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

 

[adroc_thurston]

Also, with rate of improvement in phones slowing, there is a lot less reason to upgrade, unless the phone is stolen or broken.

yeah phone cycles are slow as hell now.

 

[Antey]

It's getting very close to PC cycle. An almost 10 Years old PC (2016) now is still ''good''... 6700k + 1070. A 10 years old phone (iphone 6) is not that great today, but a 8 years old one (iphone X) is still good to go.

 

[adroc_thurston]

An almost 10 Years old PC (2016) now is still ''good''... 6700k + 1070

no it's not.
PC and phone cycles are both ~5y now.

 

[oak8292]

Good Luck AMD to get capacity

AMD will get all the capacity they contract for in their wafer supply agreement. These stories about ‘starving’ competitors don’t make sense. TSMC builds contracted capacity. Apple is getting a plant because they contracted for a plant.

The rumor is that Intel got ‘early’ N3 capacity because they gave up EUV contracts to TSMC which were installed in a TSMC R&D lab fab. I am sure they also wrote a multi year supply contract to go with the equipment.

AMD and Nvidia are getting the capacity they wrote contracts for. Writing a wafer supply contract to starve a competitor is really expensive. At TSMC if you aren’t actually paying for the finished wafers they will probably be offered at a discount to a competitor because you paid a penalty for not taking all the wafers you contracted for.

AMD paid wafer supply agreement penalties to Global Foundries and I believe that has made them conservative on contracts.

Pat Gelsinger was obviously extremely aggressive with wafer supply agreements. There was an agreement in place at TSMC and he wrote two more with both Apollo and Brookfield in SCIPs. It is one thing to have ‘shell ahead’ which is Intel strategy he was returning to however it is a big commitment to fill the fab.

No, it's just that there's enough N2 for everyone.
But also yes, between GPGPU slop and eating Intel's share, every AMD wafer shipped is a net win for TSM.

Doesn't matter, AMD and NV volumes grow TSMC wafer run rates, Apple doesn't. Apple lives in stale bread markets.

The ‘stale’ market is still over 20% of TSMC’s revenue and still the highest.

 

[adroc_thurston]

The ‘stale’ market is still over 20% of TSMC’s revenue and still the highest.

Irrelevant, it's not growing either wafer or packaging volumes.

 

[gdansk]

Phones are a big customer (for TSMC) and especially for the leading edge processes. But the point is they're not growing in importance for TSMC. Total smartphone shipment trends downward. Total data center shipment trends upward. Both have increasing die sizes to offset and augment their trend, respectively.

And even the rumor has a smaller portion of wafers going to Apple for N2 compared to N3. Not sure why that'd be ominous for AMD. Or anyone really.

 

[adroc_thurston]

And even the rumor has a smaller portion of wafers going to Apple for N2 compared to N3. Not sure why that'd be ominous for AMD. Or anyone really.

They're not even the first N2 customer anyhow.
People just see 'Apple' and start jacking off furiously for some reason.

 

[Keller_TT]

People buy multiple phones. I see that with Samsung users, they buy the Fold and the S25 Ultra. It will be more apparent with Apples user base.

Lol. No. People don't run splurging cash on something just a basic necessity. I'm still on my iPhone 13 mini. Treating it with care, and plan on using it until the hardware gives up. I don't even need new deluded iOS takes and 18.6.x it will be for a few more years.

Since my app store location is Germany, I get the benefit of AltStore too even if I live in Zurich and travel elsewhere. That negated the need for me to consider Pixel with GrapheneOS & microG.

 

[oak8292]

Irrelevant, it's not growing either wafer or packaging volumes.

Irrelevant to what? Tell me more.

 

[adroc_thurston]

Irrelevant to what?

Irrelevant to TSM growth prospects.
They want a fat slice of HPC pie and that's where their continuous invenstment goes.

 

[Josh128]

It's getting very close to PC cycle. An almost 10 Years old PC (2016) now is still ''good''... 6700k + 1070. A 10 years old phone (iphone 6) is not that great today, but a 8 years old one (iphone X) is still good to go.

Nope, Skylake is not secure boot capable, therefore not Windows 11 compatible. Gotta go newer. That is IF you need Win11, which is required by a lot of corpo ITs now.

 

[Antey]

Nope, Skylake is not secure boot capable, therefore not Windows 11 compatible. Gotta go newer. That is IF you need Win11, which is required by a lot of corpo ITs now.

it's not windows 11 compatible because microsoft wants $$

Nothing else.

 

[inquiss]

People buy multiple phones. I see that with Samsung users, they buy the Fold and the S25 Ultra. It will be more apparent with Apples user base.

4 people do. Didn't think we'd see actual intel execs on this forum, but there you go. Anandtech forums has peaked.

 

[adroc_thurston]

Didn't think we'd see actual intel execs on this forum, but there you go

good lord yes that CCG boondoggle saying everyone's buying 2 laptops now only for the market to crash 2 quarters later.

 

[inquiss]

good lord yes that CCG boondoggle saying everyone's buying 2 laptops now only for the market to crash 2 quarters later.

Real cocaine moment.

 

[oak8292]

Irrelevant to TSM growth prospects.
They want a fat slice of HPC pie and that's where their continuous invenstment goes.

I will agree that wafer volume growth is going to be in AI but I also think that Apple is relevant to TSMC's growth and I don't think you understand TSMC's business model.

TSMC does not pick customers to be winners. They took on Intel as a customer even with AMD as a customer. Either one can grow the business.

Nvidia was a prior customer with a strong relationship between Jensen and Morris. The relationship came apart at 28 nm when Nvidia blamed TSMC for their yield problems. Nvidia shifted a lot of wafers to Samsung. Nvidia is back at TSMC as are Qualcomm.

If Nvidia wants to grow at TSMC they just need to contract for wafers in the same way that Apple contracts for wafers. There are wafer supply agreements that will build any capacity. The sooner they write the contract the more likely they are to get the capacity as there are also equipment constraints. Apple has a 'stale' business that lets them plan capacity ahead of most of their competitors. This is an advantage at TSMC as they can order EUV equipment ahead and build fab shells.

Nvidia may have exciting growth but this can disappear in an instant and then who pays for depreciation on equipment TSMC or Nvidia?

 

[adroc_thurston]

but I also think that Apple is relevant to TSMC's growth

how.
Apple volume is in smartphones which bring limited Si content, stale ASPs and stale or declining volumes.

TSMC does not pick customers to be winners.

Yeah they do. But everyone does.

They took on Intel as a customer even with AMD as a customer

Both have been fabbing at TSM since times immemorial.

The relationship came apart at 28 nm when Nvidia blamed TSMC for their yield problems

20, 20nm was the explosive pile of dung.

Nvidia shifted a lot of wafers to Samsung

Everyone did.

Nvidia is back at TSMC as are Qualcomm.

That's because Samsung can't yield. Like, for reals.

 

[oak8292]

how.
Apple volume is in smartphones which bring limited Si content, stale ASPs and stale or declining volumes.

TSMC revenue from Apple increased by 14% from 2023 to 2024 to over $20 billion. Customer B, probably Nvidia, just popped up for the first year in 2024 as a 10% customer at about $12 billion. Customer C showed up as a 10% customer for one year in 2023 and slipped back below the reporting threshold. Meaning the didn’t grow fast enough to keep up.

Yeah they do. But everyone does.

TSMC amortizes equipment over 5 years. This isn’t a place for excitement it is a place for steady demand and growth. They contract for as long as they can.

20, 20nm was the explosive pile of dung.

Nvidia had problems with 28 nm at TSMC. The 20 nm only worked for Apple who wasn’t running high frequencies at the time, however it was 28 nm where Nvidia had yield problems and wanted TSMC to carry more of the start up cost.

Everyone did.

Not Apple they moved to TSMC.

That's because Samsung can't yield. Like, for reals.

The concept that technology is exciting and moving fast doesn’t work in manufacturing. Samsung isn’t yielding, Intel is struggling and TSMC is building to contract.

The PC market is ‘dead’, the mobile market is ‘stagnant’. A lot of the new revenue is in AI but a lot of that is margin and not much is in wafer cost. What is a manufacturer to do? Build to contract and backfill with secondary customers which is the same old concept.

The shiny new thing isn’t always shiny. GPU demand from crypto was a flash in the pan and a real pain at one point. Crypto ASICs at TSMC generated some revenue but they weren’t writing contracts and they were just fill in wafers. TSMC didn’t add capacity for crypto customers.

Nvidia pays for five years of wafers and they get their own plant. No favorites.

 

[adroc_thurston]

TSMC revenue from Apple increased by 14% from 2023 to 2024 to over $20 billion. Customer B, probably Nvidia, just popped up for the first year in 2024 as a 10% customer at about $12 billion. Customer C showed up as a 10% customer for one year in 2023 and slipped back below the reporting threshold. Meaning the didn’t grow fast enough to keep up.

wordcel soup, the reality is that wafer and ASP growth at TSM is driven by HPC.

TSMC amortizes equipment over 5 years. This isn’t a place for excitement it is a place for steady demand and growth. They contract for as long as they can.

wordswords.

Nvidia had problems with 28 nm at TSMC. The 20 nm only worked for Apple who wasn’t running high frequencies at the time, however it was 28 nm where Nvidia had yield problems and wanted TSMC to carry more of the start up cost.

NV had problems with 40 and 20. 28 went fine for everyone.

Not Apple they moved to TSMC.

Huawei was always there. In any case, Samsung did well when they had competitive nodes that yielded.

 

[DrMrLordX]

28 went fine for everyone.

Sure seemed like 28nm processes didn't go terribly well for AMD/Globalfoundries.

 

[adroc_thurston]

Sure seemed like 28nm processes didn't go terribly well for AMD/Globalfoundries.

They were alright?

 

[DrMrLordX]

They were alright?

It's in the past now so maybe it isn't worth revisiting it, but from what The_Stilt was telling us back when he still posted here, it wasn't . . . great. According to what I remember him saying, there was never going to be a 3M/6t or 4M/8t Steamroller now matter how many of us wished for one, at least not on GF28nm.

 

[adroc_thurston]

there was never going to be a 3M/6t or 4M/8t Steamroller now matter how many of us wished for one, at least not on GF28nm.

Well yeah they weren't frequency-optimized nodes.
Expected, really?

 

[reaperrr3]

Sure seemed like 28nm processes didn't go terribly well for AMD/Globalfoundries.

Well, AMD's CPU uArch was goo and Intel was ahead of everyone with 22nm FinFET (and later 14nm).

Even at slightly above Piledriver clocks, 4M/8t Steamroller and Excavator wouldn't have moved the needle in server sales much, if at all, and they weren't competitive in the main desktop use case (gaming) either, so... not like a few hundred Mhz of CPU clock made a real difference here.

 

[Doug S]

TSMC revenue from Apple increased by 14% from 2023 to 2024 to over $20 billion. Customer B, probably Nvidia, just popped up for the first year in 2024 as a 10% customer at about $12 billion. Customer C showed up as a 10% customer for one year in 2023 and slipped back below the reporting threshold. Meaning the didn’t grow fast enough to keep up.

TSMC said that Apple was 22% of their revenue. That's only going to increase next year with Apple having TSMC fab their modem and wifi/BT chips they used to order from others. TSMC has been reporting YoY revenue growth of 30-40% this year, and even if Apple is responsible for 14% of that again it is probably safe to assume the rest is mostly "customer B".

So Nvidia is likely to overtake Apple as the biggest customer either this year or next, but even if they do they probably don't stay ahead because at some point the AI bubble will burst. Meanwhile Apple's smartphone and Mac market will remain pretty stable. They may not be growing the number of phones they sell each year, but it isn't shrinking either. At least not until something comes along that disrupts the smartphone market the way the smartphone disrupted other markets. I'm sure that will happen someday, but clearly not anytime soon.