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]

TSMC is spending $40B+ on N2. A14 will presumably be, what, 60, 70 billion?

Since costs are higher in the US (not to mention Intel's incompetence), it'd cost Intel even more to realistically compete.

That is why they are canning 14A if customer don't want a second supplier intel is done doing this stuff

 

[511]

Anyone knows, are TSMC price hikes only for future contracts or do they apply to existing contracts too? I guess some customers like Intel, AMD, Apple already have quite some capacity booked, so these hikes should not affect them as much as some others.

Definitely newer order you can't increase price on someone already sold

 

[jpiniero]

That is why they are canning 14A if customer don't want a second supplier intel is done doing this stuff

TSMC's biggest competitor at this point is their older nodes.

And it's not without question that node progress ends, not because of technical feasibility but because it doesn't make sense cost wise.

 

[adroc_thurston]

And it's not without question that node progress ends, not because of technical feasibility but because it doesn't make sense cost wise

It'll always make sense cost-wise in segments less sensitive to raw BOM.

 

[511]

TSMC's biggest competitor at this point is their older nodes.

Not really they can adjust everything accordingly

And it's not without question that node progress ends, not because of technical feasibility but because it doesn't make sense cost wise.

i doubt this
agree with adroc

 

[Doug S]

Anyone knows, are TSMC price hikes only for future contracts or do they apply to existing contracts too? I guess some customers like Intel, AMD, Apple already have quite some capacity booked, so these hikes should not affect them as much as some others.

If you merely have it booked the price is very likely subject to increase. If you have prepayed far in advance as Apple likes to do that's a different matter. Apple likely has at least one year and perhaps more N2 capacity already locked in.

Obviously "it depends" on the actual language but covid chaos and now tariff chaos cause all sorts of real and potential supply chain issues that could increase their costs between the time orders are booked and wafers are finished. TSMC has no real competition, so contracts will be written to favor them, not the customer. In economic terms, TSMC is the price setter, and their customers are the price takers. You can complain about the price of the capacity you booked going up, but TSMC will say "there is a line of customers waiting behind you who would be happy to move up in line".

 

[LightningZ71]

Rapidus really needs to get moving...

 

[Khato]

Yeah, can certainly hope that Rapidus will fare better than Intel Foundry and Samsung. I'm not particularly optimistic, but it could happen.

 

[511]

Yeah, can certainly hope that Rapidus will fare better than Intel Foundry and Samsung. I'm not particularly optimistic, but it could happen.

Rapidus is based on IBM IP the worse you can have even worse than Samsung if I Will

 

[LightningZ71]

Perhaps it's not the IP that's sub-par, but the crew that's been attempting to implement it?

 

[511]

Perhaps it's not the IP that's sub-par, but the crew that's been attempting to implement it?

I don't have high hopes seeing GF and Samsung as well Samsung was using IBM IP as well look at them can't yield.
IP Tier List:
S -> TSMC
A-> Intel
B-> Samsung
F-> IBM(Haven't seen IBM IP in HVM since a long time.

 

[LightningZ71]

I thought that Samsung, and by inheritance GF, used IBM IP on at least some nodes?

 

[511]

I thought that Samsung, and by inheritance GF, used IBM IP on at least some nodes?

They used IBM IP but GF Bailed out after 14nm and Samsung used IBM IP for quite a while but they are not using it with SF2

 

[DavidC1]

Perhaps it's not the IP that's sub-par, but the crew that's been attempting to implement it?

Micron called DRAM the freight train that runs over all potential DRAM-killer technologies. And they highlighted it's the enormous amount of manpower, R&D and other resources going in that made it so far irreplaceable.

TSMC is the DRAM of the process world. It's so established and has such a strong foothold that it will not be easy until they have a Intel-like fumble, and even then they'll maintain it a lot for a long time to come. Just like Intel still does.

Sometimes, we don't change the tools we use even if better alternatives are out there simply because we are lazy, or don't care, or it's a hassle. The advantages have to be significant to matter. Intel, Samsung, Rapidus all has that uphill to climb. And whatever little customers they gain they have to absolutely make sure not to lose any of them. When Intel had the fake IDM with the 1.0, they should have kept tiny customers like Altera and Achronix. They lost them, because they were never serious. And they fumbled of course. Technical reasons are nowhere near enough.

 

[DrMrLordX]

People would be more than happy to use Samsung if they could get their yields up. NV has used them repeatedly in the past, and AMD was close to doing so as well. Rapidus is an unknown and Intel still doesn't have a handle on how to attract customers. Intel 3 is a decent enough node, but could they offer anyone significant volume? Probably not.

 

[511]

People would be more than happy to use Samsung if they could get their yields up. NV has used them repeatedly in the past, and AMD was close to doing so as well. Rapidus is an unknown and Intel still doesn't have a handle on how to attract customers. Intel 3 is a decent enough node, but could they offer anyone significant volume? Probably not.

Intel 3 in Ireland capacity is nice for 1 more customer but that's about it I don't think it can sustain more than one. FAB 52/62 are different game though

 

[Doug S]

When Intel had the fake IDM with the 1.0, they should have kept tiny customers like Altera and Achronix. They lost them, because they were never serious. And they fumbled of course. Technical reasons are nowhere near enough.

They didn't have a choice. They were capacity limited because they couldn't get 10nm working. They might have had bigger customers they had to turn away, we'll probably never know.

Their real problem was that they let the x86 side make the rules for the foundry side. Which meant using the same internal design tools instead of industry standard which turned off customers other than those with simple designs like FPGAs. Even if they had delivered 10nm on time and maintained a process lead over TSMC they would have major difficulties winning any ASIC/SoC type customers because of that, and because the x86 side was given priority in the fabs for running a bunch of hot lots because of the way they did their design work.

I think the latter is why they've had some pain on the design side the last few years. Not only did they have to make the transition to industry standard tooling, and to using a third party foundry, they've also learned the hard way about the need to improve the initial quality of their designs because doing a bunch of steppings isn't so easy when you don't control the fab.

 

[DZero]

I don't have high hopes seeing GF and Samsung as well Samsung was using IBM IP as well look at them can't yield.
IP Tier List:
S -> TSMC
A-> Intel
B-> Samsung
F-> IBM(Haven't seen IBM IP in HVM since a long time.

Seeing that tierlist... where is SMIC?

 

[511]

Seeing that tierlist... where is SMIC?

I would put it below Samsung tbh but that is due to lack of equipment and restrictions what would happen when both of those are lifted.

 

[DavidC1]

They didn't have a choice. They were capacity limited because they couldn't get 10nm working. They might have had bigger customers they had to turn away, we'll probably never know.

Their real problem was that they let the x86 side make the rules for the foundry side. Which meant using the same internal design tools instead of industry standard which turned off customers other than those with simple designs like FPGAs. Even if they had delivered 10nm on time and maintained a process lead over TSMC they would have major difficulties winning any ASIC/SoC type customers because of that, and because the x86 side was given priority in the fabs for running a bunch of hot lots because of the way they did their design work.

I think the latter is why they've had some pain on the design side the last few years. Not only did they have to make the transition to industry standard tooling, and to using a third party foundry, they've also learned the hard way about the need to improve the initial quality of their designs because doing a bunch of steppings isn't so easy when you don't control the fab.

That's what I meant by fake IDM. Gelsinger called it IDM 2.0 for a reason, because they claimed they wanted 3rd party vendors long ago. But just like their initial mobile efforts were only a token gesture, so was their foundry effort. It took many tries for them to get it right. But it really was in those days they should have went full in with the strategy, because it's a bit late now.

 

[DZero]

I would put it below Samsung tbh but that is due to lack of equipment and restrictions what would happen when both of those are lifted.

BTW... SMIC would be better than GloFo in that tierlist?

 

[LightningZ71]

Considering that GF hasn't made it past gf12+ with (finfets, thanks spell check) fingers, yes. SMIC is in the 6nm range, full rate 7nm, "6" nm enhancements, and volume with roughly 5nm multi pattern DUV nodes. I suspect that at their equivalent 5nm nodes, they are yielding better than Samsung and have more volume than Intel does with Intel7.

 

[Saylick]

Intel is making more management changes, including the departure of MJ Holthaus:

Intel Announces Key Leadership Appointments to Accelerate Innovation and Strengthen Execution

www.intc.com

 

[Joe NYC]

the departure of MJ Holthaus:

Which shocked no one. It's kind of interesting that LBT just pulled the plug instead of creating a fake advisory position for her.

Edit: just read the PR statement, and in fact they did create a fake advisory position for her...

 

[igor_kavinski]

in fact they did create a fake advisory position for her...

Can't have her feeling that she has been specifically mistreated.

Just some lunch money while she finds another company to drown.