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

[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.

 

[SiliconFly]

Agreed. It’s a mess. They have small Korean AI customers but that’s almost like a handout, might even be subsidized and it’s just on 4NM (SF4X). Similarly AMD is only doing one cheapo Zen 5 4c part on SF4X.

So after 4 years of playing around with 5 and 4nm, and really the 4nm only started in 2021, with the 5nm being sort of a fake 7nm shrink, they just might be good enough for a small, cheapo AMD part in 2025 on a derivative tweak of it and some dead end startups.


Also funny — Intel by contrast is on the upswing. Backside power delivery working, 18A should ship some with PTL, Nova Lake also on 18A (?), and a path to some extensions of their nodes for efficiency, performance, TSV cachet etc.

Only thing against them is cost and regulation in the West really.

Edit: well, upswing but quite volatile. I mean relative to Samsung I guess.

Panther Lake is the biggest beneficiary of all this. Should easily clock higher than its counterparts.

And Nova Lake isn't confirmed on 18A. Maybe a 18A variant, or even 14A as Intel is all in on High-NA and the timeline fits.

 

[igor_kavinski]

Panther Lake is the biggest beneficiary of all this. Should easily clock higher than its counterparts.

Intel 4 didn't really deliver on high clocks compared to Intel 7. What makes you think Intel 18A will be better on that front? It may just reduce power at the same clocks at best. 18A wafers also aren't expected to ramp up until 2026 so Panther Lake may have to wait for a refresh or whatever comes next may reap the true benefits of a refined Intel 18A.

 

[igor_kavinski]

Truck driver dies after ‘explosion’ at TSMC Arizona

Driver knocked 20ft by pressure release valve.

www.tomshardware.com

Wow. Sulphuric acid tank. That's a really bad way to go

 

[DavidC1]

Intel 4 didn't really deliver on high clocks compared to Intel 7. What makes you think Intel 18A will be better on that front?

It's not really an accurate assessment, because just like endless 14nm derivatives, their 10nm has been refined for many years. They are clocked to the stratosphere because they have little other options to compete. Trying to top that is bound to fail.

They've been beyond due aiming for another Conroe like architecture with substantially lower clocks and much better performance per clock.

We're at over 6GHz now, which the supposedly-fiery-furnace Prescott and even Tejas failed to reach. Should that itself not be an indicator that the architecture is the fault?

It may just reduce power at the same clocks at best.

Which is basically the point of a new process. And if they had an architecture that made it perform great at 4GHz, maybe 5GHz at max, then they'd have room(little) to scale clocks in the future too.

This tells me people are still too focused on clock speeds.

 

[igor_kavinski]

This tells me people are still too focused on clock speeds.

If Zen 5 hits the rumored 6 GHz even just for boost clocks, Intel needs every single MHz to prop up their coming architectures.

 

[DavidC1]

If Zen 5 hits the rumored 6 GHz even just for boost clocks, Intel needs every single MHz to prop up their coming architectures.

In the big picture, competition between duopolies are basically only perceived with both companies ending up with the similar product but different ways to get there.

The leading company pricing their final product line in accordance to the competitor further reinforces this. Once one company achieves leadership, they aim at optimizing revenue and lowering costs for increased margins.

Case in point - Zen 4 with low uarch gains but greater than expected clock gains, versus Golden Cove with bigger gains but with lower clock gains.

 

[Doug S]

Well he could hardly do worse than the last guy. The question is, can he do better?

https://www.cnbc.com/2024/05/21/sam...conductor-chief-as-ai-chip-race-heats-up.html

 

[FlameTail]

A look at TSMC's roadmap;

TSMC's Roadmap at a Glance: N3X, N2P, A16 Coming in 2025/2026

www.anandtech.com

A16 looks like a superb node. Virtue of Super Power Rail, I suppose.

That -20% power/+10% performance is vs N2P, not standard N2.

 

[Aapje]

According to the Semiwiki guy, N2 is much easier to design for than N3, which is another important factor. Supposedly, a bunch of customers plan to skip N3 and go straight to N2. I expect N16 to be even better in this regard, because splitting up power and messaging will make it much easier to route the internal IO. I do wonder how expensive backside power delivery will be to do. I fear another price increase per wafer.

 

[FlameTail]

According to the Semiwiki guy, N2 is much easier to design for than N3, which is another important factor. Supposedly, a bunch of customers plan to skip N3 and go straight to N2. I expect N16 to be even better in this regard, because splitting up power and messaging will make it much easier to route the internal IO. I do wonder how expensive backside power delivery will be to do. I fear another price increase per wafer.

Could Nvidia be one of those who are skipping N3 and going directly to N2?

Their RTX 50xz generation supposedly uses the "4NP" node, which belongs to the N5 family. Considering how the timing works out, by the time RTX 60xx lands, it could be using N2 or A16.

 

[Aapje]

I expect so, since N2 should be available for the 6000-series in late 2025.

 

[oak8292]

I expect so, since N2 should be available for the 6000-series in late 2025.

Nvidia has the cash flow to reserve capacity but the question will be yield. If they redesign for multi-die products they might be able to adopt a node earlier. However if they still want 700-800 mm2 die then they will probably still trail.

 

[Ajay]

Nvidia has the cash flow to reserve capacity but the question will be yield. If they redesign for multi-die products they might be able to adopt a node earlier. However if they still want 700-800 mm2 die then they will probably still trail.

For their AI/ML/HPC GPUs NV will almost certainly be using both large dies and multichip.

 

[Doug S]

That -20% power/+10% performance is vs N2P, not standard N2.

That's pretty much what you'd expect from BSPDN/SPR.

I wouldn't be at all surprised if A16 is very very similar if not identical to N2P other than "super power rail". N2P & A16 both entering mass production at the same time is a pretty big clue on that front.

 

[FlameTail]

That's pretty much what you'd expect from BSPDN/SPR.

I wouldn't be at all surprised if A16 is very very similar if not identical to N2P other than "super power rail". N2P & A16

A16 also has a 10% density improvement, coming from SPR.

both entering mass production at the same time is a pretty big clue on that front.

And that's interesting. Customers have the choice of either picking N2 or A16. Who will pick which, and why?

 

[Doug S]

A16 also has a 10% density improvement, coming from SPR.

And that's interesting. Customers have the choice of either picking N2 or A16. Who will pick which, and why?

They'll pick the one they can get - the one that Apple isn't taking the first 4-5 months worth of production on.

 

[FlameTail]

Samsung loses one more customer;

Exclusive: Google Pixel 10's Tensor G5 chip will be manufactured by TSMC, and we can prove it

The Google Tensor G5, the chip in the Google Pixel 10, will be manufactured by TSMC, and we have the proof.

www.androidauthority.com

After 4 generations of Tensor SoCs being manufactured on Samsung's nodes, Google will be moving to TSMC for Tensor G5 (which will be utilised in the Pixel 10 series of phones, expected to launch in 2025 October).

 

[Curious_Inquirer]

Samsung loses one more customer;

Exclusive: Google Pixel 10's Tensor G5 chip will be manufactured by TSMC, and we can prove it

The Google Tensor G5, the chip in the Google Pixel 10, will be manufactured by TSMC, and we have the proof.

www.androidauthority.com

After 4 generations of Tensor SoCs being manufactured on Samsung's nodes, Google will be moving to TSMC for Tensor G5 (which will be utilised in the Pixel 10 series of phones, expected to launch in 2025 October).

With Qualcomm also shifting all orders to TSMC I don’t really see a future for Samsung Foundry. It really looks like TSMC is getting all the orders from everyone. With Qualcomm gone they only have Tesla and Ambarella as major customers left.

https://twitter.com/x/status/1793447610055012368

 

[SpudLobby]

Samsung loses one more customer;

Exclusive: Google Pixel 10's Tensor G5 chip will be manufactured by TSMC, and we can prove it

The Google Tensor G5, the chip in the Google Pixel 10, will be manufactured by TSMC, and we have the proof.

www.androidauthority.com

After 4 generations of Tensor SoCs being manufactured on Samsung's nodes, Google will be moving to TSMC for Tensor G5 (which will be utilised in the Pixel 10 series of phones, expected to launch in 2025 October).

See @Doug S

As I said and it looks like the rumors are coming true.

This is honestly comical but also terrible.

 

[SpudLobby]

With Qualcomm also shifting all orders to TSMC I don’t really see a future for Samsung Foundry. It really looks like TSMC is getting all the orders from everyone. With Qualcomm gone they only have Tesla and Ambarella as major customers left.

https://twitter.com/x/status/1793447610055012368

View attachment 99719

Actually, it’s not even clear Tesla is secure, most likely they will follow suit and leave Samsung too for their FSD SoCs.

Tesla Moves to TSMC with Autonomous Chips

It is reported that Tesla can lead the industry in building fully self-driving electric vehicles, and TSMC is a key player. It manufactures new FSD chips for Tesla, making the "brain" of electric vehicles smarter. TSMC has never commented on single customers and market rumors. However, at the...

semiwiki.com

 

[moinmoin]

Samsung will still be doing its memory business with its foundries. That's how it started and likely also how it ends with this kind of management.

 

[LightningZ71]

I wonder if AMD will follow through with their trailing edge chips that were rumired to be moving to Samsung foundaries. They should be getting stellar pricing given Samsung's plummeting demand situation.

 

[dr1337]

I wonder if AMD will follow through with their trailing edge chips that were rumired to be moving to Samsung foundaries. They should be getting stellar pricing given Samsung's plummeting demand situation.

At this rate it really feels like AMD is going to choke on volume if they don't start using other foundries for key products. Otherwise they can never improve market share if they never make more chips, which will end their good stock prices IMO.

But also maybe they don't care and want to go apple style and keep their low market shares, would be pretty disappointing if so.

 

[DrMrLordX]

At this rate it really feels like AMD is going to choke on volume if they don't start using other foundries for key products.

As Intel demonstrated, TSMC builds capacity to meet the needs of customers if the customer is willing to pay. And AMD may not even need to pay everything upfront the way Intel did due to AMD's relationship with TSMC.

 

[Aapje]

At this rate it really feels like AMD is going to choke on volume if they don't start using other foundries for key products. Otherwise they can never improve market share if they never make more chips, which will end their good stock prices IMO.

But also maybe they don't care and want to go apple style and keep their low market shares, would be pretty disappointing if so.

What a strange take. TSMC will make everything that AMD needs, as long as they order it and are willing to pay the price.

TSMC and AMD collaborate very closely, like with the X3D-chips, so it's harder for AMD to switch than Nvidia.