Intel splashes more cash on ASML’s magic machines

[Daniel Nenni]

Questions: Whatever happened to the IFS separation.
Separate board
They should have a separate CEO
Separate finances?

There are some huge tradeoffs coming between product and foundry and they are not behaving independently. Have they announced they are not really independent.

I have not heard a word about it since Lib-Bu took the CEO job. I really do think Lip-Bu is looking to maximize Intel shareholder value. Why else would Softbank and Nvidia invest $7B?

It seems to me that would best be done when IFS is successful enough to be sold or go IPO. I'm not a finance person so this is above my paygrade but I did hear it discussed by people who would know.

When IFS starts talking about big 18A and 14A customers spinning it out will be much more viable. If IFS does not get big 18A and 14A customers it would be a moot point.

 

[tomatoma]

Last I heard Intel had two NXE:5000 HNA-EUV systems installed in the OR R&D center. Does anyone know different? I believe ASML has shipped 6 development systems in total thus far? Does anyone know how many EXE:5200 systems they will ship in 2026? From what I was told NXE:5000 HNA-EUV systems are pre production systems.

I know TSMC and Samsung both have one but I do not see them using HNA-EUV in HVM before 2030. It is not just the cost, it is scaling it for TSMC's definition of HVM, which is different than Samsung and Intel Foundry.

When will the memory makers have HNA-EUV in production?

Speaking of which, I think Intel has already said that it will introduce Hi-Na EUV mass production model.

 

[tomatoma]

Questions: Whatever happened to the IFS separation.
Separate board
They should have a separate CEO
Separate finances?

There are some huge tradeoffs coming between product and foundry and they are not behaving independently. Have they announced they are not really independent.

I think Naga Chandra, the CFO of IFS, was given a certain amount of discretion.

 

[tomatoma]

Last I heard Intel had two NXE:5000 HNA-EUV systems installed in the OR R&D center. Does anyone know different? I believe ASML has shipped 6 development systems in total thus far? Does anyone know how many EXE:5200 systems they will ship in 2026? From what I was told NXE:5000 HNA-EUV systems are pre production systems.

I know TSMC and Samsung both have one but I do not see them using HNA-EUV in HVM before 2030. It is not just the cost, it is scaling it for TSMC's definition of HVM, which is different than Samsung and Intel Foundry.

When will the memory makers have HNA-EUV in production?

I think someone from ASML said at an optical conference that the mass production model 5200 will start shipping in the second half of 2025...

 

[Fred Chen]

ASML Schedules Delivery of 56 Low-NA and 10 High-NA EUV Tools in 2027

ASML is about to manufacture significantly more Low-NA EUV and High-NA EUV machines than it originally anticipated. As you may be aware, acquiring an EUV scanner is a year-long process, during which companies make commitments that can span several years to procure these tools. According to ASML...

www.techpowerup.com

That report on SK hynix conflicts with this one: https://semiwiki.com/forum/threads/sk-hynix-builds-m15x-test-line-to-respond-to-hbm-demand.23690/

Historically, actual recognized sales has been lower than these forecasts.

 

[Barnsley]

EXE= High-NA
NXE=Low-NA
NXT=most advanced DUV (including KrF)

Thanks Fred, this will help me identify which masks we have are for what tools.

We have a good few EXE at the minute

 

[Fred Chen]

Thanks Frank , this will help me identify which masks we have are for what tools

The EUV and DUV masks should be recognizably different. The High-NA mask is probably not finalized yet.

 

[Barnsley]

The EUV and DUV masks should be recognizably different. The High-NA mask is probably not finalized yet.

Sorry for wrong name.

We have a few EXE masks on hand a minute.

 

[Fred Chen]

High-NA is expected to have a more complicated mask arrangement: https://euvlitho.com/2023/P11.pdf

 

[siliconbruh999]

The EUV and DUV masks should be recognizably different. The High-NA mask is probably not finalized yet.

Thanks Fred, this will help me identify which masks we have are for what tools.

We have a good few EXE at the minute

Fred became Frank here

 

[Daniel Nenni]

I think someone from ASML said at an optical conference that the mass production model 5200 will start shipping in the second half of 2025...

According to what I was told one 5200 system has already been shipped, to Intel hopefully.

Pat Gelsinger had said HNA-EUV will be coming in 14A. Lip-Bu Tan has not mentioned it from what I recall. Personally I think HNA-EUV will be used for 14A but it will be a small amount of layers just to say they did it. So I don't expect a big impact on the designs but yield learning for Intel will be good. Similar to what TSMC did at N7, N7+ and N6. They phased in EUV a few layers at a time.

 

[blueone]

Questions: Whatever happened to the IFS separation.
Separate board
They should have a separate CEO
Separate finances?

There are some huge tradeoffs coming between product and foundry and they are not behaving independently. Have they announced they are not really independent.

I don't think a separate BoD will accomplish anything. In the US, most BoDs are populated by useless, non-expert members. Yes-people. (As an aside, I just got a proxy vote for KLA yesterday. I almost voted against most of the board members up for election. I ended up not doing that, because it seemed futile, and they looked better than Nvidia's BoD members...)

I do think a separate CEO would be a good thing.

The separate finances part looks complex and doesn't smell good yet. Do you think that would work anytime soon?

 

[Daniel Nenni]

I don't think a separate BoD will accomplish anything. In the US, most BoDs are populated by useless, non-expert members. Yes-people. (As an aside, I just got a proxy vote for KLA yesterday. I almost voted against most of the board members up for election. I ended up not doing that, because it seemed futile, and they looked better than Nvidia's BoD members...)

I do think a separate CEO would be a good thing.

The separate finances part looks complex and doesn't smell good yet. Do you think that would work anytime soon?

I agree on the BoD, two is a waste of time and money.

IFS is separated as a business unit. Once it is successful it can be spun out for sale or IPO:

Yes Intel Should Go Private - Semiwiki

Lip-Bu Tan started as Intel CEO on March 18th of this year and some very impressive changes have already taken place. Intel started the year with more than 100,000 employees and will finish the year with around 75,000. Reporting structures have been flattened and the Intel culture is being...

semiwiki.com

 

[Fred Chen]

My understanding from BoA's count was that Intel only received 25 EUV systems during 5N4Y, and then got the 3 High-NA up to H1 this year. With so much less than what TSMC got, shouldn't expect to have the same level of EUV usage and I suppose that doesn't matter.

 

[OneEng]

Intel said 14A will ship in 2027 at the last event. That is the last I heard. Yesterday TSMC said A14 would ship in 2028. TSMC is very competitive so maybe that will move up. This is why we need Intel Foundry, competition pushes innovation.

Intel has been in the habit for quite some time now in over-promising and under-delivering. I haven't attended a class in business in quite some time, but I am pretty sure the best results are reportedly in the other direction .

But to seriously address the topic, it's still about the money, not the technology IMO.

18A cost more than a Ford class aircraft carrier. Each new generation of Lithography is costing exponentially more to get going, and is resulting in less and less improvement in PPA.

It's not just Intel that has been having trouble bringing on new nodes. It seems to me (and I am certainly not in the know like many of you guys are) that nodes going well is the exception, not the rule. TSMC appears to have the Midas touch as of late, but they appear to approach things much more conservatively than Intel.

Intel's pickle really is that they believe they can't get customers without a performance leading process node .... and they can't afford to keep pace with TSMC who has much higher volume to amortize the investment over.

This is just one old executives opinion, but it seems like Intel would be better off treating itself like the underdog it is in foundry services instead of pretending to be the industry leader (and spending like it too).

I get nauseated thinking about how much money 14A High NA might cost!

 

[siliconbruh999]

Intel has been in the habit for quite some time now in over-promising and under-delivering. I haven't attended a class in business in quite some time, but I am pretty sure the best results are reportedly in the other direction .

But to seriously address the topic, it's still about the money, not the technology IMO.

18A cost more than a Ford class aircraft carrier. Each new generation of Lithography is costing exponentially more to get going, and is resulting in less and less improvement in PPA.

It's not just Intel that has been having trouble bringing on new nodes. It seems to me (and I am certainly not in the know like many of you guys are) that nodes going well is the exception, not the rule. TSMC appears to have the Midas touch as of late, but they appear to approach things much more conservatively than Intel.

Intel's pickle really is that they believe they can't get customers without a performance leading process node .... and they can't afford to keep pace with TSMC who has much higher volume to amortize the investment over.

This is just one old executives opinion, but it seems like Intel would be better off treating itself like the underdog it is in foundry services instead of pretending to be the industry leader (and spending like it too).

I get nauseated thinking about how much money 14A High NA might cost!

Don't forget Intel's craze of the highest perf node got us to so many innovation in Nodes i would say let them cook.

 

[user nl]

According to what I was told one 5200 system has already been shipped, to Intel hopefully.

I hope as well!!

Let INTEL do some hard R&D work on High-NA EUV to figure out some problems with High-NA HVM. And let INTEL guide ASML on how to make these high-NA EUV-tools much more economical in HVM.

Let INTEL be helping ASML to transition these high-NA EUV-tools to a common module-based platform for low-NA and high-NA EUV (that's ASML's current strategy), so that these EUV-tools become more similar to build for low-NA and high-NA EUV versions. This will reduce the manufacturing costs for EUV-tools, and the implementation/fab costs for HVM with these novel high-NA EUV tools.

Both ASML and INTEL are somewhat in trouble and could be shaking hands because their advanced High-NA EUV tools and Leading Edge foundry-costs, respectively, are (way) too expensive currently. So, the two companies INTEL and ASML are both under quite some pressure.

Add the China progress on their DUV lithography development domestically, and their EUV R&D domestically (of course they prefer to keep rather quiet about both of these), and it seems that ASML may be loosing more China business in the next 10 years, than they did in the past 10 years. Unless the West opens up China lithography tool sales again for ASML, like they do perhaps for NVIDIA's and AMD's AI-chips, who knows ......

Furthermore, TSMC is way too busy with large-scale trusted ramping of 2 nm for all the global leading-edge customers and the revolution in 3D photonic-packaging. My gut feeling is that the photon-revolution in 3D large scale advance packaging will be much more of a determining factor the coming decade to reduce the energy-consumption of leading-edge HPC/AI in large connected networks of HPC-GPUs/CPUs.

High-NA EUV transistor shrinking will perhaps not be the key determining factor for leading-edge logic Foundries the coming 10 years is my feeling......

 

[Y.H]

I don't think a separate BoD will accomplish anything. In the US, most BoDs are populated by useless, non-expert members. Yes-people. (As an aside, I just got a proxy vote for KLA yesterday. I almost voted against most of the board members up for election. I ended up not doing that, because it seemed futile, and they looked better than Nvidia's BoD members...)

I do think a separate CEO would be a good thing.

The separate finances part looks complex and doesn't smell good yet. Do you think that would work anytime soon?

Frank Yeary is perhaps a reminder on the usefulness of BoD

 

[IanD]

According to what I was told one 5200 system has already been shipped, to Intel hopefully.

Pay Gelsinger had said HNA-EUV will be coming in 14A. Lip-Bu Tan has not mentioned it from what I recall. Personally I think HNA-EUV will be used for 14A but it will be a small amount of layers just to say they did it. So I don't expect a big impact on the designs but yield learning for Intel will be good. Similar to what TSMC did at N7, N7+ and N6. They phased in EUV a few layers at a time.

N7 was all-DUV, N7P had some EUV layers but incompatible design rules so no IP porting. IIRC (we didn't use it) N6 was a shrunk N7 (with compatible layout) so still all-DUV. N5 was the first "proper" EUV process, and then N4 was a shrunk N5.

 

[Fred Chen]

N7 was all-DUV, N7P had some EUV layers but incompatible design rules so no IP porting. IIRC (we didn't use it) N6 was a shrunk N7 (with compatible layout) so still all-DUV. N5 was the first "proper" EUV process, and then N4 was a shrunk N5.

I just checked TSMC's site now, it seems to be consistent with as you say:

https://www.tsmc.com/english/dedicatedFoundry/technology/logic/l_5nm

In 2020, TSMC became the first foundry to move 5nm FinFET (N5) technology into volume production and enabled customers’ innovations in smartphone and high-performance computing (HPC) applications. TSMC’s N5 technology is the Company’s second technology to use EUV lithography and achieved the same success as its predecessor, the N7+ process.