Discussion Leading Edge Foundry Node advances (TSMC, Samsung Foundry, Intel)

[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.

 

[SpudLobby]

What's also interesting and makes me curious is what's going on with Intel's test chips for Qualcomm's design. Is it about the HD cells' leakage?

 

[Geddagod]

What's also interesting and makes me curious is what's going on with Intel's test chips for Qualcomm's design. Is it about the HD cells' leakage?

Could be. Maybe SRAM as well, considering SRAM density had always been a weakness for Intel?

 

[jpiniero]

Nothing timely about it. Seems like old regurgitated information mostly. Pretty odd thing to publish now.

That Qualcomm bailed because Intel missed deadlines is new info.

 

[SpudLobby]

That Qualcomm bailed because Intel missed deadlines is new info.

That’s not what the article says RE: deadlines. Makes 0 mention of that actually.

 

[jpiniero]

That’s not what the article says RE: deadlines. Makes 0 mention of that actually.

Missing the performance targets really in the timeline given.

 

[LightningZ71]

It is my opinion that the huge thorn in Intel's side is AMD's inroads into the very high marginserver/DC class products. Those margins were likely a key factor in their ability to finance their fab development. Loosing volume AND having to sell at lower ASPs per processor than before is likely having a snowballing impact on their margin to reinvest in fab R&D. They likely really needed a competitive GPU unit to make up for that, and they aren't there yet.

 

[SpudLobby]

Question for someone, anyone really.

The "5NM" Samsung process used in the 888, 8 Gen 1 is derivative of Samsung's 7NM process, and the density is more comparable to TSMC N6 yes? 8 Gen 1 was claimed 4NM LPX fwiw but this is just 5NM LPE/LPP again.

Point being, independent of the parametric yields issues that negatively affected performance and power, that "5nm" process was really more an N6, N7P competitor. We saw this logically with the power gap between the 8 Gen 1 and 8 Gen 1 +. Some of that was probably about yields though.

And the difference with Samsung's 4NM LPE/LPP (not LPX) is that 4NM in this case is a completely different process, with higher density, different EUV layers and hopefully more performance, power benefits. TSMC's N5 -> N4 is an optical shrink with some design rule differences but fairly small differences, whereas Samsung's 4NM is what their 5NM should have been all along afaict.

Hoping to see SF4 get good parametric yields and be competitive with N5/N4. It's not impressive if Samsung needs GAAFETs to compete with e.g. N4P, basically punching below their weight. Hopefully they realize this though.

 

[SpudLobby]

Also, does anyone know what depreciation schedules are like for leading processes even vaguely? E.g. will N3, N5 wafer prices come down substantially in 5 years or only minimally?

 

[Lodix]

Question for someone, anyone really.

The "5NM" Samsung process used in the 888, 8 Gen 1 is derivative of Samsung's 7NM process, and the density is more comparable to TSMC N6 yes? 8 Gen 1 was claimed 4NM LPX fwiw but this is just 5NM LPE/LPP again.

Point being, independent of the parametric yields issues that negatively affected performance and power, that "5nm" process was really more an N6, N7P competitor. We saw this logically with the power gap between the 8 Gen 1 and 8 Gen 1 +. Some of that was probably about yields though.

And the difference with Samsung's 4NM LPE/LPP (not LPX) is that 4NM in this case is a completely different process, with higher density, different EUV layers and hopefully more performance, power benefits. TSMC's N5 -> N4 is an optical shrink with some design rule differences but fairly small differences, whereas Samsung's 4NM is what their 5NM should have been all along afaict.

Hoping to see SF4 get good parametric yields and be competitive with N5/N4. It's not impressive if Samsung needs GAAFETs to compete with e.g. N4P, basically punching below their weight. Hopefully they realize this though.

Yes SD888 and SD8 Gen 1 use Samsung 5nm. But it is not correct to call it just a derivative, because it has some meaningful changes and improvements to density compared to their 7nm. And In density Samsung 5/4nm is comparable to TSMC's 5/4nm. In terms of efficiency is difficult to say, because as you mention when this SoC were made Samsung had terrible yields that are now corrected.
Also to mention Samsung 3nm should be comparable to TSMC 3nm in density and efficiency.

This is their latest information. But we don't know how a good yielding 4LPP fairs.

 

[A///]

It is my opinion that the huge thorn in Intel's side is AMD's inroads into the very high marginserver/DC class products. Those margins were likely a key factor in their ability to finance their fab development. Loosing volume AND having to sell at lower ASPs per processor than before is likely having a snowballing impact on their margin to reinvest in fab R&D. They likely really needed a competitive GPU unit to make up for that, and they aren't there yet.

yes but one must consider that the current intel products are years late. these would have been going up against older epycs not that there's much competition to go by that mark anyway now looking back. if intel's future nodes are compelling and they can offer enough volume they can take traffic away from tsmc simply by selling their ifs 2.0 fab time near cost. it'd be abrazen decision but they'd gain consumers and tsmc would react by lowering prices to be competitive. Intel's future nodes may not be as great as future tsmc, 9% slower with slightly higher power usage but intel takes it on the chin by selling their fab wafer time near cost. it weakens tsmc's ability to reinvest even if apple is their number 1 girl in the neighbourhood. even apple may be enticed for their computers minus mobiles like the air or their phone or ipad where efficiency is king.

even pat said he'd love to have amd fab with them. intel wouldn't steal any design from amd. that'd be a blow for intel's reputation or what remains of it. intel wants to make it painful for tsmc to operate and force tsmc to 2nd place by taking their lunch money and slowly using that money to improve their own process while tsmc makes do with scraps. again.

 

[clemsyn]

yes but one must consider that the current intel products are years late. these would have been going up against older epycs not that there's much competition to go by that mark anyway now looking back. if intel's future nodes are compelling and they can offer enough volume they can take traffic away from tsmc simply by selling their ifs 2.0 fab time near cost. it'd be abrazen decision but they'd gain consumers and tsmc would react by lowering prices to be competitive. Intel's future nodes may not be as great as future tsmc, 9% slower with slightly higher power usage but intel takes it on the chin by selling their fab wafer time near cost. it weakens tsmc's ability to reinvest even if apple is their number 1 girl in the neighbourhood. even apple may be enticed for their computers minus mobiles like the air or their phone or ipad where efficiency is king.

But the government of Taiwan will help out TSMC against any form of competition from Intel which makes it tougher for Intel to compete.

 

[A///]

But the government of Taiwan will help out TSMC against any form of competition from Intel which makes it tougher for Intel to compete.

this is true, but we can see how it's working out for sk and samsung. samsung could be considered a state owned company but they're solidly in third place. us gov won't help intel like they do their own but tsmc was "asked" to build in the us and by the time they open up their 5nm won't be state of the art. it's hard to say what's going to happen in the next few years but I see a resurgence in intel. team blue took a giant viagra when gelsinger came back.

 

[clemsyn]

this is true, but we can see how it's working out for sk and samsung. samsung could be considered a state owned company but they're solidly in third place. us gov won't help intel like they do their own but tsmc was "asked" to build in the us and by the time they open up their 5nm won't be state of the art. it's hard to say what's going to happen in the next few years but I see a resurgence in intel. team blue took a giant viagra when gelsinger came back.

I really hope so but I am on the other side of the fence. I think it would be very difficult for Intel even with Pat and the "Chips Act" this time around. Brian K messed up this company pretty bad.

 

[A///]

I really hope so but I am on the other side of the fence. I think it would be very difficult for Intel even with Pat and the "Chips Act" this time around. Brian K messed up this company pretty bad.

brian was the cherry on top of the cake while he boinked a woman from work. intel made a lot of terrible decisions years before he sat in the ceo seat. intel's board hired on dumbasses like murth the father of death. never discount a turning ship. either intel remains on this path forever or they mess up so bad they end up in a worse position than amd ever was to the nth degree. at the end of the day intel will still atrtract people wanting assuring platforms even if they're not the best or wanting to avoid the diy ups and downs with amd unless you're buying tr or epyc. what's stopping intel now in the immediate is their power usage. if their power usage was the same as amd's do you think a lot of people would go with amd? I don't. either is fine if you don't pay a lot for electricity but many in other countries are getting their asses chapped by their utilities and are paying 5-7x what the normal north americaner does. amd makes sense for them. gelsinger is trying to reignite the flame in intel. intel spent way too long under multiple ceos thinking they were the big badasses of the hw world because no one could challenge them until tsmc did and succeeded and they thought amd would never come back, but they did. they've been eating humble and crow pie for a few years now.

the only glimpse of intel's future nodes we have is jhh's commentary about intel's future nodes. like i said, he could have said that to make tsmc anxious but why? nvidia recently expanded their wafer orders because they're killing it due to the ai goldrush. why redesign for intel or mess with tsmc who'll tell them to take a hike, again. why would jhh give a future competitor in the gpu space because arc isn't very useful for now a competitive edge? it's too easy to write off intel because they've been screwing the pooch for years now. give it time.

you lot no one in specific are way too focused on people like mlid, that fat english man, raichu, kobra kimi or whomever it is that's the flavor the week. too many people focused on ever changing bull. mark my words, if intel maintains their stagnant course, it'll be amd who'll highly segment and pull and intel. they already do to a small degree now with certain hardware ranges. why would amd waste money on r&d when they know they can milk more money? and for those of us who are older this isn't our first time on this rodeo. This isn't the first time intel's been knocked on their head. it isn't the first time amd has outperformed the competition. it also isn't the first time apple was faster than a pc. Give me my old haircut, and hair for that matter alongside the lost colour and my questionable taste in clothes and it would be the 90s again.

 

[moinmoin]

or they mess up so bad they end up in a worse position than amd ever was to the nth degree.

Only way they could manage that is by going bankrupt. There really wasn't much room below AMD's lowest point.

they've been eating humble and crow pie for a few years now.

Have they? Must have missed it then.

the only glimpse of intel's future nodes we have is jhh's commentary about intel's future nodes. like i said, he could have said that to make tsmc anxious but why?

He was saying nice things about a potential partner. That's standard business practice.

 

[Ajay]

brian was the cherry on top of the cake while he boinked a woman from work. intel made a lot of terrible decisions years before he sat in the ceo seat. intel's board hired on dumbasses like murth the father of death. never discount a turning ship

Barrett held the ship steady, behaving more like his previous predecessors - though not having the same standing. Otellini just didn't get it, IMHO. Intel just kept getting larger and larger - adding more business units, etc. He also missed the boat on phones. Sure, that was hard to see back then, but it is such a historically bad decision. Krzanich - was supposedly the one who would come in and straighten out Intel's process development problems. He created more problems instead. Boinking his secretary just happened to be 'noticed' at the right time to get rid of him.

He was saying nice things about a potential partner. That's standard business practice.

Exactly. You are doing it wrong as an ODM if you aren't keeping discussions open with at least the top three semiconductor manufacturers in the world. One never knows who may have just the right process for performance, density or price at any given time. I suspect that IFS will start out competing on price so that they can develop a customer base.

 

[Exist50]

What's also interesting and makes me curious is what's going on with Intel's test chips for Qualcomm's design. Is it about the HD cells' leakage?

When developing a new process, it's not just yield that you have to tune for and slowly ramp up, but also performance. And actually those two are often inversely related for a given feature. So along the way, various PDK milestones come with performance and yield expectations, e.g. PDK 0.5 will be 85% of the final target performance at 1.0DD. Numbers made up, but you get the point.

This is very important for customers, because hitting these milestones not only shows how realistic the claimed targets are to be hit, but they also directly feed into design stability. Larger perf or yield gaps increase the risk/number of design-breaking changes, which causes a lot of extra work for the design partner. Historically, Intel could just tell its design teams to deal with it, but that doesn't fly in foundry.

 

[A///]

Only way they could manage that is by going bankrupt. There really wasn't much room below AMD's lowest point.

Some people earnestly believe Intel is on that path after having a relaxed approach due to AMD's poor performance for a decade, not being very competitive in mobile, losing in dc and being within margin of error in performance in desktop while using more power.

Have they? Must have missed it then.

Compare their quarterly from over a year ago to now. Different tone, acknowledging their competitor is racking on them big time.

He was saying nice things about a potential partner. That's standard business practice.

And those words carry weight. Those words boosted Intel's after hours trading. Instilling confidence in a platform as a third party does more than patting Gelsing's bum.

 

[A///]

Barrett held the ship steady, behaving more like his previous predecessors - though not having the same standing. Otellini just didn't get it, IMHO. Intel just kept getting larger and larger - adding more business units, etc. He also missed the boat on phones. Sure, that was hard to see back then, but it is such a historically bad decision. Krzanich - was supposedly the one who would come in and straighten out Intel's process development problems. He created more problems instead. Boinking his secretary just happened to be 'noticed' at the right time to get rid of him.

you know ajay for all random things you say in politics section from time to time you have a very insightful look into history. This is all correct and I've had a luncheon with two of these men listed here. Barett was also the last real engineer that steered intel, up until gelsinger was brought in. I've been pessimistic about Intel but I support gelsinger here, and I'm neither an Intel employee nor an investor. it wasn't the secretary alone, he got too political which raised the ire of intel folks.

 

[A///]

Exactly. You are doing it wrong as an ODM if you aren't keeping discussions open with at least the top three semiconductor manufacturers in the world. One never knows who may have just the right process for performance, density or price at any given time. I suspect that IFS will start out competing on price so that they can develop a customer base.

The words carry weight. There are ramifications not for nvidia but for intel if they were false statements and the board wants to drop management if ifs isn't as good. Jensen's words bumped intel's after hour trading. IFS could charge at just above cost and fund it through the internal products side. It'll be rough, but if the nodes are as good as rumors by some and jensen's opinion suggest then tsmc may be in for a world of hurt.

if the ifs nodes are slightly less efficient but cost much less than TSMC's 2 and 3 nodes then intel will sweep up. Last I looked the 3nm wafers by tsmc were 20k minimum. I don't know the break even cost for Intel but if they shot for 10% above that investors will overlook the lack of profit for the customer base. if tsmc has their cake taken from them they won't have the capital to expand and must ask the Taiwanese government for handouts. That's not cheap, and that's on the basis they get it right the first time.

interesting times ahead.

 

[Doug S]

if tsmc has their cake taken from them they won't have the capital to expand and must ask the Taiwanese government for handouts.

TSMC can't "have their cake taken from them". Intel doesn't have anything remotely like the capacity necessary to run the volume of wafers TSMC does on new nodes so even if Intel's nodes were better AND cheaper their effect on TSMC would be limited.

Intel would have to open up a lead, AND invest sufficiently through the end of the decade to build capacity necessary to take major customers like Apple, Qualcomm, and Nvidia (I'm assuming AMD wouldn't go to Intel unless hell froze over or Intel spun off the foundry business) before they could get TSMC to "where they won't have the capital to expand".

That's much more likely to be Intel's fate if they can't make IFS work. Doesn't matter how good their process is or how cheap their wafer pricing is if they don't understand the foundry business well enough to get those major customers on board. They failed miserably on their first attempt to operate a foundry, and the news so far on their second attempt doesn't sound that encouraging.

 

[A///]

TSMC can't "have their cake taken from them". Intel doesn't have anything remotely like the capacity necessary to run the volume of wafers TSMC does on new nodes so even if Intel's nodes were better AND cheaper their effect on TSMC would be limited.

tsmc can only output from what they get in. if they don't have the clients they don't have the income and if they don't have the income they can't keep the volume up. this is how things work. intel could undersell their own wafers and slowly leach clients as they build volume up.

Intel would have to open up a lead, AND invest sufficiently through the end of the decade to build capacity necessary to take major customers like Apple, Qualcomm, and Nvidia (I'm assuming AMD wouldn't go to Intel unless hell froze over or Intel spun off the foundry business) before they could get TSMC to "where they won't have the capital to expand".

amd going to intel is not an issue. what is intel going to do? copy their architecture and open themselves up to massive lawsuits? Be real here. If Nvidia go to intel for their consumer lineup next gen that's not good for tsmc. if it performs well, it may encourage others to dip their toes into team blue's waters. qualcomm is already open to such an idea. and my old employer broadcom also wants to try ifs out in the future. you're here talking in absolutes as if tsmc will forever remain king. intel would be smart to spin out the fabs like mobileye and not completely separate. Apple being the biggest tsmc customer has been one of their main cash cows for R&D.

 

[A///]

That's much more likely to be Intel's fate if they can't make IFS work. Doesn't matter how good their process is or how cheap their wafer pricing is if they don't understand the foundry business well enough to get those major customers on board. They failed miserably on their first attempt to operate a foundry, and the news so far on their second attempt doesn't sound that encouraging.

Intel's prior attempt failed because they played the our way or the highway game, they limited what partners could do and the heirarchy was a mess. IFS reports to gelsinger and only gelsinger, and in the future any other ceo. Ever since their failure they've been working with the fabless companies to understand the pain points caused the first time and to avoid those altogether in their current attempt. This isn't a one to one repeat of the first attempt. The only repeat here is the name.

the lack of confidence people are willing to give, just to see how it pans out is astonishing. before they've had their first set of wafers come out of production and into the eval and packaging phase people are already saying it's a waste of time based on rumors they've read online, whereas JHH has seen the preliminary results of their own hardware, which is what I gathered from JH's comments.

People seem to be more content with intel selling off its fabs and relying on TSMC to do everything for them because whoopty doo they messed up for a few years and are slowly recovering. None of you recall the six years intel spent in the back seat in the 90s and 2000s even with their bribing that only saw them get ahead with Core.

that last part. what news are you referring to? twitter rumors? partners working with intel who've made public statements? intel themselves? twitter rumors or youtube dipsticks like mlid have zero credibility no matter how many times they're correct. by the time they hear something it's gone through a half a dozen people. if jensen says intels' future nodes look good to him that's all that matters for now. people seem to give tsmc a pass despite tsmc jumbling 3nm for ages only to recently get it working in some order for apple and who knows the true yield rate for them on their modified 3nm.

 

[FlameTail]

Yes SD888 and SD8 Gen 1 use Samsung 5nm. But it is not correct to call it just a derivative, because it has some meaningful changes and improvements to density compared to their 7nm. And In density Samsung 5/4nm is comparable to TSMC's 5/4nm. In terms of efficiency is difficult to say, because as you mention when this SoC were made Samsung had terrible yields that are now corrected.
Also to mention Samsung 3nm should be comparable to TSMC 3nm in density and efficiency.

This is their latest information. But we don't know how a good yielding 4LPP fairs.

I still think TSMC 5nm/4nm is substantially better than Samsung 5nm/4nm in performance and efficiency (even if Samsung fixed their yields) because TSMC uses superior processes and materials (such as cobalt- which iirc Samsung hasn't used)

 

[moinmoin]

Those words boosted Intel's after hours trading.

Ain't that nice?

Personally I think stock trading is bad at accurately reflecting the actual worth of businesses. Intel still did put a lot of importance into it, with stock buybacks and dividends, at times when it should have spent into its business instead.

Only thing is, that didn't even help them. Current market valuation of select companies (feel free to add more):

• Apple: 2833 B
• Nvidia: 982.3 B
• TSMC: 512.6 B
• Samsung Electronics: 376.9 B
• AMD: 192.4 B
• Intel: 129.8 B
• Qualcomm: 129.0 B
• MediaTek: 39.0 B
• GlobalFoundries: 32,75 B