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]

I thought they used some sort of ARM SoC? A72 based in the past, unless they've updated it to something more recent. Dunno the size. The SRAM is for early work well before risk production begins, and not used for risk/production go live decisions.

For N2 they used SRAM iirc but the thing used for risk production is way smaller than for MP

 

[Win2012R2]

No one gets this data except for customers so keep dreaming

One does not need access to actual data to be able to deduce correctly from behavior of those who had it.

Depends how you define profit.

It's already defined by relevant accounting standards, changing that definition is a sure way to jail for execs, check out Sarbanes-Oxley.

They can still release a product at 20% yield at a profit... but not much profit.

No, they can't, especially now with Foundry having it's own P&L that they can't hide easily.

 

[511]

One does not need access to actual data to be able to deduce correctly from behavior of those who had it.

LoL yeah sure you can't judge how one can behave with the data

 

[Win2012R2]

LoL yeah sure you can't judge how one can behave with the data

Yes you can when it's all or nothing kind of event: no serious customer signed up for it and I am 100% sure all serious customers had the data: we'd know from TSMC reporting if somebody big jumped the ship.

 

[jpiniero]

Yes you can when it's all or nothing kind of event: no serious customer signed up for it and I am 100% sure all serious customers had the data: we'd know from TSMC reporting if somebody big jumped the ship.

There isn't much 18A capacity. So even if there was real interest from third parties, it'd have to be something that wasn't that big.

 

[Win2012R2]

There isn't much 18A capacity. So even if there was real interest from third parties, it'd have to be something that wasn't that big.

But if it's a great "on-track" process why not there isn't much capacity? This is contrary to official Intel's Fab plans to have big customers to share costs.

Occam's razor says this is because 18A is bust, I mean how much clearer that can be when we know that Nova Lake will use N2.

 

[511]

There isn't much 18A capacity. So even if there was real interest from third parties, it'd have to be something that wasn't that big.

Well one Arizona Fab is at least 20K WSPM for 18A reference TSMC N2 Capacity will be 60K WSPM next year.

Yes you can when it's all or nothing kind of event: no serious customer signed up for it and I am 100% sure all serious customers had the data: we'd know from TSMC reporting if somebody big jumped the ship.

Just cause no one signed that doesn't mean it's bad for Intel lol external vs Intel Internal use is quite the difference.

 

[Win2012R2]

Just cause no one signed that doesn't mean it's bad for Intel

It's terrible for Intel actually because it means they can't sign up anybody serious for 14A, which is evident from their own disclosures.

Well one Arizona Fab is at least 20K WSPM for 18A reference TSMC N2 Capacity will be 60K WSPM next year.

This needs to be adjusted by actual yields, which in case of Intel clearly not good enough to say the least.

 

[511]

It's terrible for Intel actually because it means they can't sign up anybody serious for 14A, which is evident from their own disclosures.

14A is in evaluation phase for at least this year and the next and it won't have the issue of having not good enough PDK and has FinFlex aka NanoFlex TurboCell.

This needs to be adjusted by actual yields, which in case of Intel clearly not good enough to say the least.

Keeps dreaming about 18A yield issues when there is none 😛

 

[Win2012R2]

Keeps dreaming about 18A yield issues when there is none

If there are no issues then why Intel uses N2 for Nova Lake?

it won't have the issue of having not good enough PDK

How can you possibly know???

 

[511]

If there are no issues then why Intel uses N2 for Nova Lake?

Capacity/Cost? It's 3 freaking N2 tiles and people are acting like 18A has issues. You forget that by NVL Clearwater Forest/PTL/WCL will be out and DMR will be out as well.

How can you possibly know???

Cause everyone has said it it's built in collaboration with Foundry Customers.

 

[Win2012R2]

Cause everyone has said it it's built in collaboration with Foundry Customers.

There are no 14A customers - Intel says so in their filing, and they warn if they don't find those customers then might just not bother investing into 14A, but PDK is of course perfect.

Capacity/Cost? It's 3 freaking N2 tiles and people are acting like 18A has issues. You forget that by NVL Clearwater Forest/PTL/WCL will be out and DMR will be out as well.

And why would not Intel have capacity for it? Cost should be better for Intel since it's inhouse - saving on 53%+ gross margin TSMC wants. So what's the problem?

I reckon the problem is clocks - server parts can get away with lower clocks, but for client 18A just can't compete, clearly either parametric yields are rubbish (most likely) or just using way too much power (unlikely since that did not stop Intel before).

Thing is the whole thing does not make sense - paying $$$ to competitor to make lower margin client parts, only reason is they can't clock competitively 18A they have now and know it won't happen even in a year.

 

[511]

There are no 14A customers - Intel says so in their filing, and they warn if they don't find those customers then might just not bother investing into 14A, but PDK is of course perfect.

Believe what you will 14A is being designed with input from industry

And why would not Intel have capacity for it? Cost should be better for Intel since it's inhouse - saving on 53%+ gross margin TSMC wants. So what's the problem?

Cost should be better in house the must be compelled to use N2 for other reasons

I reckon the problem is clocks - server parts can get away with lower clocks, but for client 18A just can't compete, clearly either parametric yields are rubbish (most likely) or just using way too much power (unlikely since that did not stop Intel before).

Nope and Nope to both Raichu leaded Parametric Yield already

Thing is the whole thing does not make sense - paying $$$ to competitor to make lower margin client parts, only reason is they can't clock competitively 18A they have now and know it won't happen even in a year.

oh please they are just using 1-2 Tile in each SKU From TSMC they have the room to adjust it according to their needs. If you think you know more about what to source from and from where you are free to apply at Intel.

 

[DZero]

There are no 14A customers - Intel says so in their filing, and they warn if they don't find those customers then might just not bother investing into 14A, but PDK is of course perfect.

Funny story, if the Chinese weren't kicked out the industry like that (now depends on TSMC but is slowly going away), by now, they would be the clients of Intel at least from Allwinner or Rockchip. Heck, even UNISOC would jumped with Intel.

 

[marees]

Funny story, if the Chinese weren't kicked out the industry like that (now depends on TSMC but is slowly going away), by now, they would be the clients of Intel at least from Allwinner or Rockchip. Heck, even UNISOC would jumped with Intel.

The Chinese will now go to Korean design companies & fabs

 

[511]

The Chinese will now go to Korean design companies & fabs

and US will block them xDd

 

[DZero]

and US will block them xDd

China don't need them anymore. US made China to boost themselves...

They should make India or Vietnam produce in house processors if they want clients now.

 

[mpumalanga]

Believe what you will 14A is being designed with input from industry

Cost should be better in house the must be compelled to use N2 for other reasons

Nope and Nope to both Raichu leaded Parametric Yield already

oh please they are just using 1-2 Tile in each SKU From TSMC they have the room to adjust it according to their needs. If you think you know more about what to source from and from where you are free to apply at Intel.
View attachment 129387

You mean the difficult part they leave to TSMC, because Intel can't do it themselves.

 

[511]

You mean the difficult part they leave to TSMC, because Intel can't do it themselves.

what difficult part ?

 

[regen1]

Believe what you will 14A is being designed with input from industry

Yeah, this was already the case with IDM2.0 projections where 14A was to be more suitable for external customers from the ground up than 18A family but they might be making even more changes to it, result of which has to seen.
Even for 18A family, 18AP was the one projected as more complete for external customer usage relative to vanilla 18A.


This was June 2023 projection. It's pretty clear initial products were internal.
All the foundry execution remains to be seen. Many of the foundry guys have been laid off and some have left.

oh please they are just using 1-2 Tile in each SKU From TSMC they have the room to adjust it according to their needs. If you think you know more about what to source from and from where you are free to apply at Intel.
View attachment 129387

The decision to move Novalake-H(4P+8E) compute tiles was probably taken under interim Product CEO, can't say for certain. That tile is very significant in terms of area and the volume of shipments. If 18AP turns out to be good enough, this might be one of the worst decisions. It's margin, foundry and PR disaster. Capacity can not be an excuse here.
It's only good in case 18AP is quite significantly worse than N2P in performance.

 

[511]

The decision to move Novalake-H(4P+8E) compute tiles was probably taken under interim Product CEO, can't say for certain.

It was taken by products and the product VP sucks imo at Intel no relevant Engineering Experience

 

[DrMrLordX]

It's terrible for Intel actually because it means they can't sign up anybody serious for 14A, which is evident from their own disclosures.

Ehh. We haven't had any smoking gun moments like when Broadcom publicly declared 18A to be inadequate for their needs. 14A is still too far off for us to know anything about it.

China don't need them anymore. US made China to boost themselves...

SMIC can't sub in for TSMC yet, and possibly never will. There's only so much they can do.

 

[LightningZ71]

I don't think that people fathom the amount of resources that China is throwing at their foundries (and gas turbine blade technology). While they certainly trail the leading edge at present, they are working every possible angle at advancement and won't be behind in what they are able to produce for more than three or four more years.

 

[DrMrLordX]

I don't think that people fathom the amount of resources that China is throwing at their foundries (and gas turbine blade technology). While they certainly trail the leading edge at present, they are working every possible angle at advancement and won't be behind in what they are able to produce for more than three or four more years.

The money won't matter if they can't get EUV up and running.

 

[LightningZ71]

I have no doubts that they will get there. They may go a different route than ASML took, but they'll get there.