Discussion Leading Edge Foundry Node advances (TSMC, Samsung Foundry, Intel) - [2020 - 2025]

DisEnchantment · Apr 24, 2020

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.

---------------------------------------------------------------------------------------------------------------------------------------------------

FEEL FREE TO CREATE A NEW THREAD FOR 2025+ OUTLOOK, I WILL LINK IT HERE

Saylick · Feb 20, 2025

uzzi38 said:
Yep, they can't be compared at all.

Both plots are both very, very good though. Like it's really good signs for both N2 and 18A, they both look like they're in great condition.

I’m seeing far too many Xitter posts of people comparing the two plots directly, mostly by those who clearly don’t have a background in this stuff, as a way of claiming 18A superiority.

The density looks great, sure, but seeing as how SRAM density improvements have stagnated it’s not all surprising to me to see everyone at roughly the same spot on that aspect.

Geddagod · Feb 20, 2025

@SiliconFly What you are doing on twitter man is pathetic dude. But I also shouldn't have fallen for the obvious rage bait, that's on me.

Also, I love silicon? Really? Should be I love INTEL silicon.

Hitman928 · Feb 20, 2025

ajsdkflsdjfio said:
I genuinely cannot find any material claiming that intel planned to turn a profit with their foundry by 2025, the year where 5N4Y was to finish. I browsed through articles from the date of Pat Gelsinger signing on until 12/31/2021 and found only claims about process leadership.

View attachment 117670
They only claimed process leadership by 2025 for a reason, because they understand that after process leadership comes building out fabs, attracting customers etc etc. Even with customers it takes a while to turn a profit from such large investments, so I don't think anybody expected foundry to profit remotely close to the time they finished the 5N4Y plan. The demand for 18a thus far is likely less than what Pat might've imagined it to be, but this isn't set in stone and might change for the better in the next year or so as 18a draws interest from more people given it's relatively impressive specifications.

Intel 3 might've been the node they first planned to bring on foundry customers, but 18A was their golden egg so to speak, the turning point where people might change to team blue. The node where they advertise that they finally caught up with TSMC and the node where they expect EDA to be on parity (whatever that means) with TSMC, allowing easier onboarding of customers.

View attachment 117673

Either way I think rather than foundry being profitable or not, the main issue with Pat's tenure (whether directly his fault or not) in the eyes of the board was the poor performance of the company otherwise like with Gaudi and bleeding market share in DCAI. If even just their DCAI/Consumer products held the status quo during Pat's plan, I feel that the board would've been 10x happier and gave him more time to carry out his plan. Regardless, I think that these stem from #1 a poorly functioning foundry, #2 lackluster engineering teams. These issues largely predate Pat, but I would say that he definitely did not do enough to plug the leaks happening outside of foundry. But at the very least Intel design teams seem to be releasing mostly on time after meteor lake/sapphire rapids. I mean if you look at some of their delays for their products in 2010-2020 they are on the magnitude of years rather than months, which were affordable to intel back then, but would kill intel right now.

I'd agree with this and with the idea that Pat was too optimistic with building out fabs, not necessarily because of a lack of customers but just in general a lack of capital due to poor intel performance and other factors like German subsidies stalling out and delaying of CHIPS act funding. I mean if you look at Intel's projected volumes for their nodes, 18A will only be a fourth of their overall capacity by 2026, and EUV nodes as a whole only a third. I think they could do with more capacity, but obviously they don't have the capital to build right now. He carried out a game plan that would've been great for Intel 5-10 years earlier, but painful for current Intel. Either way, Intel before Pat was trending downwards even if they were starting at a higher position. If they can hold onto foundry and survive unbroken through 2025, I think they'll have passed through the valley and begun to climb the mountain.

I never said they talked about profits, I said they didn't start talking about lack of profits until much later on, which is true. Before that, they threw up a bunch of hype talking about customer pipelines and engagements and lifetime deals, etc., which is what they did to make people think they had large customers in the works. Problem was, that pipeline didn't turn into signed contracts. Exact same thing happened with Gaudi. It wasn't until later when it was clear they had no customers for Intel 3 and their financials started to turn south in a major way that they were somewhat forced to admit the truth about the state of IFS and profitability.

If you want to see why Pat was fired in visual terms, this is it (remember that Pat started as CEO in Feb 2021):

Pat was spending basically everything they could on building out fab capacity for IFS to be prepared to rival TSMC, or as Pat put it, "I've bet the whole company on 18a." Follow that by no major customer wins to speak of and multiple fab build-outs around the world being suspended, there was very little to show for all this spending. The only reason their cash flow wasn't worse, is that because they ran out of money and had to literally mortgage parts of their foundry's future for money to build-out their fab capacity.

There were, of course, other issues, but his almost tunnel vision focus on building IFS caused or exacerbated those other issues. He gave the foundry division an unlimited budget while making the design side change their operations to be more efficient. The need for greater efficiency on the design side was long overdue, but it just shows where his focus was in terms of spending and opportunity but it didn't work out like he expected, even though 18a seems to be shaping up to be a competitive node. The CEO is beholden to the board and the board is beholden to the share holders. Share holders speak in terms of money, not technology. Pat spent all the money while revenues dropped and seemingly had no intention of changing course. When enough time passed that he was forced to discuss the state of things, he could only promise the hope for profitability on his plan years into the future. That's why he was fired.

P.S. EDA parity basically means how well industry standard tools are supported for chip design on Intel fabs as before IFS 2.0, Intel by and large used only internally developed tools for chip design which no one outside the company knew how (or wanted) to use.

Hitman928 · Feb 20, 2025

Saylick said:
I’m seeing far too many Xitter posts of people comparing the two plots directly, mostly by those who clearly don’t have a background in this stuff, as a way of claiming 18A superiority.

The density looks great, sure, but seeing as how SRAM density improvements have stagnated it’s not all surprising to me to see everyone at roughly the same spot on that aspect.

I just briefly looked through the slides shown, but I believe Intel also requires BSPD to get the last 10% to match TSMC's SRAM density. Obviously that's fine from a technology stand point as they should (at least according to them) have it ready to go this year, but from a customer perspective, many will not want to use BSPD due to the added costs and (maybe?) thermal issues.

Hitman928 · Feb 20, 2025

511 said:
Not capitalizing on AI Boom was partly Swan's as well Pat's biggest fault was cancelling Rialto Bridge

This is one of the things I agreed with Pat on. The problem was, no one wanted it. He tried to have them instead focus on a solution people would want (Falcon Shore GPU) but that too was going to be too little to late, so it got scrapped as well.

cannedlake240 · Feb 20, 2025

Hitman928 said:
I just briefly looked through the slides shown, but I believe Intel also requires BSPD to get the last 10% to match TSMC's SRAM density. Obviously that's fine from a technology stand point as they should (at least according to them) have it ready to go this year, but from a customer perspective, many will not want to use BSPD due to the added costs and (maybe?) thermal issues.

Pretty sure they claimed the opposite, that Powervia increases area by 1.1x. Which slide or claim is it?

uzzi38 · Feb 20, 2025

Geddagod said:
@SiliconFly What you are doing on twitter man is pathetic dude. But I also shouldn't have fallen for the obvious rage bait, that's on me.
View attachment 117709
View attachment 117710
Also, I love silicon? Really? Should be I love INTEL silicon.

You didn't see my one and only set of interactions with him on Twitter?

Hilarious. I really wish I had screenshots, because he deleted everything.

Hitman928 · Feb 20, 2025

cannedlake240 said:
Pretty sure they claimed the opposite, that Powervia increases area by 1.1x. Which slide or claim is it?

BSPD, not PowerVia specifically. Though they are using PowerVia, just outside the bitcell.

https://twitter.com/x/status/1892251370192843253

https://twitter.com/x/status/1892250623191511142

Saylick · Feb 20, 2025

uzzi38 said:
You didn't see my one and only set of interactions with him on Twitter?

Hilarious. I really wish I had screenshots, because he deleted everything.

I saw that exchange... basically, they claimed that modern CPU cores only use HP SRAM cells, but then you and Cheese corrected them quickly with AMD slides showing their use of HD cells for many Zen generations lol

Doug S · Feb 20, 2025

Hitman928 said:
He got somewhat close to 5 nodes in 4 years, you could argue foundry leadership

No one can argue that. What has Intel shipped a single unit of that's better than the mass volumes of N3E that TSMC has shipped? Maybe they will be ahead with 18A, we'll see. But no one can argue that until something made with 18A chips is available for retail sale and ships to consumer hands.

Hitman928 · Feb 20, 2025

Doug S said:
No one can argue that. What has Intel shipped a single unit of that's better than the mass volumes of N3E that TSMC has shipped? Maybe they will be ahead with 18A, we'll see. But no one can argue that until something made with 18A chips is available for retail sale and ships to consumer hands.

The bolded is basically what I was meaning to say.

Saylick · Feb 20, 2025

Doug S said:
No one can argue that. What has Intel shipped a single unit of that's better than the mass volumes of N3E that TSMC has shipped? Maybe they will be ahead with 18A, we'll see. But no one can argue that until something made with 18A chips is available for retail sale and ships to consumer hands.

This reminds me of AMD's attempt to usurp Intel in servers in how AMD had the better server chip for many years (starting with Rome) but since they couldn't ship it in mass volume + Intel being the trusted vendor, it took many years and many generations of consistent execution for AMD to start making a dent. Yet to this day, AMD still doesn't have >50% market share in servers.

I see 18A the same way. Sure, it possibly can have a slight edge against N2 in certain characteristics, but until Intel can 1) pump out this node in vast volumes, 2) build a PDK that is comparable to TSMC, 3) attract companies to use the node, and 4) consistently execute while also surviving for a few more years, this will be a "let's wait and see" moment rather than the "OMG, Intel is BACK" that some are shouting from the rooftops.

Doug S · Feb 20, 2025

Saylick said:
This reminds me of AMD's attempt to usurp Intel in servers in how AMD had the better server chip for many years (starting with Rome) but since they couldn't ship it in mass volume + Intel being the trusted vendor, it took many years and many generations of consistent execution for AMD to start making a dent. Yet to this day, AMD still doesn't have >50% market share in servers.

I see 18A the same way. Sure, it possibly can have a slight edge against N2 in certain characteristics, but until Intel can 1) pump out this node in vast volumes, 2) build a PDK that is comparable to TSMC, 3) attract companies to use the node, and 4) consistently execute while also surviving for a few more years, this will be a "let's wait and see" moment rather than the "OMG, Intel is BACK" that some are shouting from the rooftops.

Well I agree from the standpoint of the foundry business. Intel can't win much business no matter how good 18A is because they don't have the wpm capacity to handle that much.

But it would matter from the standpoint of Intel's x86 business, if they had a process advantage over AMD. That could make up for some of their sins as far as ridiculous power consumption, or allow them to clock higher / be faster at the same ridiculous power consumption of today.

ajsdkflsdjfio · Feb 20, 2025

Hitman928 said:
I never said they talked about profits, I said they didn't start talking about lack of profits until much later on, which is true. Before that, they threw up a bunch of hype talking about customer pipelines and engagements and lifetime deals, etc., which is what they did to make people think they had large customers in the works. Problem was, that pipeline didn't turn into signed contracts. Exact same thing happened with Gaudi. It wasn't until later when it was clear they had no customers for Intel 3 and their financials started to turn south in a major way that they were somewhat forced to admit the truth about the state of IFS and profitability.

If you want to see why Pat was fired in visual terms, this is it (remember that Pat started as CEO in Feb 2021):

Pat was spending basically everything they could on building out fab capacity for IFS to be prepared to rival TSMC, or as Pat put it, "I've bet the whole company on 18a." Follow that by no major customer wins to speak of and multiple fab build-outs around the world being suspended, there was very little to show for all this spending. The only reason their cash flow wasn't worse, is that because they ran out of money and had to literally mortgage parts of their foundry's future for money to build-out their fab capacity.

There were, of course, other issues, but his almost tunnel vision focus on building IFS caused or exacerbated those other issues. He gave the foundry division an unlimited budget while making the design side change their operations to be more efficient. The need for greater efficiency on the design side was long overdue, but it just shows where his focus was in terms of spending and opportunity but it didn't work out like he expected, even though 18a seems to be shaping up to be a competitive node. The CEO is beholden to the board and the board is beholden to the share holders. Share holders speak in terms of money, not technology. Pat spent all the money while revenues dropped and seemingly had no intention of changing course. When enough time passed that he was forced to discuss the state of things, he could only promise the hope for profitability on his plan years into the future. That's why he was fired.

P.S. EDA parity basically means how well industry standard tools are supported for chip design on Intel fabs as before IFS 2.0, Intel by and large used only internally developed tools for chip design which no one outside the company knew how (or wanted) to use.

I can see your point about most of what you are saying here, minus the point about wasting all this money for building out fabs. Yes Pat planned to build out too many fabs to the point which Intel could no longer afford and as you say there are no "major" fab customers to speak of which results in canceled/delayed fab build outs. Yes some money was wasted on planning/preparing for those fabs to be built but the vast majority of the money for those canceled fabs has not been spent It's not like Intel spent all this money to build out fab capacity which will end up not being used. A good portion of canceled fabs were nothing more than plans, and the ones that did begin construction are being mostly delayed like the one in Arizona. My point about Intel never mentioning profitability was to say that Pat never intended for foundry to be a get rich quick scheme, but always knew that it would take years, longer than just 4 years, to turn a profit and pay off all the investment required to regain Intel's technology parity.

As for the fabs that have opened recently or in the next 3 years, I would not call them wasted capacity. All indicators point to 18A being an attractive node not only for tentative customers but for Intel themselves. With the fabs that did end up getting built, I fail too see how they will be underutilized in the coming years. If you look at intel's own projections for 18A/14A capacity. In fact many people have doubts as to whether Intel will have enough leading edge capacity for both their products and potential customers. Pat may have hyper focused on foundry but versus the path that they were on before Pat, I feel like Intel is better off now than before. Either way I think we can agree that the effect Pat has had on the company, whether amazing or detrimental, can only be seen in the next few years.

jdubs03 · Feb 20, 2025

Doug S said:
Well I agree from the standpoint of the foundry business. Intel can't win much business no matter how good 18A is because they don't have the wpm capacity to handle that much.

But it would matter from the standpoint of Intel's x86 business, if they had a process advantage over AMD. That could make up for some of their sins as far as ridiculous power consumption, or allow them to clock higher / be faster at the same ridiculous power consumption of today.

Gotta start somewhere. At this point there are probably only two outcomes: gets back to growth or bankruptcy. @Saylick just remember products built on 18A will be available half a year before any products built on N2.

jpiniero · Feb 20, 2025

jdubs03 said:
Gotta start somewhere. At this point there are probably only two outcomes: gets back to growth or bankruptcy.

Or they dump the fabs.

Saylick · Feb 20, 2025

jdubs03 said:
Gotta start somewhere. At this point there are probably only two outcomes: gets back to growth or bankruptcy. @Saylick just remember products built on 18A will be available half a year before any products built on N2.

Right, but I suspect also in limited quantities.

jdubs03 · Feb 20, 2025

Saylick said:
Right, but I suspect also in limited quantities.

Yeah true. That’s going to be the main problem they’ll need to solve. Getting full capacity and expanding that. But if 18A if provably on the level that should inspire confidence and lead to more contracts.

511 · Feb 21, 2025

Intel 18A | See Our Biggest Process Innovation

See how the Intel 18A semiconductor manufacturing process is the foundation of the systems foundry for the AI era.

www.intel.com

18A is available for the external

lightisgood · Feb 21, 2025

511 said:
Intel 18A | See Our Biggest Process Innovation

See how the Intel 18A semiconductor manufacturing process is the foundation of the systems foundry for the AI era.

www.intel.com

18A is available for the external

IFS done it.
i18A certainly shows off 1 year lead at GAAFET and 2 years lead at BSPDN vs. TSMC.

511 · Feb 21, 2025

lightisgood said:
IFS done it.
i18A certainly shows off 1 year lead at GAAFET and 2 years lead at BSPDN vs. TSMC.

IMO Intel issues is only half solved designs horrible PPA remains a issue(both Arc and P core ) E core is fine Arc is supposed to be addressing it with Xe3P

Thibsie · Feb 21, 2025

lightisgood said:
IFS done it.
i18A certainly shows off 1 year lead at GAAFET and 2 years lead at BSPDN vs. TSMC.

Where are the products ? Announcements are nice but this is Intel we're talking about. I'll believe it when I'll see it.
And just to be clear, I'd like to see it. Nobody needs a monopoly.

Hitman928 · Feb 21, 2025

ajsdkflsdjfio said:
I can see your point about most of what you are saying here, minus the point about wasting all this money for building out fabs. Yes Pat planned to build out too many fabs to the point which Intel could no longer afford and as you say there are no "major" fab customers to speak of which results in canceled/delayed fab build outs. Yes some money was wasted on planning/preparing for those fabs to be built but the vast majority of the money for those canceled fabs has not been spent It's not like Intel spent all this money to build out fab capacity which will end up not being used. A good portion of canceled fabs were nothing more than plans, and the ones that did begin construction are being mostly delayed like the one in Arizona. My point about Intel never mentioning profitability was to say that Pat never intended for foundry to be a get rich quick scheme, but always knew that it would take years, longer than just 4 years, to turn a profit and pay off all the investment required to regain Intel's technology parity.

As for the fabs that have opened recently or in the next 3 years, I would not call them wasted capacity. All indicators point to 18A being an attractive node not only for tentative customers but for Intel themselves. With the fabs that did end up getting built, I fail too see how they will be underutilized in the coming years. If you look at intel's own projections for 18A/14A capacity. In fact many people have doubts as to whether Intel will have enough leading edge capacity for both their products and potential customers. Pat may have hyper focused on foundry but versus the path that they were on before Pat, I feel like Intel is better off now than before. Either way I think we can agree that the effect Pat has had on the company, whether amazing or detrimental, can only be seen in the next few years.

I'm sure Pat wasn't promising big profits for IFS right away, but there's no way the plan was to spend 10s of billions of dollars on additional capex, go into $50B of debt, mortgage half the new fab, invest in land and fab build-outs to suspend them a couple of years later, all while attracting no major wafer contracts and having to sell off profitable pieces of the company. Pat's plan failed, there's really no way around it. At some point, the board isn't going to let you keep going when the financial situation is this bad. Now, on the technology side, his plan largely succeeded, not perfectly but for the most part, but that's only half the battle he was fighting. The business side of the plan did not work out. Maybe they can pull out a win in the end, but the board seems to be done with IFS, so I don't know if it'll have much more time to try and pull out a win.

511 · Feb 21, 2025

There is only one way to recover money from 18A is to build more products on it doesn't matter for whome.
The first step should be the products come back to internal nodes for majority of the products if there is no significant differences between node offering.

DZero · Feb 21, 2025

511 said:
IMO Intel issues is only half solved designs horrible PPA remains a issue(both Arc and P core ) E core is fine Arc is supposed to be addressing it with Xe3P

P core must be reformed or make the E core become the new P core.

Discussion Leading Edge Foundry Node advances (TSMC, Samsung Foundry, Intel) - [2020 - 2025]

Golden Member

Diamond Member

Golden Member

Diamond Member

Diamond Member

Diamond Member

Senior member

Platinum Member

Diamond Member

Diamond Member

Diamond Member

Diamond Member

Diamond Member

Diamond Member

Member

Golden Member

Lifer

Diamond Member

Golden Member

Golden Member

Senior member

Golden Member

Golden Member

Diamond Member

Golden Member

Senior member