Taiwanese media: TSMC will get its 1st High-NA EUV this month

Posted by k0ug0usei@reddit | hardware | View on Reddit | 28 comments

https://ec.ltn.com.tw/article/breakingnews/4794855

Translation of key points:

TSMC will get its 1st High NA EUV in September, company is finalizing transportation plan.
TSMC negotiated with ASML to get 15% discount.
Will only be introduced in 16A process the earliest.

[-]

monocasa@reddit

So all of those stories about how Intel bought all of the High NA machines for 2024 were BS.

https://www.reddit.com/r/intel/comments/1codf6v/report_intel_bought_all_of_asmls_highna_euv/

[-]

theQuandary@reddit

Everything about this report seems weird.

ASML shipped the first EUV prototype (3100) in 2010 and shipped second-gen 3300 EUV by 2013. TSMC and Samsung weren't using EUV in non-risk production until 2018-2019. It takes time to develop new processes and even changes from EUV to high-NA EUV will have repercussions for the process that take time to find and solve.

ASML said that Intel was shipped the first machine in Dec 2023 (it went to Intel's lab in Oregon) and Intel announced they had set it up by April 2024. If TSMC got their first machine in September, they probably don't have it operational until Jan 2025 to begin finding out how to use it with their manufacturing process. That's an 8-month lead in R&D on how to use these machines with their process efficiently.

TSMC getting one machine for research is also no indication that they are getting more machines any time soon. Even if Intel otherwise bought up all the capacity, it's not unlikely that both Samsung and TSMC would be allowed to buy machines for R&D purposes. Not allowing this would not only tick off ASML's customers, but would likely wind up involving Korean and Taiwanese governments lodging complaints about anti-competitive behavior.

ASML high-NA machine production supposedly 5-6 per year at present (ASML shipped around 140 machines in the decade from 2012-2022, so this number isn't surprising). Intel has talked about possibly using these for 14a in 2026, but it seems to be really pushing the limits. TSMC hitting the same schedule for their A16 process with 8 months less time seems even more absurd.

[-]

Strazdas1@reddit

Inten intends to use HIGH-EUV machines in 2027. TSMC intends to use them in 2030. This isnt for current products.

[-]

theQuandary@reddit

That's why I said the report is weird. Here's a Google-translated bit of the article

Greet Storms, vice president of product management at ASML, pointed out last week that the company’s current EUV customers have already placed orders for High NA EUV equipment. The company will work closely with customers and hopes to move towards mass production in 2026.

I don't think I'm misreading the article's assertion.

As to Intel, we have them stating that they plan to use high-NA in 14a in 2026 according to this slide from Feb, 2024

https://www.intel.com/content/dam/www/public/us/en/newsroom/posts/intel-foundry-node-roadmap-2024.png.rendition.intel.web.1920.1080.png

(here's the article)

I believe they won't be using it until at least 14a "Enhanced" (I guess they didn't want to call it 14a++) and may be forced to wait even longer.

[-]

Strazdas1@reddit

14A is 2027 as per Intels last time i saw it. Your image seems to indicate the same.

I dont know if Intel will make it work by 2027 or not, but its not like they are behind others. The only competition for high-NA has their schedule 3 years later so if Intel takes till 2028 to make it work thats still 2 years ahead of TSMC.

[-]

Kryohi@reddit

Pretty much every story about a company buying "all" of something is BS.

It was for Apple buying all N3B capacity, it is for Intel buying all high-NA machines and many other examples. They might be the only ones buying something at a certain time, but only because competitors have different plans or timelines.

[-]

monocasa@reddit

Apple's is generally true. They have a unique relationship with TSMC.

[-]

Kryohi@reddit

They likely often have some privileges around volumes, timing and contracts, but in the case of N3B most competitors simply waited for the actually cost-effective N3 node, i.e. N3E.

[-]

monocasa@reddit

I mean, Lunar Lake is N3B too.

[-]

Exist50@reddit

LNL was intended to arrive in '23, hence the node choice.

[-]

DaBIGmeow888@reddit

It was a nice spin by Intel though before their stock cratered.

[-]

Jeffy299@reddit

Could be testing machine to iron out the issues before the real production with the High-NA machines starts.

[-]

Legal-Insurance-8291@reddit

Of course it's a test machine, same with the one Intel has installed in Oregon.

[-]

lusuroculadestec@reddit

or Pat Gelsinger called Christophe Fouquet and backed out of buying a few in the past couple of months.

[-]

Vushivushi@reddit

ASML's forecasts were way too weak to have sold out.

[-]

MixtureBackground612@reddit

Intel like to waste money on marketing instead

[-]

k0ug0usei@reddit (OP)

Or maybe Intel cut back a bit for cost saving.

[-]

Oockland@reddit

Sodium is coming to chips

[-]

PastaPandaSimon@reddit

It'll be now added to wafers

[-]

champignax@reddit

Sweet!

[-]

NerdProcrastinating@reddit

I wonder, how much of a premium high NA will add to the wafer cost?

Compensating for the extra cycles needed from a halving of the reticle size must be a challenge. I suppose it will only make sense to use for processes that would otherwise require multi-patterning of regular NA EUV.

[-]

CEO: but i told you to buy it