh really?

Intel could've powered the original iPhone, but decided against it, says ex-CEO Otellini

Intel's Paul Otellini has admitted, on his last day as Intel's CEO, that the original iPhone could've been powered by an Intel CPU -- but Intel decided against it.

www.extremetech.com

Had nothing to do with NIH. They lost because they believed iPhone would have barely sold and wasn't worth sacrificing their cash cow. Otellini did not have the vision for it. This is how the mindset of finance works. Pure numbers.

They were probably too ingrained with the Wintel ideology (for lack of a better word) back then.

 

[DavidC1]

They were probably too ingrained with the Wintel ideology (for lack of a better word) back then.

Just like they said with Microsoft, they did EVERYTHING to sell Windows. Everything else, the browser, Office, was just something to prop up Windows.

Intel has the same mentality. Problem is that there's internal war going on with the company, and it probably was brought on by Andy Grove's leadership. What works under one leadership may not work for most though.

The real problem is again the main cash cow. The war mentality means they would also supress other divisions that could be better and help the company.

I couldn't see it. It wasn't one of these things you can make up on volume. And in hindsight, the forecasted cost was wrong and the volume was 100x what anyone thought."

Repeat that elsewhere.

 

[lucasworais]

Afaik ASML pretty much has a monopoly on state of the art lithography machinery. I find it very hard to believe that even a highly subsidized entity could miraculously make up for years and years worth of R&D that ASML has undertaken to get where they are today. They already produced the machinery to get down to below 2nm. Right now SMIC is at 7nm. They won’t have access to EUV and that is required for 3nm. Seems like there’s a big obstacle in their way to bridge that three year difference.

ASML has a monopoly only on EUV machines. China is probably going to use nanoimprint (NIL) technology to achieve 5nm and beyond.

 

[Doug S]

Oh really?

Intel could've powered the original iPhone, but decided against it, says ex-CEO Otellini

Intel's Paul Otellini has admitted, on his last day as Intel's CEO, that the original iPhone could've been powered by an Intel CPU -- but Intel decided against it.

www.extremetech.com

Had nothing to do with NIH. They lost because they believed iPhone would have barely sold and wasn't worth sacrificing their cash cow. Otellini did not have the vision for it. This is how the mindset of finance works. Pure numbers.

That article says nothing about an "x86 CPU", it says an "Intel CPU". Since Apple wasn't designing their own stuff from day one, it would have been an "Intel CPU" with ARM cores in the iPhone, rather than the Samsung CPU with ARM cores that Apple actually shipped in the first few generations of iPhone.

The "chip Apple was interested in" referred to in the article was StrongARM/XScale, then the best performing ARM design available, which Intel owned having acquired it from DEC in the previous decade. They ended up selling it to Marvell in 2006, after they'd made the decision they didn't want to be in the business of selling SoCs to Apple for the iPhone - that's probably around the time they started tossing around the idea of Atom and trying to do "x86 everywhere".

According to Jobs when he first approached Intel inquiring about StrongARM/XScale, they tried to really sell him on x86, but he when he stopped laughing he told them he was only interested in ARM.

 

[Hitman928]

New article saying that Broadcom is not happy with the results from their test chips on 18a.

https://www.reuters.com/technology/intel-manufacturing-business-suffers-setback-broadcom-tests-disappoint-sources-2024-09-04/

 

[igor_kavinski]

New article saying that Broadcom is not happy with the results from their test chips on 18a.

https://www.reuters.com/technology/intel-manufacturing-business-suffers-setback-broadcom-tests-disappoint-sources-2024-09-04/

Honestly not surprised. They got into the habit of getting away with stupendous claims since the 14nm era and this shows nothing has changed in that regard.

 

[OriAr]

New article saying that Broadcom is not happy with the results from their test chips on 18a.

https://www.reuters.com/technology/intel-manufacturing-business-suffers-setback-broadcom-tests-disappoint-sources-2024-09-04/

The process isn't supposed to move into HVM until Q2 2025, so yields still being below what's acceptable for HVM isn't exactly the gotcha news people think it is.
Question is what yields will be come June 2025.

 

[Gideon]

New article saying that Broadcom is not happy with the results from their test chips on 18a.

https://www.reuters.com/technology/intel-manufacturing-business-suffers-setback-broadcom-tests-disappoint-sources-2024-09-04/

The main issue seems to be yields (AFAIK) not the performance of the node?

The process isn't supposed to move into HVM until Q2 2025, so yields still being below what's acceptable for HVM isn't exactly the gotcha news people think it is.
Question is what yields will be come June 2025.

Yep!

 

[jdubs03]

Seems like peeps are really trying to lay it on Intel right now. It’s the cool thing to do.

 

[Hitman928]

The process isn't supposed to move into HVM until Q2 2025, so yields still being below what's acceptable for HVM isn't exactly the gotcha news people think it is.
Question is what yields will be come June 2025.

They are supposed to be "manufacturing ready" by the end of this year but only open up to external customer production next year.

This news actually correlates with what I had heard from a friend of a friend, that the company they worked for wasn't happy with 18a's progress (PDK readiness and yield). Now, after Pat's latest comments on it, I figured I had outdated information, but this article makes me think the info I got may not be all that old. Obviously, the customers know that Intel is telling them they aren't going into production for their wafers until next year, but these customers also work with other manufacturers and have certain expectations of production readiness this close to tape-out based on that.

I will say that I think it could be a misunderstanding in that Intel's communication on fab readiness and tape-in/out timelines has always been weird compared to the rest of the industry. So, when Intel says manufacturing ready and that really means risk production only, that may be causing confusion. All of this is just speculation on my part though.