[karlgkk]

> If china succeeds in leapfrogging ASML

they wont. full stop. they've even admitted as much in industry. they'll be extremely happy if they're (only!) a few years behind tsmc

at this point, what will be a real earthshaker is if china manages to get past 7nm. smic has gotten a long way but even that company's 7nm process has serious limitations (much higher cost and worse yields)

and anyways, aside from a handful of use cases (ai being one tbh), 7nm chips are more than viable for any general purpose task. a "leapfrog" is quickly diminishing from a "need" to a "want", and the resulting governmental support is fading as well. of course, it'll still be a high priority for the chinese, but it's not like top of the list.

[fooker]

You are two years behind in your assessment, and technology moves pretty fast especially when you're trying to catch up.

You could buy phones with SMIC 7nm chips as early as mid 2024, that means yields were good enough around mid 2023.

This indicates they'd be on track to do 5nm this year, which is what the news articles indicate. The impressive part is that this is catching up with ASML+TSMC combined. There's no other company or government in the world that has achieved this vertical in the last few decades. China is willing to sink Manhattan project level resources into this, for good reason.

[karlgkk]

I am not two years out of date. Everything you said aligns with my assessment. Their 5nm yields are horrendous.

I guess my mistake was assuming they would ship it considering how rough it is financially, but they probably will for bragging rights.

> You could buy phones with SMIC 7nm chips as early as mid 2024, that means yields were good enough around mid 2023.

That was a halo product

> The impressive part is that this is catching up with ASML+TSMC combined.

Without euv, they are mining diminishing returns.

> There's no other company or government in the world that has achieved this vertical in the last few decades.

No, they had the advantage of copying and learning from decades of industry experience.

Don’t get me wrong, it’s impressive, but expecting Chinese leadership in this space by 2030 is foolish.

Of course, I’m never going to rule out anyone’s long term success, but there is no indication they’re in any position of leadership.

[fooker]

> Without euv, they are mining diminishing returns

Not sure why you are basing your whole argument on euv. There's nothing magical about it that prevents them from stealing/reinventing it.

[karlgkk]

> Not sure why you are basing your whole argument on euv.

I'm not. We were talking about "leapfrogging" and that naturally requires a technology that enables processes beyond EUV's capabilities.

In fact, many hours before you responded to my post, I responded to my own post with this: "My rough best guess is that they’ll be shipping euv-involved chips by 2028."

> There's nothing magical about it that prevents them from stealing/reinventing it.

If it was so easy they'd have done it already.

No, it's just insanely complicated to implement and the innovations required are not just the conceptual technology itself, but the high precision manufacturing prowess required to actually execute it.

And again, I think they're probably gonna have it before 2030. Hell, I could be convinced that they're going to start taping out EUV-based chips by the end of next year, although I would require a beer wager for that :)

But we were talking about leapfrogging, and that's "only" parity.

[karlgkk]

Oh and btw we’re talking about leapfrog.

My rough best guess is that they’ll be shipping euv-involved chips by 2028. That’s not a leapfrog, that’s parity

[fooker]

What would you say the chances ot leapfrogging are if China manages to invade Taiwan in 2028?

[karlgkk]

That's still not a leapfrog. That's just gaining parity lol

[corimaith]

Pretty sure they achieved that with DUV multipatterning, which isn't leapfrogging ASML at all.

[fooker]

If they keep pouring unlimited money at it and hiring the right people, I'd predict parity within two years.

[corimaith]

Instead of using vague abstractions like "pouring unlimited money" and "hiring the right people", I'd hope that predictions would be predicated more on the actual specifics of the progress being made, i.e what are specific engineering problems to be overcome and what is their progress in doing so. If it's not, it's really just astrology one is using.

The Chinese perspective itself certainly isn't anything close to these vague abstractions, outside of vague anti-western polemics or nationalist chest-beating. After all with the same logic we're pouring a "manhattan's worth of funding" into "AI", dosen't mean we're going to be reaching Gen-AI anytime soon!

[jacquesm]

Industrial espionage is a thing. China has so far managed to get every tech they ever wanted and I don't see why EUV could not be stolen. Everything, even very advanced technology, can be reverse engineered, especially if you already know conceptually what it does. Software and data (and blueprints) can be stolen as well. ASML and TSMC have a lot of security in place but at least on HN I would assume everybody knows that that does not guarantee perfection. If the knowledge is out there, it will spread.

[karlgkk]

> I don't see why EUV could not be stolen

> Everything, even very advanced technology, can be reverse engineered, especially if you already know conceptually what it does

This isn't true. Maybe for software it is. Manufacturing is one of the hardest stages, and China lacks tooling for the ultra-high precision engineering required to actually implement this process.

It's kind of like building a nuclear bomb - conceptually it's easy. Hell, the first nuke was dropped before the microwave oven was commercially available.

The real challenge is manufacturing the damn thing. Refining all of that uranium is not an easy task - even today.

China has spent billions of USD equivalent trying to copy EUV. They have had access to EUV installations and fully disassembled them (funny story: they broke it putting it back together).

They are highly motivated, they have a ton of money, and frankly, they're no a bunch of dummies.

And yet, they still don't have it. (fwiw i think they will by the end of 2030)

[fooker]

> dosen't mean we're going to be reaching Gen-AI anytime soon!

I disagree. We already have, by anyone's standards from 2021.

We keep shifting the bar, somewhat intentionally so that progress doesn't stall.