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]

So, we are still not certain why it would cost them a ton of money.

Ramping aka running wafer on a new fab costs money they need to have ARL sold for few years to make a profit out of that ramp.

 

[adamge]

It boggles the mind to consider a scenario where Intel, knowing that 20A was a massive investment and the fruits of massive new technology development, would not have a strong cache of product designs to manufacture on it.

How broken is Intel that when they finally get their new process online, they don't have products to fill the production line. Instead they ice the process and start working on the next one.

 

[511]

It boggles the mind to consider a scenario where Intel, knowing that 20A was a massive investment and the fruits of massive new technology development, would not have a strong cache of product designs to manufacture on it.

How broken is Intel that when they finally get their new process online, they don't have products to fill the production line. Instead they ice the process and start working on the next one.

Blame swan lol he simply decided to outsource almost everything and didn't fund the fabs so once the roadmap is set it is difficult to change it.
Even the broken 10nm node had a single sku.
20A is not launching anyways.

 

[igor_kavinski]

Blame swan lol he simply decided to outsource almost everything and didn't fund the fabs so once the roadmap is set it is difficult to change it.

It's not just one person to blame. It's like they forgot how to make sane decisions. If your CEO says do something stupid, there needs to be pushback from the employees. It's almost like they hated the CEO (be it Krzanich or Swan or Gelsinger) and wanted them to fail and just did as they were told, without giving any counterfeedback to help the CEOs make informed decisions. Almost like everyone was fed up with the amount of work they had to do and the lousy pay they were getting so almost everyone decided in unison, they pay the CEO such a handsome package, let him do the thinking and the strategic planning. We are just drones and we gonna do as told, even if every cell in our body cries otherwise.

 

[511]

It's not just one person to blame. It's like they forgot how to make sane decisions. If your CEO says do something stupid, there needs to be pushback from the employees. It's almost like they hated the CEO (be it Krzanich or Swan or Gelsinger) and wanted them to fail and just did as they were told, without giving any counterfeedback to help the CEOs make informed decisions. Almost like everyone was fed up with the amount of work they had to do and the lousy pay they were getting so almost everyone decided in unison, they pay the CEO such a handsome package, let him do the thinking and the strategic planning. We are just drones and we gonna do as told, even if every cell in our body cries otherwise.

List of all good decision made by last 3 Ex -CEO
Kranzich:
None
Swan:
Outsourcing to TSMC was a good decision in his position.
Pat:
Change node dev methodology aka use minimalist node and add features to it later this made it to 5N4Y(I4/20A were minimalist node)
Can businesses that are not profitable
Do not allow infinite steppings and skip pre silicone Validation
Separation between Foundry/design

Bad decision:
Kranzich:
His existence as a Intel CEO was a bad decision and every decision he made
Swan:
Handed over something he didn't have experience with so he thought of getting rid of problem aka the foundry.
Pat:
Overbuilding Foundry shells
Missing AI GPUs.

 

[igor_kavinski]

Pat:
Overbuilding Foundry shells
Missing AI GPUs.

Pretty sure he made those decisions based on the lies he was told (oh everything's going great! Just greenlight this new process and we will have technology leadership in no time!). Pat should've ratted out the complete list of employees to the board who helped him make the bad decisions but he was a religious fool who didn't even think of getting revenge.

 

[oak8292]

Pretty sure he made those decisions based on the lies he was told (oh everything's going great! Just greenlight this new process and we will have technology leadership in no time!). Pat should've ratted out the complete list of employees to the board who helped him make the bad decisions but he was a religious fool who didn't even think of getting revenge.

Pat is old school Intel and based on ‘old school Intel’ process leadership is everything, the essence of ‘Moore’s Law’. Pat hired a lot of people to get process leadership and as hail Marys go it needs volume. I would say it was all Pat. To come in after process development failed epically on 10 nm and the first EUV process 7/4 nm and then claim you are going to do 5 nodes in four years which is epically expensive isn’t somebody below you telling you they can do it. That is Intel CEO level brashness.

Also Swan oversaw the roll out, or lack of roll out of the first EUV process and the inability to contractually deliver Aurora. There hadn’t been wafer volume growth for years and Fab 34 in Ireland and fab 42 in Arizona would probably be enough capacity for the decreasing ‘economic’ volume. Swan did not have the Intel vision of ‘build it and they will come’. He didn’t swing for the fence.

It is interesting the sideline quarterbacking. Swan, Krzanich and Pat each had their own styles. The fact that the ‘bounty’ of semiconductors at Intel is not keeping pace with the ‘industry’ probably has a lot to do with innovators dilemma. Intel had a market they did not want to disrupt with processors that sold for a lot less than $50. That was essentially where they lost the game and TSMC took over in wafer volume growth. If Intel had a process that was compatible with $10 ARM processors they could still be going strong. You can blame that on Krzanich for not seeing that at a minimum Atom needed to be both low cost and better than ARM, not trailing to insure it didn’t disrupt Core.

 

[dttprofessor]

pat is right, fab is future!

 

[jpiniero]

Also Swan oversaw the roll out, or lack of roll out of the first EUV process and the inability to contractually deliver Aurora. There hadn’t been wafer volume growth for years and Fab 34 in Ireland and fab 42 in Arizona would probably be enough capacity for the decreasing ‘economic’ volume. Swan did not have the Intel vision of ‘build it and they will come’. He didn’t swing for the fence.

A big part of the reason Intel shut down 10 nm was because they needed the space back for more 14 nm. Going to Quad Core on mobile (plus more cores on Desktop) and the Apple modem meant they didn't have enough capacity.

 

[511]

Pat is old school Intel and based on ‘old school Intel’ process leadership is everything, the essence of ‘Moore’s Law

This is true by the way process leadership matters if you don't have it your products are not the best.

 

[igor_kavinski]

This is true by the way process leadership matters if you don't have it your products are not the best.

Guess AMD missed the memo?

You know what they say. Ignorance is bliss. Nobody told AMD that having fabs is mandatory to have the best CPU so they are just ignorantly and blissfully going about being the absolute worse they can be, meaning a CPU company without fabs.

They learned the hard way that design excellence matters and what matters even more is iterating on your design meaningfully. Not have a great design and try to coast on its success for five years or more.

 

[511]

Guess AMD missed the memo?

You know what they say. Ignorance is bliss. Nobody told AMD that having fabs is mandatory to have the best CPU so they are just ignorantly and blissfully going about being the absolute worse they can be, meaning a CPU company without fabs.

They learned the hard way that design excellence matters and what matters even more is iterating on your design meaningfully. Not have a great design and try to coast on its success for five years or more.

Tell me what would your design would look like without a good fab process to fab it ? Designs matters but process matters even more this should have been clear by now.

AMD uses the best process for their highest Margin Market DC.

 

[igor_kavinski]

Tell me what would your design would look like without a good fab process to fab it ? Designs matters but process matters even more this should have been clear by now.

No.5 and No.6 using great design with a process that should be considered garbage in 2025 yet here is AMD hogging the list.

Design wins over anything else.

Apple also wins because of design. They were winning in perf/watt with their phone chips long before M1 debuted. Intel's problem is that they always relied on process. The moment that was taken away, their design related flaws came to light for the entire world to see.

Even N3B couldn't save Intel. Apple using the same process didn't churn out a turd as bad as Arrow Lake.

 

[511]

View attachment 127598

No.5 and No.6 using great design with a process that should be considered garbage in 2025 yet here is AMD hogging the list.

Design wins over anything else.

Apple also wins because of design. They were winning in perf/watt with their phone chips long before M1 debuted. Intel's problem is that they always relied on process. The moment that was taken away, their design related flaws came to light for the entire world to see.

I agree Intel's design relied on the process but that doesn't mean they don't know how to make a good design ADL vs Zen 3 12600K/12700K is better than a 5800XT/5600XT the only reason so many Zen3 is selling is due to platform longevity there ain't no way 5800XT is better than a 12700K.

ARL is a mess of a desktop design.

 

[Doug S]

Pat:
Overbuilding Foundry shells
Missing AI GPUs.

What do you think he missed about AI GPUs? That Intel should have been more aggressive about working with Nvidia to fab some of them AI GPUs, especially given their packaging capacity which has been a problem for TSMC? If that's what you're saying I fully agree. He missed the boat, and Intel should have been able to get at least a small piece of that action. Even if in order to win the business they had to offer them a price that didn't leave a whole lot of room for profit, making some of Nvidia's AI GPUs would give their foundry efforts some real street cred that is sorely lacking.

Or are you saying Intel should have been in the AI GPU game itself? Intel has always been terrible when it comes to GPUs, and aspiring to enter that market doesn't mean they'd be at all successful. AMD has made only a tiny dent in it so far and they're miles ahead of Intel when it comes to GPUs! IMHO if he'd tried most likely he'd have wasted a bunch of money and ended up trying to sell a subpar product no one wanted.

I suppose since Nvidia can't keep up with demand, maybe Intel could have sold a few Intel AI GPUs as sloppy seconds, like AMD does. I just think trying to enter a new highly competitive market against an incumbent who is a decade ahead of you (and a hopeful up and comer who is also way ahead of you) is a bad idea when you're in trouble. Intel needs laser focus on two things: getting the foundry working, and getting their x86 house in order, before they worry about entering new markets based on a hope and a prayer.

 

[Josh128]

Suffice to that that if AMD had been stuck a node behind Intel all this time (since 2017), things would look a lot different in the CPU market today. AMD wouldnt be selling EPYCs and TRs for $10k+ for sure. Their PR and marketshare wouldnt have progressed enough for that to happen.

ARL vs Zen 5 is the first CPU gen from AMD since the Athlon 64 days (or ever?) that AMD was able to match and/or beat Intel CPUs despite having a full node disadvantage. The jig is up now for Intel, I think. The days of them "sitting on 20%+ IPC" designs to pull out for a rainy day are over.

It will be truly interesting to see Intel vs AMD on the same node, 2nm. Will they be evenly matched or will AMD continue to pull away as they have done with Zen 5 despite not having node advantage?

 

[511]

Suffice to that that if AMD had been stuck a node behind Intel all this time (since 2017), things would look a lot different in the CPU market today. AMD wouldnt be selling EPYCs and TRs for $10k+ for sure. Their PR and marketshare wouldnt have progressed enough for that to happen.

ARL vs Zen 5 is the first CPU gen from AMD since the Athlon 64 days (or ever?) that AMD was able to match and/or beat Intel CPUs despite having a full node disadvantage. The jig is up now for Intel, I think. The days of them "sitting on 20%+ IPC" designs to pull out for a rainy day are over.

It will be truly interesting to see Intel vs AMD on the same node, 2nm. Will they be evenly matched or will AMD continue to pull away as they have done with Zen 5 despite not having node advantage?

N3B is not much better than N4P according to even TSMC lol as for NVL vs Zen6 i expect NVL to have edge in MT and Zen 6 is going to be slightly ahead in ST(mid single digit %).

What do you think he missed about AI GPUs? That Intel should have been more aggressive about working with Nvidia to fab some of them AI GPUs, especially given their packaging capacity which has been a problem for TSMC? If that's what you're saying I fully agree. He missed the boat, and Intel should have been able to get at least a small piece of that action. Even if in order to win the business they had to offer them a price that didn't leave a whole lot of room for profit, making some of Nvidia's AI GPUs would give their foundry efforts some real street cred that is sorely lacking.

Or are you saying Intel should have been in the AI GPU game itself? Intel has always been terrible when it comes to GPUs, and aspiring to enter that market doesn't mean they'd be at all successful. AMD has made only a tiny dent in it so far and they're miles ahead of Intel when it comes to GPUs! IMHO if he'd tried most likely he'd have wasted a bunch of money and ended up trying to sell a subpar product no one wanted.

I suppose since Nvidia can't keep up with demand, maybe Intel could have sold a few Intel AI GPUs as sloppy seconds, like AMD does. I just think trying to enter a new highly competitive market against an incumbent who is a decade ahead of you (and a hopeful up and comer who is also way ahead of you) is a bad idea when you're in trouble. Intel needs laser focus on two things: getting the foundry working, and getting their x86 house in order, before they worry about entering new markets based on a hope and a prayer.

They could have done both tbh packing for Nvidia plus selling theirs what's the point of having both design and foundry under one roof if you miss both opportunity also Intel has been terrible due to their mindset they never designed their process for GPUs and they kept canning projects one after another for their GPU.
Xe GPU IP will keep improving though PTL's highlight is it's GPU.

 

[desrever]

The problem with Intel doing packaging is that EMIB is very different then what everyone else does and more expensive thus nobody outside of Intel will want to use it until proven and cheaper.

 

[jpiniero]

Or are you saying Intel should have been in the AI GPU game itself? Intel has always been terrible when it comes to GPUs, and aspiring to enter that market doesn't mean they'd be at all successful. AMD has made only a tiny dent in it so far and they're miles ahead of Intel when it comes to GPUs! IMHO if he'd tried most likely he'd have wasted a bunch of money and ended up trying to sell a subpar product no one wanted.

Do you forget the multiple companies Intel bought specifically for AI, only to be ignored?

And that was before Wall Street's orgasms over anything AI.

 

[desrever]

Do you forget the multiple companies Intel bought specifically for AI, only to be ignored?

And that was before Wall Street's orgasms over anything AI.

I think part of the problem when Intel acquires companies is they force them onto Intel foundry and then the foundry can't deliver so everything is delayed and cancelled.

 

[511]

Habana Movidius and a few more

The problem with Intel doing packaging is that EMIB is very different then what everyone else does and more expensive thus nobody outside of Intel will want to use it until proven and cheaper.

EMIB is not that expensive as you are making out to be the closest EMIB Alternative is COWOS-L which is cheaper than EMIB.
As for proven Intel had shipped Millions of DC CPUs using EMIB and with 600-700mm2 die size Ponte Vehico is shipped with EMIB that should be enough Amazon uses EMIB.

 

[511]

TIL

https://twitter.com/x/status/1948025647143538780

 

[Doug S]

TIL

Yes definitely count in me with "I never would have guessed cow farts can impact semiconductor yields". Also wouldn't have guessed that filtering out methane is apparently so difficult that buying farms to get rid of the cows is the cheaper alternative.

 

[Joe NYC]

From Intel's 10Q, under Risk Factors:

"If we are unable to secure a significant external foundry customer for Intel 14A, our next generation semiconductor manufacturing process technology, we may pause or discontinue our pursuit of next generation leading-edge process technologies, which may have significant strategic business, financial, operational and reputational risks and repercussions."

https://www.intc.com/filings-reports/all-sec-filings/content/0000050863-25-000109/0000050863-25-000109.pdf

 

[Josh128]

It really looks like its over for Intel as we know it. The infamous "bet" has been lost, it seems.