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

 

[Geddagod]

It's possible that 18A-P is different enough from 18A that it has poor yields and needs work. I'd even go so far to say that's the only plausible answer for why they'd do one single 4P 0E chip on 18A-P when 18A already had much bigger chips.

I would be shocked if there is. There is no density uplift between the two nodes according to Intel, and the perf/watt uplift of 8% is good, but not ground breaking.

 

[511]

They are selling the test wafers. That's what Panther/Nova? is.

no? that's ridiculous lol test wafer have very bad yields for upcoming nodes

 

[Win2012R2]

They are selling the test wafers.

To whom - Intel's product group?

 

[511]

I would be shocked if there is. There is no density uplift between the two nodes according to Intel, and the perf/watt uplift of 8% is good, but not ground breaking.

that's the same as N2->A16 but without any new BSPDN

 

[jpiniero]

no? that's ridiculous lol test wafer have very bad yields

Not a problem. Sell what you get from it.

Surely part of this is a scheme to convince "Investors" that the Foundry is doing well.

 

[511]

Surely part of this is a scheme to convince "Investors" that the Foundry is doing well.

their yield is not bad as reuturs makes out to be also foundry is not doing well due to Intel advanced accounting and them playing pin pong in internal balance sheet

 

[DKR]

Mercury says they gained more server market share in Q1 2025 alone (2.1%) than they did all of 2024 (1.5%).

Now that my X posts are all over this conversation. Here is a good one imo about these so called % gains of AMD over Intel in Data center.
TLDR - Intel lost more of the server CPU volume due to AI sucking the cloud & enterprise capex spend than AMD winning. This is not to say AMD is not gaining but their share gains are being overexaggerated by %.

https://twitter.com/x/status/1948442296556683744

I also have some charts on how fast the server CPUs ramp. TLDR - Usually 10+Qtrs for Intel but much faster for AMD, my guess is mainly due to Intel's bias towards Enterprise more

https://twitter.com/x/status/1948442336205119622

 

[DKR]

And the reason is because they need N2 to hope to be competitive, not because of any artificial capacity constrains on 18A.

I agree. I think Intel needs N2P to be competitive with AMD this time. AMD has moved to N2 and that changed the equation for Intel and they changed the 4+8 from 18A-P to N2. Just as Pat was retired, the CO-CEOs started talking about having a significant % of wafers outsourced to TSMC to keep IFS on its toes. If I am not mistaken that is about the time, AMD's Zen 6 taped out at TSM. Maybe that led them to change it. I highly doubt 18A-P is having any more yield issues than 18A on PTL which has a 4+8 die. Just my guess, of course.

 

[DKR]

Another more charitable take for Intel would be 18A is capacity constrained and Intel is focusing to use the capacity for their external customers like DoD, Amazon & Microsoft. It is the right thing to do imo to prioritize external customers than Intel products if Foundry is to succeed.

 

[Doug S]

that's the same as N2->A16 but without any new BSPDNView attachment 128663

The TSMC slide you've included says these numbers are compared to N2P. So you'd have to combine the N2->N2P numbers with these to calculate the uplift of N2->A16.

 

[jpiniero]

Now that my X posts are all over this conversation. Here is a good one imo about these so called % gains of AMD over Intel in Data center.

Which is crazy given how noncompetitive Intel Server is. People buying Cascade Lake still is even crazier.

 

[DZero]

Now that my X posts are all over this conversation. Here is a good one imo about these so called % gains of AMD over Intel in Data center.
TLDR - Intel lost more of the server CPU volume due to AI sucking the cloud & enterprise capex spend than AMD winning. This is not to say AMD is not gaining but their share gains are being overexaggerated by %.

https://twitter.com/x/status/1948442296556683744

I also have some charts on how fast the server CPUs ramp. TLDR - Usually 10+Qtrs for Intel but much faster for AMD, my guess is mainly due to Intel's bias towards Enterprise more

https://twitter.com/x/status/1948442336205119622

This is really bad, look at the graphics.


By 2020 have 35% in total, but by 2024 only have 22% at best... where the 13% remaining went? ARM? RISC-V?

 

[Doug S]

This is really bad, look at the graphics.

View attachment 128665
By 2020 have 35% in total, but by 2024 only have 22% at best... where the 13% remaining went? ARM? RISC-V?

Check that Y axis again. That's not percentage, it is in millions of units. You're asking where 13 million units went. Presumably some of it went ARM (stuff like Graviton that hyperscalers designed for themselves) but some of it is as the tweet says displacement of investment by AI. i.e. companies investing in Nvidia GPUs not any sort of CPUs (not directly, but via the cloud)

 

[DKR]

This is really bad, look at the graphics.

View attachment 128665
By 2020 have 35% in total, but by 2024 only have 22% at best... where the 13% remaining went? ARM? RISC-V?

Some went to ARM through Hyperscaler internal silicon program (ARM is still only <7% of total server volume- likely increases with all the GB200 shipping this year.) But major impact is many Hyperscalers like Google, Amazon extended their useful life of their servers around COVID time, likely the same happened in Enterprise too. Also emergence of AI at the end of 2022 (ChatGPT) sucked the life out of capex spend on traditional servers. This is why total volume decreased but Intel got hit pretty bad compared to AMD. AMD having better product started gaining share but its only modest. Obviously when you calculate % of shares, if the total volume comes down, even small gains will get magnified.

AMD capitalized on this and started forwarding this Mercury research reports on time to every industry analysts as well.

https://twitter.com/x/status/1923072060613439640

 

[Saylick]

So it's been rumored now that the government is considering taking a stake in nationalizing Intel's fabs. How does this work from a shareholder's perspective? Do shareholders get a one-time dividend for the portion of Intel that got bought out? Does the market cap of Intel (the design side) get adjusted upwards or downwards from having IFS be spun out?

 

[DZero]

Check that Y axis again. That's not percentage, it is in millions of units. You're asking where 13 million units went. Presumably some of it went ARM (stuff like Graviton that hyperscalers designed for themselves) but some of it is as the tweet says displacement of investment by AI. i.e. companies investing in Nvidia GPUs not any sort of CPUs (not directly, but via the cloud)

Thanks for the correction, but.,.. isn't it worse considering that both are slowing down their sellings then?

 

[Josh128]

So it's been rumored now that the government is considering taking a stake in nationalizing Intel's fabs. How does this work from a shareholder's perspective? Do shareholders get a one-time dividend for the portion of Intel that got bought out? Does the market cap of Intel (the design side) get adjusted upwards or downwards from having IFS be spun out?

Usually, shareholders dont make out so great during these prop ups. Sometimes there are exceptions. What sucks is, it likely gives Intel an unfair advantage in chip design, against competitors like AMD. It will be interesting to see how the goverment navigates and /or separates the foundries from chip design.

This is the Grok response to the question: how did the Bank of America goverment bailout affect shareholders that held stock at that time? I got two different responses.

The 2008-2009 TARP bailout provided Bank of America $45B in capital injections, plus asset guarantees. Shareholders faced severe stock price drops (70% in 2008, to ~$3 low in 2009), dilution from government warrants convertible to common shares, and dividend cuts to $0.01/quarter for 3 years. Repayment in 2009 via new stock issuance further diluted holdings. Lawsuits over nondisclosure of Merrill Lynch losses yielded $2.43B settlement. Overall, holders suffered major losses during the crisis.

• The 2009 Bank of America bailout, involving a $20 billion injection and a $100 billion toxic asset guarantee, stemmed from the 2007-2008 financial crisis, triggered by the subprime mortgage collapse, with Merrill Lynch’s deteriorating balance sheet adding pressure; a Reuters report from January 16, 2009, highlights how this made Bank of America the largest bailout recipient after Citigroup.
• Shareholders faced immediate dilution as the government’s rescue required a dividend cut from 32 cents to 1 cent per share and imposed executive pay caps, with a study from the Journal of Financial Economics (2011) showing that such bailouts often reduced shareholder value by 20-30% due to government equity stakes and restrictions.
• Contrary to mainstream views of bailouts as pure rescues, evidence suggests they shifted risk to taxpayers while preserving executive control, with Bank of America’s CEO retaining power despite the crisis, challenging the narrative of accountability, as noted in a critical analysis by the National Bureau of Economic Research (2010).

 

[johnsonwax]

So it's been rumored now that the government is considering taking a stake in nationalizing Intel's fabs. How does this work from a shareholder's perspective? Do shareholders get a one-time dividend for the portion of Intel that got bought out? Does the market cap of Intel (the design side) get adjusted upwards or downwards from having IFS be spun out?

If it's a split and government stake, the shareholders will get proportional stake in the two resulting companies, the design side will see a corresponding drop in market cap due to the loss of shares, and the market will re-evaluate the value of just the design company (could go up or down). On the foundry side they'll mostly likely issue new shares, dilute the shit out of the existing shareholders, and they'll just take a bath in return for the company getting effectively bailed out. Shareholders shouldn't expect to get to both run up the stock on the bailout news and then demand to be made whole for the amount they just ran the stock up as the new shares are issued.

My guess is that the feds take an equity stake and not a loan/etc. They've been doing this in other areas of late and it's pretty in keeping with how Trump does things (even though it's the antithesis of what neoliberal Dem/GOP would think is acceptable). Democrats are going to have to eat sh*t for a while for their failure to embrace interventionist economic policy sooner and cede that space to Trump.

 

[johnsonwax]

Usually, shareholders dont make out so great during these prop ups. Sometimes there are exceptions. What sucks is, it likely gives Intel an unfair advantage in chip design, against competitors like AMD. It will be interesting to see how the goverment navigates and /or separates the foundries from chip design.

I don't see how that's the case. The feds are going to prop up the foundry business, not the design one. This seems to presume the company splits before the equity stake is taken. If the feds invest in the foundry business, it becomes equally available to Intel (design) and AMD.

 

[Josh128]

I don't see how that's the case. The feds are going to prop up the foundry business, not the design one. This seems to presume the company splits before the equity stake is taken. If the feds invest in the foundry business, it becomes equally available to Intel (design) and AMD.

This is how it will likely go down, and if the gov injects 20-50 billion into them, you damn well better get ready for Trump to have some immediate demands in what steps they take next. Who knows how it will all play out.

 

[Saylick]

If it's a split and government stake, the shareholders will get proportional stake in the two resulting companies, the design side will see a corresponding drop in market cap due to the loss of shares, and the market will re-evaluate the value of just the design company (could go up or down). On the foundry side they'll mostly likely issue new shares, dilute the shit out of the existing shareholders, and they'll just take a bath in return for the company getting effectively bailed out. Shareholders shouldn't expect to get to both run up the stock on the bailout news and then demand to be made whole for the amount they just ran the stock up as the new shares are issued.

My guess is that the feds take an equity stake and not a loan/etc. They've been doing this in other areas of late and it's pretty in keeping with how Trump does things (even though it's the antithesis of what neoliberal Dem/GOP would think is acceptable). Democrats are going to have to eat sh*t for a while for their failure to embrace interventionist economic policy sooner and cede that space to Trump.

Thanks for weighing in. There’s a handful of Xitter people I follow who happen to be bagholders of Intel stock and they’ve been praying for a split of IFS and the design side, mainly because they believe a split will unlock value on the design side and allow them to recover their paper losses. I never really understood how that could work since it was my belief that their shares of the foundry side would offset any gains from the design side.

 

[Josh128]

*duplicate post removed.

 

[poke01]

The 18-A reference non-x86 SoC is based on ARM.

 

[DZero]

The 18-A reference non-x86 SoC is based on ARM.

If Intel didn't abandoned ARM in their moment... now would be on another light.

 

[DavidC1]

If Intel didn't abandoned ARM in their moment... now would be on another light.

Intel's faults are Intel's, not their ISA if that's what you are implying. Their StrongARM division was executing badly.

If x86 was open and not artificially protected by corrupt lawyers that couldn't see beyond money, then Intel(and AMD) would have real competition. I bet you Apple would make a comparably efficient SoC.