Discussion Leading Edge Foundry Node advances (TSMC, Samsung Foundry, Intel)

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.

SiliconFly · Apr 4, 2024

FlameTail said:
2023: A17 = N3B
2024: A18 = N3E
2025: A19 = N3P
2026: A20 = N2+BSPDN
2027: A21 = N2P

Is it true Apple is skipping N2 in 2025 and going to N2+BSPD in 2026? Or is it just a guess?

(If possible, pls cite some source of the leak; assuming it's a leak.)

Well, if the above speculation/leak/fact is true, it's not looking good for TSMC N2 in 2025.

Hitman928 · Apr 4, 2024

SiliconFly said:
Good for you. Any references/links by any chance?

PPAC has been around a while now. Quick search brought up this IEEE paper from 2021 but the term predates that even.

https://ieeexplore.ieee.org/document/9592598

Curious_Inquirer · Apr 4, 2024

igor_kavinski said:
Pasting a paid article would probably break some forum rule...

Can you PM me the article as well? And thank you.

Doug S · Apr 4, 2024

SiliconFly said:
Is it true Apple is skipping N2 in 2025 and going to N2+BSPD in 2026? Or is it just a guess?

(If possible, pls cite some source of the leak; assuming it's a leak.)

Well, if the above speculation/leak/fact is true, it's not looking good for TSMC N2 in 2025.

You can infer that Apple will skip N2 based on the timing. It is scheduled to enter mass production in H2 2025, too late for Apple. N2+BSPDN follows six months later, exactly on schedule for Apple. I don't think that's any coincidence. They've always scheduled around Apple, N3 has been an exception because of the problems they ran into.

TSMC's stated reason for splitting off BSPDN is that some customers have told them they do not want to incur the increased cost. So presumably there are customers for that, either due to cost or due to the opportunity to be at the front of the line for a new TSMC process for once instead of fighting with others to take the scraps Apple leaves behind in the first 3-6 months.

SiliconFly · Apr 4, 2024

Doug S said:
You can infer that Apple will skip N2 based on the timing. It is scheduled to enter mass production in H2 2025, too late for Apple. N2+BSPDN follows six months later, exactly on schedule for Apple. I don't think that's any coincidence. They've always scheduled around Apple, N3 has been an exception because of the problems they ran into.

TSMC's stated reason for splitting off BSPDN is that some customers have told them they do not want to incur the increased cost. So presumably there are customers for that, either due to cost or due to the opportunity to be at the front of the line for a new TSMC process for once instead of fighting with others to take the scraps Apple leaves behind in the first 3-6 months.

Tbh, N2 in 2025 is a bit too soon for smaller customers too. And it looks like big players like Apple, NVidia, AMD & Intel aren't using N2 in 2025 either. So, N2 is going to face quite a bit of problem late next year giving 18A a clear opportunity. If Intel doesn't screw-up big time (like usual), they can easily take the lead.

Doug S · Apr 4, 2024

SiliconFly said:
Tbh, N2 in 2025 is a bit too soon for smaller customers too. And it looks like big players like Apple, NVidia, AMD & Intel aren't using N2 in 2025 either. So, N2 is going to face quite a bit of problem late next year giving 18A a clear opportunity. If Intel doesn't screw-up big time (like usual), they can easily take the lead.

But what major customers is Intel getting for 18A? They might be able to claim "process superiority" (in some factors, certainly not in density) but if they're only using it for their own stuff plus a few niche customers it is hardly hurting TSMC at all.

Hitman928 · Apr 4, 2024

Doug S said:
But what major customers is Intel getting for 18A?

None so far, according to their CFO:

David Zinsner said:
We probably won't win anybody's major volume in 18A. We'll win some smaller SKUs, and that's all we need, to be honest with you.

https://seekingalpha.com/article/4676516-intel-corporation-intc-morgan-stanley-technology-media-and-telecom-conference

Hitman928 · Apr 4, 2024

SiliconFly said:
Tbh, N2 in 2025 is a bit too soon for smaller customers too. And it looks like big players like Apple, NVidia, AMD & Intel aren't using N2 in 2025 either. So, N2 is going to face quite a bit of problem late next year giving 18A a clear opportunity. If Intel doesn't screw-up big time (like usual), they can easily take the lead.

I think going into volume production at the end of 2025 fits into AMD's timeline pretty well.

FlameTail · Apr 4, 2024

SiliconFly said:
Good for you. Any references/links by any chance?

From semianalysis

FlameTail · Apr 4, 2024

SiliconFly said:
Is it true Apple is skipping N2 in 2025 and going to N2+BSPD in 2026? Or is it just a guess?

(If possible, pls cite some source of the leak; assuming it's a leak.)

Well, if the above speculation/leak/fact is true, it's not looking good for TSMC N2 in 2025.

Not leak. Speculation.

FlameTail · Apr 4, 2024

Hitman928 said:
I think going into volume production at the end of 2025 fits into AMD's timeline pretty well.

Zen6?

Hitman928 · Apr 4, 2024

FlameTail said:
Zen6?

Yes. I'm not a leaker, it just seems to line up well, schedule wise, and AMD has a couple of product lines that could easily afford to purchase leading edge production with good volume.

FlameTail · Apr 5, 2024

Samsung 5LPE -> 4LPP seems to be good efficiency bump.

E1380 = 9.6441281139 points/W
E1480 = 11.5401459854 points/W

That's a 20% jump in power efficiency. Even better, we have actually gone up the frequency-power curve, as the A78 core in E1480 is running at 2.75 GHz, whereas the A78 in E1380 runs at 2.4 GHz.

and this is not even Samsung's best 4nm node.

Doug S · Apr 5, 2024

Hitman928 said:
I think going into volume production at the end of 2025 fits into AMD's timeline pretty well.

But when has AMD ever aggressively used TSMC's latest process immediately after it is available?

Some might excuse that by saying "well Apple always takes all the initial supply" but if you're looking for a deep pocketed company that will be next in line behind Apple to take that supply that's Nvidia not AMD.

Heck if Nvidia doesn't get some real competition soon (and the AI bubble doesn't burst) they might overtake Apple as TSMC's biggest customer by revenue two or three years from now.

gdansk · Apr 5, 2024

Doug S said:
But when has AMD ever aggressively used TSMC's latest process immediately after it is available?

You have to go back some time but if I recall correctly Jaguar was on TSMC's 28nm before Apple A7.

beginner99 · Apr 5, 2024

Doug S said:
and the AI bubble doesn't burst

It will burst or rather slowly deflate. Once large fraction of companies are setup on azure AI or whatever other services exist and they don't need to install new servers en masse daily, then the volume of cards will simply start to go down. Then companies realize the millions they are spending per year APi fees and internal app development can't really be justified, usage will be limited and will go down.

So I don't see a burst, just a slow decline.

One could argue video generation models will further drive need for GPUs. certainly some demand yes but I see video much less generally usable than LLMs.

jpiniero · Apr 5, 2024

gdansk said:
You have to go back some time but if I recall correctly Jaguar was on TSMC's 28nm before Apple A7.

Yeah. 28 nm and earlier... AMD/RTG and/or NV would be the first to use a TSMC node.

Doug S · Apr 5, 2024

beginner99 said:
It will burst or rather slowly deflate. Once large fraction of companies are setup on azure AI or whatever other services exist and they don't need to install new servers en masse daily, then the volume of cards will simply start to go down. Then companies realize the millions they are spending per year APi fees and internal app development can't really be justified, usage will be limited and will go down.

So I don't see a burst, just a slow decline.

One could argue video generation models will further drive need for GPUs. certainly some demand yes but I see video much less generally usable than LLMs.

Sure but the question is when does that limit get hit. We could see first signs of a slowdown by this time next year, or maybe they are selling every card they can make through the end of the decade. I think a reasonable case could be made for either scenario, there are a lot of unknowns and it is hard to separate the hype from the actual cost savings for corporate implementations (i.e. replacement of the hordes of offshore CSRs, etc.)

A couple years ago the first time Apple hit $3 trillion market cap I cut back my holdings significantly, both because that had grown to far too much of my overall portfolio and because I figured they'd start hitting roadblocks like regulation or inability to further expand their current markets. One of the stocks I bought was Nvidia. I wish I had bought a LOT more of course, but FWIW I'm sticking with it despite its massive growth over the past couple of years because I think it has more room to run and will eventually leave Apple and Microsoft in the dust and stands the best chance of becoming the first stock to eclipse $4 trillion market cap.

igor_kavinski · Apr 5, 2024

Doug S said:
One of the stocks I bought was Nvidia. I wish I had bought a LOT more of course, but FWIW I'm sticking with it despite its massive growth over the past couple of years because I think it has more room to run and will eventually leave Apple and Microsoft in the dust and stands the best chance of becoming the first stock to eclipse $4 trillion market cap.

It's a very daring bet to assume that the combined forces of Microsoft, Google, AMD, Intel and Tenstorrent won't be able to challenge and diminish Nvidia's role in the AI space. Sure they must've some of the brightest people on the planet but not ALL of the brightest people on the planet. Someone on the other side making a big enough breakthrough in AI and then patenting it could easily put Nvidia's market leading position in peril.

Hitman928 · Apr 5, 2024

Doug S said:
But when has AMD ever aggressively used TSMC's latest process immediately after it is available?

Some might excuse that by saying "well Apple always takes all the initial supply" but if you're looking for a deep pocketed company that will be next in line behind Apple to take that supply that's Nvidia not AMD.

Heck if Nvidia doesn't get some real competition soon (and the AI bubble doesn't burst) they might overtake Apple as TSMC's biggest customer by revenue two or three years from now.

If AMD continues to execute at their current level, I don't think having the money to spend will be the issue if they feel it is justified and I think AMD has justification for at least some select product lines now to use bleeding edge nodes. It also comes down to roadmap. Apple not only got the first dibs on new nodes because they had the money and volume, but because they had the consistent demand for it. Even if (theoretically) a company came in with super deep pockets and wanted first customer status on a node but had no dependable plans for the next node, I don't think that they would have any chance of leap frogging Apple because TSMC knows Apple will spend now and at each subsequent node as well. With AMD's roadmap and execution, they're starting to get into that position as well. Obviously it helps that timing wise, Apple will probably wait for N2+ anyway so there wouldn't be conflict in that regard.

As far as NV goes, maybe. Looking at their release dates for A100 and H100, late 2025 seems a little early to start pumping high volume for their next gen, since B100 won't be launching until later this year. If B100 launches sooner or they just accelerate between gens, they could potentially also take that first customer spot.

igor_kavinski · Apr 5, 2024

Doug S said:
stands the best chance of becoming the first stock to eclipse $4 trillion market cap.

If you've watched the latest Veritasium video on the number 37, I predict there's a 37 percent chance that Nvidia is able to eclipse 3.7 trillion market cap.

SiliconFly · Apr 5, 2024

Hitman928 said:
I think going into volume production at the end of 2025 fits into AMD's timeline pretty well.

Not practical imho. Just cos TSMC pulled in N2 by a year, doesn't automatically mean AMD can easily follow suit. Completing a design earlier than planned, validating it and releasing it to manufacturing early by an year is not a reasonable scenario. We can't expect N2 based Zen6 parts next year I think.

Doug S said:
But when has AMD ever aggressively used TSMC's latest process immediately after it is available?

True. Cutting edge nodes are freaking expensive and only super premium product makers like Apple can afford them. The rest have to wait for their turn.

Hitman928 · Apr 5, 2024

SiliconFly said:
Not practical imho. Just cos TSMC pulled in N2 by a year, doesn't automatically mean AMD can easily follow suit. Completing a design early, validating it and releasing it to manufacturing early by an year is not a reasonable scenario I think. We can't expect N2 based Zen6 parts next year.

TSMC going HVM late 2025 means products on shelves in 2026. The change in N2's schedule would have been communicated to key partners before we heard anything about it. Should be plenty of time to get a Zen 6 variant on N2, if they so choose.

Edit: Where are you getting the N2 pull in, anyway? Are you sure you're not thinking of N3(B)?

SiliconFly · Apr 5, 2024

Hitman928 said:
TSMC going HVM late 2025 means products on shelves in 2026. The change in N2's schedule would have been communicated to key partners before we heard anything about it. Should be plenty of time to get a Zen 6 variant on N2, if they so choose.

Edit: Where are you getting the N2 pull in, anyway? Are you sure you're not thinking of N3(B)?

Even if they communicate, the real problem lies with the customer not having enough time to finish design/validation. Takes many years. Can't usually be altered midway in the cycle. N2 actually (without BSPD).

Hitman928 · Apr 5, 2024

SiliconFly said:
Even if they communicate, the real problem lies with the customer not having enough time to finish design/validation. Takes many years. Can't usually be altered midway in the cycle. N2 actually (without BSPD).

How long ago do you think N2 was known? When did it get pulled in by 6 months?

Discussion Leading Edge Foundry Node advances (TSMC, Samsung Foundry, Intel)

Golden Member

Golden Member

Diamond Member

Junior Member

Platinum Member

Golden Member

Platinum Member

Diamond Member

Diamond Member

Platinum Member

Platinum Member

Platinum Member

Diamond Member

Platinum Member

Platinum Member

Platinum Member

Diamond Member

Lifer

Platinum Member

Lifer

Diamond Member

Lifer

Golden Member

Diamond Member

Golden Member

Diamond Member