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.

 

[Hougy]

Are you really asking that? There are countries in South America, Bana Republics that can't sell their Bananas to Russia. If they do they will be Cut off from the established World Order from Which the USA and Co(NATO/UK/EU) Rule with an Iron Fist.


A single $1 Banana can't be sold to Russia without the USA Consent. Let alone a few Billion Dollars worth of State Of The Art Equipment that would bring technology parity to China.

Politically, it's very different. You can do that with Russia because you can say they are evil since they invaded a country and committed atrocious war crimes.

What was the justification for not allowing China to have the same access to the same tech as the rest of the world, to compete on equal footing? The US doesn't give the impression of being the guys here.

What was the threat if ASML decided to continue selling to China? Did they promise to impose severe economic sanctions to all of Netherlands and make the other countries do the same? Couldn't ASML just have told told the US "If you want to shoot yourself in the foot and stop buying our valuable EUV equipment that's fine, your loss. There's plenty of demand and we can sell it to the rest of the world." ?

 

[gdansk]

I still don't understand how the US was able to forbid ASML from selling EUV equipment to China and TSMC from selling 7nm or better chips to Chinese companies. Could someone please explain me how this was possible? Why would these two follow the US' orders?

A significant portion of ASML is in the US. Notably EUV light sources. They literally cannot make EUV machines without US exports. They bought Cymer, after all. Plus ASML was already booked well beyond their capacity for EUV machines anyway so why fight about it.

As for TSMC, well, that's more political and probably goes back to the Hong Kong crackdown.

 

[A///]

A significant portion of ASML is in the US. Notably EUV light sources. They literally cannot make EUV machines without US exports. They bought Cymer, after all. Plus ASML was already booked well beyond their capacity for EUV machines anyway so why fight about it.

As for TSMC, well, that's more political and probably goes back to the Hong Kong crackdown.

It isn't just that. There are components in their machines with US origins. Components made by individual companies that may also have offices or HQs in the US. Patents pertaining to the US.

It's a mix-mash of how the US can order foreign companies to operate. From my understanding revolving sanctions against other countries that encompass more than tech, it refers to patents, office locations, component origins, and more. I tried following the sanctions against a certain axis of evil country a while back and got lost in the depth of the sanctions.

 

[Roland00Address]

2025 for N2? IFS might be stealing customers from TSMC after all. Wouldn’t it be nuts if Intel lured Apple away?

How would that work since part of Apple's magic is the TSMC various interconnects?

I am not saying it is impossible, no in fact the opposite. I am curious of all the details, for I have curiosity about things in the realm of magic and mystery. =)

 

[Roland00Address]

I still don't understand how the US was able to forbid ASML from selling EUV equipment to China and TSMC from selling 7nm or better chips to Chinese companies. Could someone please explain me how this was possible? Why would these two follow the US' orders?

Politically, it's very different. You can do that with Russia because you can say they are evil since they invaded a country and committed atrocious war crimes.

What was the justification for not allowing China to have the same access to the same tech as the rest of the world, to compete on equal footing? The US doesn't give the impression of being the guys here.

What was the threat if ASML decided to continue selling to China? Did they promise to impose severe economic sanctions to all of Netherlands and make the other countries do the same? Couldn't ASML just have told told the US "If you want to shoot yourself in the foot and stop buying our valuable EUV equipment that's fine, your loss. There's plenty of demand and we can sell it to the rest of the world." ?

As the OddLots podcast explained.

Wait let me explain what OddLots is...​

OddLots is a Bloomberg podcast that focus on Supply Chains and other aspects of the economy. They have several good silicon focus episodes with the foundries. I am talking about the one they did with ASML a few months ago located here. (nov 2021)​

​

ASML, the Obscure Powerhouse at the Cutting Edge of Chip Technology

<p>This year has brought fresh awareness to the complexity of the semiconductor supply chain. Taiwan Semiconductor, the big manufacturer, has become a household name. But there's another giant that hardly anyone outside of the chip industry has heard of. ASML is a Dutch company that's at the...

player.fm

​

As the OddLots podcast explained that the secret sauce for ASML is not them being better at the technology of making the parts to make the smaller chips, but instead being a nexus of suppliers for there are 10,000+ different companies parts inside an ASML EUV machine. Parts from all over the world including the US. ASML is not a technology company just by itself, just as important is being a supply chain company.

Since US law allows the US to put on an embargo on exports of technologies manufactured in the US to be natural security related, those parts can be shipped to ASML but with rules of who ASML can ship to. Thus an EUV machine can have a restricted places it can go, and the EUV machine is then sent to other places in the world including not the US, but can not go places such as China.

As a bad science fiction novel published in the 60s, from a publisher known for their Auto Manuals and not science fiction.​

• The power to destroy a thing is the absolute control over it.
  • Paul-Muad'Dib to the Guild navigators, at his confrontation with the Emperor Shaddam IV.
• The thing the ecologically illiterate don't realize about an ecosystem is that it's a system. A system! A system maintains a certain fluid stability that can be destroyed by a misstep in just one niche. A system has order, a flowing from point to point. If something dams the flow, order collapses. The untrained might miss that collapse until it was too late. That's why the highest function of ecology is the understanding of consequences.
  • Pardot Kynes in "Appendix I: The Ecology of Dune"

It may stink, it may offend your moral sentiments, but this is not new and we can talk about the past of the 1800s where technology was stolen and brought to the US (Francis Cabot Lowell and others) and now there is no reciprocality, but that is kind of beyond this forum isn't it? 🙂

 

[jpiniero]

Maybe they changed enough, maybe they didn't. Someone was lying somewhere at Intel before. Either the lying was knowingly from the top, or they were being lied to from below and believed the lies. If it was the former, then switching the guy at the top will fix that. If it was the latter, it doesn't fix it unless the people who were doing the lying have also been replaced.

I'm sure upper management was well aware that things with 10 nm were not going well; even though they weren't honest with the public about it. I'd expect sunshine & rainbows from Pat regardless of what's actually going on.

 

[Doug S]

I'm sure upper management was well aware that things with 10 nm were not going well; even though they weren't honest with the public about it. I'd expect sunshine & rainbows from Pat regardless of what's actually going on.

One would hope they were aware but I wouldn't be sure of it. I've personally witnessed line managers blowing sunshine up the ass of their bosses, and people in charge not knowing what was going on. Since I was a consultant it wasn't my problem so I ignored it.

Seriously considered picking up some put options on the stock after that gig ended, but like Buffett's famous quote about market irrationality I figured lying to your bosses could likewise continue for longer than the expiry date on my put contracts (plus I'd have to consult an attorney to see if that would have constituted insider trading...)

 

[DrMrLordX]

It's hard to believe that Intel could ship 18a anything by 2024. It's 2022 and they still aren't shipping 7nm/Intel 4. They probably won't have anything on Intel 4 by the end of 2022, except maybe Loihi 2 to select customers. Meteor Lake? They haven't even launched Raptor Lake yet, which is just another 10ESF/Intel 7 product.

Intel must iterate through Intel 4, Intel 3, and Intel 20a before hitting 18a. They are going to do all that in two years? No. Absolutely not. Not unless IBM is doing all the work for them. And even then I have my doubts!

 

[igor_kavinski]

It's hard to believe that Intel could ship 18a anything by 2024.

Mark your calendar for 31st December, 2024. That's when Intel will launch and sell their 18A mobile Core i3 in China. Promise fulfilled.

 

[DrMrLordX]

Mark your calendar for 31st December, 2024. That's when Intel will launch and sell their 18A mobile Core i3 in China. Promise fulfilled.

Funny, but I don't see them doing that even. SRAM test cells maybe . . .

 

[igor_kavinski]

Why haven't they fired Sohail Ahmed yet? - post regarding Intel Corp. layoffs (thelayoff.com)

Sohail Invented TMGs hire cheap H1B's to cut cost model. PhD requirement was set to eliminate US born candidates, we clearly don't need PhDs to sustain tools.
Turns out the good H1Bs wont come to Intel because of its reputation as a sweat shop, so they've just been hiring from third tier universities with graduates from unrelated majors.
The H1Bs are only biding their time till they get their papers in order. Ofcourse Intel has lost its manufacturing lead.
Sohail implemented this model with tacit support from Paul and BK. Time for him to take the responsibility and leave.

Is there any truth to this? This makes me wary of Raja Koduri even more. The solution to every problem is not an H1B visa.

 

[igor_kavinski]

Can Intel Catch TSMC in 2025? - SemiWiki

So all depends on Intel making 4nm possible by end of 2022.

 

[DrMrLordX]

This makes me wary of Raja Koduri even more. The solution to every problem is not an H1B visa.

1). Raja has nothing to do with Intel's fabs, at least not directly.
2). Eh. Really?

 

[uzzi38]

https://investor.tsmc.com/english/encrypt/files/encrypt_file/reports/2022-04/f127e5c5d893581fc65f3c7a8d223f0f689663da/TSMC%201Q22%20Transcript.pdf

TSMC says N3 yield is good and they are sticking to the plan where they start the ramp at the end of the year. N3E is scheduled a year later but could get pulled in.

Alas they are saying 2025 for N2 ramp, maybe even 2H.

Yield might be good but it doesn't mean anything if the cost per wafer is still beyond ridiculous.

Base N3 is still a dud, and that's not changing any time soon. With N2 being as late as it is, I certainly hope TSMC are able to improve on N3E even further as an alternative for it's customers in 2025.

 

[jpiniero]

Yield might be good but it doesn't mean anything if the cost per wafer is still beyond ridiculous.

TSMC claims they have multiple tapeouts for N3 and a "high level of customer engagement" on it. The CPW is I'm sure outrageous but I guess there's plenty who will pay it.

 

[uzzi38]

TSMC claims they have multiple tapeouts for N3 and a "high level of customer engagement" on it. The CPW is I'm sure outrageous but I guess there's plenty who will pay it.

And you'll see some big names who aren't too, I guarantee you that.

 

[DisEnchantment]

Base N3 is still a dud, and that's not changing any time soon.

Considering the results achieved by their competition, calling it a dud is a bit of a stretch, don't you think?

27% more efficient, and 1.6x logic scaling based on latest OIP Ecosystem Forum numbers, granted performance is not so great, but for the bulk of ARM SoCs it is not like they can clock any higher even if process allows.
Another thing too, one of the main focus for N3 was on HPC as per TSMC's material. >1.5x scaling for HPC cells (Compared to 1.38x for N5).
Samsung's 7LPP-->5LPE-->5LPP-->4LPE-->4LPP-->3GAE progression is just edging >1.6x logic scaling.

I think the folks doing chiplet SoC architecture will be fine.
IO already partitioned (e.g. AMD's IOD), SRAM was already partitioned (e.g. AMD's 3D V-Cache ) and SRAM is traditionally energy efficient.
Keep all the heavy logic on the most efficient and performant node available.

And you'll see some big names who aren't too, I guarantee you that.

TSMC's statement mentioned >2x NTOs for N3 already.

 

[Ajay]

TSMC's statement mentioned >2x NTOs for N3 already.
View attachment 60157

What is an NTO?

 

[DisEnchantment]

What is an NTO?

It is New Tape Out, fresh physical design/implementation of the chip using the target process PDK and libraries and the usual stuffs like parasitic extraction, design rule checks, layout etc.
The output of NTO is GDSII data sent to TSMC who prints the masks and optimize them with the designer.

 

[NTMBK]

Why haven't they fired Sohail Ahmed yet? - post regarding Intel Corp. layoffs (thelayoff.com)



Is there any truth to this? This makes me wary of Raja Koduri even more. The solution to every problem is not an H1B visa.

That forum is full of disgruntled ex-employees with an axe to grind. Of course they blame everything on immigrants.

 

[Ajay]

That forum is full of disgruntled ex-employees with an axe to grind. Of course they blame everything on immigrants.

Well, used to be full of. We've lost so many AT members, it’s a bit sad. It used to be almost impossible to keep up with a thread like this from all the posts made each day.

 

[Thunder 57]

That forum is full of disgruntled ex-employees with an axe to grind. Of course they blame everything on immigrants.

H1B's are bad.

 

[Doug S]

Can Intel Catch TSMC in 2025? - SemiWiki

So all depends on Intel making 4nm possible by end of 2022.

Is he getting those "performance improvement" numbers? I assume these are Intel claims? It seems ludicrous to expect Intel will get 20% 7nm -> 4nm, 18% 4nm -> 3nm (and with only a 1.09x density increase??), 15% 3nm -> 20A, and 10% 20A -> 18A (this with only a 1.06x density increase?)

Now obviously these are like TSMC's numbers, you get x% performance improvement at constant power or y% power reduction at constant performance. And "constant power" roughly depends on making the same design - i.e. you can't have a chip with bigger cores and caches and 50% more transistors overall in the smaller process and expect that will be "constant power".

Still, color me highly skeptical that if Intel does deliver something they will decide to call "18A" by the end of 2024 that it will be that big of a performance jump over what they have today.

 

[uclaLabrat]

18A means each "feature" is going to be less than 10 silicon atoms in size. Thats patently ludicrous.

 

[Doug S]

18A means each "feature" is going to be less than 10 silicon atoms in size. Thats patently ludicrous.

No it doesn't, node names stopped having anything to do with physical dimensions of a transistor long ago.