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.

 

[Doug S]

for a process tech 1 year is nothing. it takes 3-5 years of development. TSMC is probably at somthing beyond N2 class as well. Samsung did mention 1.4 as well. Its just an announcement but any major change will be known like when Forksheets will be adopted.

When TSMC announces "volume production" they are really producing in volume. They've had volume production of a process begin in April/May and shipped it in millions of iPhones in September.

I wasn't talking about Intel announcing "we have this process on the roadmap", just their ridiculous claims of volume production that don't hold up to scrutiny. Maybe by "volume production" they mean what TSMC means when they say "risk production"? That's the only explanation that makes sense at least based on Intel 4. We'll have to see how it pans out with Intel 3, 20A and 18A to get a feel for how long it takes before they announce volume production and actual chips hitting the street.

 

[DavidC1]

If it takes a year from when Intel announces "18A ready for volume production" and the first chips reaching the market, as was the case for Intel 4, then we are still far away from 18A actually mattering.

Sierra Forest is H1, so that's pretty quick, certainly much faster than predecessors. For 20A, they claim H1 2024 for production, and rumors say late Q4/Early Q1 for Arrowlake. We'll see how Clearwater Forest does, but if it's same H2/H1 deal as SRF, that's pretty quick too.

 

[FlameTail]

What is the difference between Forests and Rapids?

 

[FlameTail]

This 5% decrease is meme tier.

 

[Tigerick]

What is the difference between Forests and Rapids?

Rapids - P cores. Forest - E cores

 

[Ajay]

View attachment 91200
This 5% decrease is meme tier.

Poor Samsung. Terrible nodes for Logic, excellent for Memory/Storage.

 

[H433x0n]

When TSMC announces "volume production" they are really producing in volume. They've had volume production of a process begin in April/May and shipped it in millions of iPhones in September.

I wasn't talking about Intel announcing "we have this process on the roadmap", just their ridiculous claims of volume production that don't hold up to scrutiny. Maybe by "volume production" they mean what TSMC means when they say "risk production"? That's the only explanation that makes sense at least based on Intel 4. We'll have to see how it pans out with Intel 3, 20A and 18A to get a feel for how long it takes before they announce volume production and actual chips hitting the street.

Eh, TSMC N3B was announced for volume production in late December 2022, they shipped in products in September 2023.

I think there’s a lot of misinfo on this topic. All it means is that it hit satisfactory defect density and has associated IP (PDK) that qualifies it for HVM. I don’t think Intel or TSMC being deceptive when they say a process is ready for HVM.

 

[mikegg]

Eh, TSMC N3B was announced for volume production in late December 2022, they shipped in products in September 2023.

Only because the iPhone always ships in September.

 

[H433x0n]

Only because the iPhone always ships in September.

I don’t doubt that.

Similarly, it’s pretty clear MTL was half baked when it launched with unfinished microcode. So who’s to say it couldn’t have been launched 3-4 months earlier?

The fabs can and should announce when nodes have qualified for HVM independently of when a product is ready.

 

[FlameTail]

Poor Samsung. Terrible nodes for Logic, excellent for Memory/Storage.

Even in memory/storage, their leadership position is eroding.

 

[Doug S]

Only because the iPhone always ships in September.

N3B is also a highly problematic process (to put it charitably)

I'm sure "N3" (before that scarlet letter "B" was added) was at first roadmapped to arrive on TSMC's usual cadence and used for A16 in 2022. Judging by the timing of first rumors circulating of its problems/delay it seems they recognized some serious issues when they were trying to move it into whatever stage they call the one before risk production (taking it out of the research phase and moving into a real line to begin all the steps they need to take before risk production can start)

If N3B had any other customers with designs ready before Apple it would have shipped sooner. But Intel is the only other customer for it (and no one can yet claim with absolute certitude their stuff will use N3B instead of N3E) so having a production ready process with no chips to produce is going to delay things.

Now it is possible the same thing happened with Intel 4. Maybe the fab was ready to go and Intel could have shipped Intel 4 chips months earlier, but they had to go through multiple steppings to get things working, and the delay is attributable to their chip architects not their process engineers. If that's the case Intel is unlikely to be forthcoming, given the scrutiny they are under from Wall Street.

 

[DrMrLordX]

Now it is possible the same thing happened with Intel 4. Maybe the fab was ready to go and Intel could have shipped Intel 4 chips months earlier, but they had to go through multiple steppings to get things working, and the delay is attributable to their chip architects not their process engineers. If that's the case Intel is unlikely to be forthcoming, given the scrutiny they are under from Wall Street.

Too bad Intel either cancelled every other product on Intel 4 or moved them to a different process.

 

[trivik12]

When TSMC announces "volume production" they are really producing in volume. They've had volume production of a process begin in April/May and shipped it in millions of iPhones in September.

I wasn't talking about Intel announcing "we have this process on the roadmap", just their ridiculous claims of volume production that don't hold up to scrutiny. Maybe by "volume production" they mean what TSMC means when they say "risk production"? That's the only explanation that makes sense at least based on Intel 4. We'll have to see how it pans out with Intel 3, 20A and 18A to get a feel for how long it takes before they announce volume production and actual chips hitting the street.

TSMC announced volume start of N3B end of last year.

Look at their own news site. https://www.tomshardware.com/news/tsmc-kicks-off-3nm-production

 

[Doug S]

TSMC announced volume start of N3B end of last year.

Look at their own news site. https://www.tomshardware.com/news/tsmc-kicks-off-3nm-production

Yes, and?

That date is irrelevant because they had to wait for months until their one and only customer for N3B actually needed chips.

 

[FlameTail]

Question: Considering how huge the demand for Nvidia's H100 card is, and how the supply from TSMC is limited, why don't they port it to Samsung's node and thus dual-source from both TSMC and Samsung?

It will allow them to supply more cards, as their is clearly terrific demand for them.

 

[Ajay]

Question: Considering how huge the demand for Nvidia's H100 card is, and how the supply from TSMC is limited, why don't they port it to Samsung's node and thus dual-source from both TSMC and Samsung?

It will allow them to supply more cards, as their is clearly terrific demand for them.

Hold on a bit, I’ll get Jensen of the phone…

 

[H433x0n]

Question: Considering how huge the demand for Nvidia's H100 card is, and how the supply from TSMC is limited, why don't they port it to Samsung's node and thus dual-source from both TSMC and Samsung?

It will allow them to supply more cards, as their is clearly terrific demand for them.

They would but it’s not possible for a few reasons. 1) The existing cooling and power delivery wouldn't suffice if ported to Samsung and 2) The PDKs that Samsung offers are notoriously bad.

 

[Tigerick]

Hold on a bit, I’ll get Jensen of the phone…

Jensen: Been there, done that....run

 

[FlameTail]

2) The PDKs that Samsung offers are notoriously bad.

Perhaps that would explain why almost every single chip fabbed by Samsung in recent times has not been good, ranging from mediocre to straight up disastrous.

Exynos 990, Exynos 2100, Exynos 2200, Snapdragon 888, Snapdragon 8 Gen 1, Tensor, Tensor G2, Tensor G3, Exynos 1280, Exynos 1380...

 

[FlameTail]

Unfortunately, Intel does not have a whale of a customer for the Intel foundry group as of yet. Making chiplets internally does not compare to TSMC manufacturing complex SoCs for whales like Apple and Qualcomm.

2024 Big Race is TSMC N2 and Intel 18A - Semiwiki

There is a lot being said about Intel getting the lead back from TSMC with their 18A process. Like anything else in the semiconductor industry there is much more here than meets the eye, absolutely. From the surface, TSMC has a massive ecosystem and is in the lead as far as process technologies...

semiwiki.com

This is a profound statement. Intel may have the best node in the industry, but it's no use if they don't have customers.

Apple, Qualcomm, Nvidia, AMD are all "whales". Which one of them will most likely jump ship to Intel Foundry?

 

[Ajay]

2024 Big Race is TSMC N2 and Intel 18A - Semiwiki

There is a lot being said about Intel getting the lead back from TSMC with their 18A process. Like anything else in the semiconductor industry there is much more here than meets the eye, absolutely. From the surface, TSMC has a massive ecosystem and is in the lead as far as process technologies...

semiwiki.com

This is a profound statement. Intel may have the best node in the industry, but it's no use if they don't have customers.

Apple, Qualcomm, Nvidia, AMD are all "whales". Which one of them will most likely jump ship to Intel Foundry?

Qualcomm maybe. AMD never - unless Intel actually spins off IFS (which they don’t plan on doing).

 

[igor_kavinski]

Which one of them will most likely jump ship to Intel Foundry?

I don't think anyone is dying to fab their designs with Intel. Their primary concern would be delays, yields, hardheadedness of proud Intel semi engineers and the conflict of interest of sharing their designs with a competitor. TSMC does not make any CPUs of their own. That's why everyone has no qualms about them. Even if IFS were to be spun off, I find it hard to believe that no one would have the slightest distrust. After all, it's Intel. A fiercely competitive company. And all is fair in competition, including peeking at the designs of customers being fabbed in their sister company. Nope. I would not trust Intel with that. Not unless I dedicated significant resources to ensuring that my IP is viewed only by eyes that really need to see them.

 

[clemsyn]

2024 Big Race is TSMC N2 and Intel 18A - Semiwiki

There is a lot being said about Intel getting the lead back from TSMC with their 18A process. Like anything else in the semiconductor industry there is much more here than meets the eye, absolutely. From the surface, TSMC has a massive ecosystem and is in the lead as far as process technologies...

semiwiki.com

This is a profound statement. Intel may have the best node in the industry, but it's no use if they don't have customers.

Apple, Qualcomm, Nvidia, AMD are all "whales". Which one of them will most likely jump ship to Intel Foundry?

Didn't Intel received a large prepayment from an unnamed customer already to use their 18A process?

Intel's Foundry Business Gets a Huge Vote of Confidence | The Motley Fool

An unnamed customer has put up cash to secure Intel 18A manufacturing capacity.

www.fool.com

 

[Doug S]

Didn't Intel received a large prepayment from an unnamed customer already to use their 18A process?

Intel's Foundry Business Gets a Huge Vote of Confidence | The Motley Fool

An unnamed customer has put up cash to secure Intel 18A manufacturing capacity.

www.fool.com

And they can use whatever definition they want for "large". Could easily have been the US government i.e. government adjacent military industrial complex company, and have nothing to do with the "whale" customers mentioned above.

Qualcomm is likely to dabble with Intel just as they do with Samsung, to keep everyone honest on pricing and because they have so many different product lines they can always find something to do with capacity they buy. Also gets them a foot in the door if they decide to make a move in the future.

But AMD won't consider them unless it is spun off, and now that Intel is trying to compete in AI/GPU Nvidia will probably feel the same. Intel simply doesn't have the capacity to handle Apple (don't forget how just making modems for ~ half of Apple's yearly iPhone shipments was causing them all sorts of problems) and that won't change until the Ohio fabs are online, and they have a cozy relationship with TSMC that will make them difficult to pry loose.

Intel has their work cut out for them trying to come within distant shouting distance of TSMC, but they can probably overtake Samsung without much difficulty.

 

[Ajay]

And they can use whatever definition they want for "large". Could easily have been the US government i.e. government adjacent military industrial complex company, and have nothing to do with the "whale" customers mentioned above.

Qualcomm is likely to dabble with Intel just as they do with Samsung, to keep everyone honest on pricing and because they have so many different product lines they can always find something to do with capacity they buy. Also gets them a foot in the door if they decide to make a move in the future.

But AMD won't consider them unless it is spun off, and now that Intel is trying to compete in AI/GPU Nvidia will probably feel the same. Intel simply doesn't have the capacity to handle Apple (don't forget how just making modems for ~ half of Apple's yearly iPhone shipments was causing them all sorts of problems) and that won't change until the Ohio fabs are online, and they have a cozy relationship with TSMC that will make them difficult to pry loose.

Intel has their work cut out for them trying to come within distant shouting distance of TSMC, but they can probably overtake Samsung without much difficulty.

In addition, I think that Intel is going to have to proved that 18A is a slam dunk for their own processors before anyone one would be willing to lay down big orders (at least in pre-production). If PPA is very good, and PDKs are well designed and well integrated into third party tools - they really could steal allot from Samsung Logic foundry business. So, the pressure is really on for Intel to deliver. Honestly, I hope they do.