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

 

[511]

As I speculated, with zero customer and zero commitment, Intel 18A is a complete flop process. And if people still not able to get what CFO said about 100% failure rate, well now that Intel might totally scrapped the plan to offer 18A to external customer. Are you getting it yet?

Yeah OFC Commitment from MSFT/AMZN doesn't exist 🤣

If I am BOD, I will fire Pat also due to his BS execution. He was trying to hide that facts IFS cannot provide good yield by going chiplet solution. Yes, chiplet is the future solution only if the monolithics design cannot sustain PPA any further. As I compared upcoming PTL-H and AMD's Gorgon Point: AMD able to offer 24 threads and 16 CU, meanwhile Intel PTL is having 12+4 threads and 12XE (Thanks TSMC!). Do you think PTL has a chance to perform faster than Gorgon Point? Now that Gorgon Point's NPU rumored to hit 65TOPS; 15 TOPS faster than PTL. And what process node AMD is using???

Well yield is not a problem for 18A as of rn as for NPU nobody cares all I care about is CPU+GPU and I am glad that Intel didn't waste die space for NPU like they did on LNL it's fine if the NPU is 50 TOPS cause the GPU is 120 TOPS.

And if you think AMD is sitting idle and let Intel catch up a bit, hoho let's wait for AMD's Soundwave...but that is topic for another thread.

NO One is sitting Idle nowadays that's the fun

 

[Tigerick]

Ok, this explains why Clearwater Forest was delayed. It may not have been just packaging.

Another thing is this is in line with Exist50's very gloomy projections. He said they won't have leadership even with 14A.

What Exist50's claim about 14A with High-NA is bold and unfounded. And yet, I strongly believe about Intel's failure to execute and perform is a pattern. It is due to working culture and habit, in case people are wondering why TSMC able to execute almost flawlessly every new node: check out TSMC's Project NightHawk.

With PTL's competing with AMD Gorgon Point (N4P), yep 18A comparable to N3P (from CC Wei) is way over-estimate. Thus, 14A is only catching up to TSMC's N2 process. By that time, TSMC will move on to their true A14 process. IFS will forever be catching up to TSMC without customers commitment: the same situation as SF. That's why SF and IFS are rumored to focus on fine-tuninng the current process, namely 18A and SF2.

It's actually good move of LBT to dump external foundry business. By doing that, Intel could save billions dollar especially with no customers committed. The delay of Microsoft's ARM CPU might be the final nail to the coffin...

Stephen Robinson, the Chief architect of the E cores since at least Tremont is taking over the next gen core project.

I'm expecting the dead end P cores that started since Sandy Bridge to be abandoned and go with the E core route, which is: no uop cache, clustered decode, no SMT, no thoughtless expansions.

Intel did, and would wish to.

They are unappealing to work, that's why they leave Intel. The solution is to "fix" the mess they are having. Poaching at this point just wastes money.

Do you not remember a slide from Intel saying Intel 3 and 4 is the same thing?

I said when they downgraded 18A performance from 15% + 10% to 15% that it's basically rebranded 20A. If you claim X offers 10% extra performance, but suddenly you take that away, what else is there?

 

[511]

I will be having the last laugh when 90% of this turns out to be false 😂 except for dumping 18A for external customers it can or can't be true.

 

[Io Magnesso]

While that's true, it's also important to realize that many foundries are . . . foundering.

Remember when Samsung saved GF by licensing 14LPE and (I guess) LPP? GF wound up making processes that were arguably better than what Samsung licensed to them initially. What if Intel, Samsung, GF, and possibly others had a cross-licensing agreement and formed a multinational fab consortium? Everybody gets to look at everybody else's patents and trade secrets, and then they make the best nodes they can with what's available.

Survive together or die alone. Take your pick!

IBM has provided process technology to various fabs We also provided technology to the UMC.
The latest project at the moment is 2nm of Rapidas
Originally, it seems to improve the 2nm process node for research at IBM to make it suitable for mass production.

 

[Io Magnesso]

What Exist50's claim about 14A with High-NA is bold and unfounded. And yet, I strongly believe about Intel's failure to execute and perform is a pattern. It is due to working culture and habit, in case people are wondering why TSMC able to execute almost flawlessly every new node: check out TSMC's Project NightHawk.

With PTL's competing with AMD Gorgon Point (N4P), yep 18A comparable to N3P (from CC Wei) is way over-estimate. Thus, 14A is only catching up to TSMC's N2 process. By that time, TSMC will move on to their true A14 process. IFS will forever be catching up to TSMC without customers commitment: the same situation as SF. That's why SF and IFS are rumored to focus on fine-tuninng the current process, namely 18A and SF2.

It's actually good move of LBT to dump external foundry business. By doing that, Intel could save billions dollar especially with no customers committed. The delay of Microsoft's ARM CPU might be the final nail to the coffin...

R&D with 24 hours...
Unfortunately it's not something that can be imitated

 

[moinmoin]

While that's true, it's also important to realize that many foundries are . . . foundering.

Remember when Samsung saved GF by licensing 14LPE and (I guess) LPP? GF wound up making processes that were arguably better than what Samsung licensed to them initially. What if Intel, Samsung, GF, and possibly others had a cross-licensing agreement and formed a multinational fab consortium? Everybody gets to look at everybody else's patents and trade secrets, and then they make the best nodes they can with what's available.

Survive together or die alone. Take your pick!

That's actually happening occasionally. E.g. ASML's current position was essentially intended, with much of the industry rallying behind them when they needed financial support for developing EUV back in 2012. Intel, TSMC and Samsung all invested in ASML as a result of that (much of what's divested again by now).

 

[QuickyDuck]

18A for external customers doesn't seem to pass new CEO's 50% margin goal. Rule is rules. /s

 

[511]

18A for external customers doesn't seem to pass new CEO's 50% margin goal. Rule is rules. /s

The CEO clearly said going forward as in new projects not in flight projects

 

[QuickyDuck]

The CEO clearly said going forward as in new projects not in flight projects

Projects for Amazon and Microsoft will remain, no more new one 🙏

 

[DrMrLordX]

With PTL's competing with AMD Gorgon Point (N4P), yep 18A comparable to N3P (from CC Wei) is way over-estimate.

That could be due to Panther Lake having design issues. But with few-to-no other examples of consumer hardware that will be using Intel 18a, it'll be difficult to differentiate.

IBM has provided process technology to various fabs We also provided technology to the UMC.
The latest project at the moment is 2nm of Rapidas
Originally, it seems to improve the 2nm process node for research at IBM to make it suitable for mass production.

Yeah IBM went out of the main fab business years ago but retained their research fabs/function. I'm saying the industry needs to expand on this model somehow. Intel and Samsung seem to be choking on their own hubris. Less-cutting-edge fabs like TI and GF maybe need to be brought into the modern age.

@moinmoin

That was a step in the right direction, though you'll notice that it only pertained to a very specific part of the node design/development process (albeit an incredibly important part).

 

[Win2012R2]

Yeah OFC Commitment from MSFT/AMZN doesn't exist

They are fully committed to sample it in order to secure better pricing from TSMC, which was probably main motivation for everybody else who engaged.

Even if 18A was free it would have represented unacceptable risk of not getting on time or working in the first place.

Why risk when reliable TSMC is here and any price increases can be passed onto customers wanting latest tech?

 

[511]

They are fully committed to sample it in order to secure better pricing from TSMC, which was probably main motivation for everybody else who engaged.

Even if 18A was free it would have represented unacceptable risk of not getting on time or working in the first place.

Why risk when reliable TSMC is here and any price increases can be passed onto customers wanting latest tech?

Nope those deals are real as hell just not a large volume also the customer for those are MSFT and Amazon themselves and we will see in 1-1.5 year when intel will ship most of their product stack containing 18A.

 

[Win2012R2]

just not a large volume

1) How large is it?
2) How do you know that?

 

[511]

1) How large is it?

Both of them are supposed to be chips for their cloud units so pretty low

2) How do you know that?

it was announced somewhere can't remember

 

[Win2012R2]

Both of them are supposed to be chips for their cloud units so pretty low

How is it viable commercially to take large fixed cost of designing cutting edge stuff and only go for low volume?

Either way it's not good for Intel if volumes are low, granted it was totally unreasonable to expect that anybody would risk high volumes with this - example of Altera and 10nm will be on people's minds, plus Intel is far more sick financially now.

The whole IFS (for external customers) is a fantasy land, probably sucks precious little financial resources that Intel has still got.

 

[511]

How is it viable commercially to take large fixed cost of designing cutting edge stuff and only go for low volume?

Microsoft and Amazon has the pockets for these things

Either way it's not good for Intel if volumes are low, granted it was totally unreasonable to expect that anybody would risk high volumes with this - example of Altera and 10nm will be on people's minds, plus Intel is far more sick financially now.

The whole IFS (for external customers) is a fantasy land, probably sucks precious little financial resources that Intel has still got.

don't forget this IFS is keeping them alive together with client and DC should have done IFS when they were the leader in 14nm era with proper EDA along with client GPUs and they wouldn't be in this position

 

[maddie]

Microsoft and Amazon has the pockets for these things

What does this have to do with "How is it viable commercially to take large fixed cost of designing cutting edge stuff and only go for low volume?"

Having lots of capital doesn't mean you waste it. That's a fool's behavior, and it's what made Intel throw away accumulated earnings by chasing pointless ventures over the years.

 

[johnsonwax]

I've been saying for quite some time that there is only one path for Intel's fab here and it's a government backed, ULA-style consortium for domestic production. The reality of the situation, and why Biden pushed to invest in this direction is that the DOD has aspirations for advanced AI powered military tech, and that's going to require more advanced processes than what the non-Intel domestic fabs can do. ITAR limits who DOD can buy from, and transferring ITAR tech to a player like TSMC even if they are manufacturing in the US gives China additional incentive to invade Taiwan. The best option for the federal government is to salvage Intels fab business, and the only way that's going to happen is to lash it to the American fabless companies - AMD, Qualcomm, Nvidia, Apple, Broadcom, get them to take a stake in the business and guarantee a certain volume of demand from each. Apple doesn't need to move their flagship production there (they could if they wanted), they have plenty of other products that could go to Intel.

The key to semi has always been volume, ever since Intel cut prices on ICs to boost volume because marginal costs were negligible, what was needed was volume to cover the fixed up-front costs. That problem has only gotten bigger as those fixed costs have grown. Intel failed to find that volume market which is why they are where they are. We keep treating this market like other markets. It's not like other markets. This is an industry that has grown beyond even national scale. In order to have a 'competitive market' you need to subdivide it sufficiently which robs the industry of the volume needed to maintain the current technological pace. We can trade competition for progress, but you can't have both. The US can tip up a domestic market around Intel to maintain a domestic control over tech, but it can't be done in a free market manner. If the US wants a strategic advantage then they need to build an economy independent of market forces here. Otherwise, TSMC is going to bury Intel because TSMC has the volume from all of those domestic fabless companies, and Intel doesn't, and nobody other than the federal government, not even Apple, has the cash to change that. The federal government has a choice to make, they don't realize they have a choice to make because they're idiots, and the free market will do what it always seeks to do - create a monopoly and that will be TSMC.

 

[511]

What does this have to do with "How is it viable commercially to take large fixed cost of designing cutting edge stuff and only go for low volume?"

Having lots of capital doesn't mean you waste it. That's a fool's behavior, and it's what made Intel throw away accumulated earnings by chasing pointless ventures over the years.

It's for their cloud business they are not throwing anything well when TSMC started Intel also took the risk with them why won't AMZN/MSFT won't take risk with their old business partner if you don't take risk you can't be first.

I've been saying for quite some time that there is only one path for Intel's fab here and it's a government backed, ULA-style consortium for domestic production. The reality of the situation, and why Biden pushed to invest in this direction is that the DOD has aspirations for advanced AI powered military tech, and that's going to require more advanced processes than what the non-Intel domestic fabs can do. ITAR limits who DOD can buy from, and transferring ITAR tech to a player like TSMC even if they are manufacturing in the US gives China additional incentive to invade Taiwan. The best option for the federal government is to salvage Intels fab business, and the only way that's going to happen is to lash it to the American fabless companies - AMD, Qualcomm, Nvidia, Apple, Broadcom, get them to take a stake in the business and guarantee a certain volume of demand from each. Apple doesn't need to move their flagship production there (they could if they wanted), they have plenty of other products that could go to Intel.

The key to semi has always been volume, ever since Intel cut prices on ICs to boost volume because marginal costs were negligible, what was needed was volume to cover the fixed up-front costs. That problem has only gotten bigger as those fixed costs have grown. Intel failed to find that volume market which is why they are where they are. We keep treating this market like other markets. It's not like other markets. This is an industry that has grown beyond even national scale. In order to have a 'competitive market' you need to subdivide it sufficiently which robs the industry of the volume needed to maintain the current technological pace. We can trade competition for progress, but you can't have both. The US can tip up a domestic market around Intel to maintain a domestic control over tech, but it can't be done in a free market manner. If the US wants a strategic advantage then they need to build an economy independent of market forces here. Otherwise, TSMC is going to bury Intel because TSMC has the volume from all of those domestic fabless companies, and Intel doesn't, and nobody other than the federal government, not even Apple, has the cash to change that. The federal government has a choice to make, they don't realize they have a choice to make because they're idiots, and the free market will do what it always seeks to do - create a monopoly and that will be TSMC.

TSMC has some ownership by Taiwan's Government also how about actually giving Intel the money with stringent requirements than the joke Chips act is and setting favorable policies for TI/Micron/GF/Intel.

 

[johnsonwax]

Having lots of capital doesn't mean you waste it. That's a fool's behavior, and it's what made Intel throw away accumulated earnings by chasing pointless ventures over the years.

It's not entirely pointless ventures, but understand that investors want to get paid more now than they want companies to thrive later. Investors are greedy, selfish, and shortsighted, and they are now incredibly powerful. Every dollar a company might have available to invest in R&D is a dollar investors want returned to them. Investors are bleeding the place dry. I should know, I am one and I see it everywhere. And it's never been this bad before.

 

[511]

And in the end the one who pays for it are the Employees of the company. No one spends as much as time for the company as the employees they get paid but they suffer as well.
In recent time layoffs and stuff has increased too much

 

[johnsonwax]

It's for their cloud business they are not throwing anything well when TSMC started Intel also took the risk with them why won't AMZN/MSFT won't take risk with their old business partner if you don't take risk you can't be first.

TSMC has some ownership by Taiwan's Government also how about actually giving Intel the money with stringent requirements than the joke Chips act is and setting favorable policies for TI/Micron/GF/Intel.

Well, yeah, the CHIPS act was fairly decent sentiment married to neoliberal execution with the usual results from such a pairing. Capitalism makes no promises to be compatible with national strategic interest. But lawmakers know who pays them.

 

[Hitman928]

How is it viable commercially to take large fixed cost of designing cutting edge stuff and only go for low volume?

Either way it's not good for Intel if volumes are low, granted it was totally unreasonable to expect that anybody would risk high volumes with this - example of Altera and 10nm will be on people's minds, plus Intel is far more sick financially now.

The whole IFS (for external customers) is a fantasy land, probably sucks precious little financial resources that Intel has still got.

If memory serves, the MS deal isn’t really an IFS deal, it’s really a design win for a custom chip from Intel, which of course they will use 18a to fabricate. Under Pat, they decided to count it under IFS to help prop up the 5N4Y IFS 2.0 story.

 

[511]

Earnings for this quarter will be fun cause they said on external customer is going to tape out on 18A if only some analyst doesn't mess things up.

 

[Io Magnesso]

How is it viable commercially to take large fixed cost of designing cutting edge stuff and only go for low volume?

That's what many companies do
You can usually do it if you're a big company