Exclusive-How China built its ‘Manhattan Project’ to rival the West in AI chips

[Fred Chen]

Also true that Moore's law no longer applies to transistors. We now look at the whole chip (PPA) packaging and all.

Isn't this called "More than Moore"?

 

[Barnsley]

My Linkedin is full of this story

 

[count]

I have been saying on this forum from some time that China would catch up on EUV and the industry seconds source will not be Intel but China. I think it’s just a matter of time.

 

[Fred Chen]

the industry seconds source will not be Intel but China. I think it’s just a matter of time.

They'd need the gate all around nano-sheet and buried power rails. Must be the truly secret development.

 

[Fred Chen]

I had been wondering for a while how China could possibly get ASML EUV parts through sales. Now it seems there might be an answer: https://news.eeworld.com.cn/manufacture/ic511949.html

"We know that GlobalFoundries had two early EUV machines, one of which was installed, but ultimately sold both when it decided not to pursue the leading 7nm process. One of these machines was ordered by a domestic company, but as far as we know, it was not installed due to restrictions imposed by the US."

The Reuters story is still made bogus by this news, there's no need for second-hand markets or DUV parts from Nikon or Canon.

It does beg the question of whether the parts got to shipped to China and were just lying around in crates. Presumably, ASML should have reclaimed them.

 

[ashwinbrao1]

I believe that we are making a mistake by focusing on when China catches up with the "best in class" technology available in the west. Instead, we should be looking at the minimum viable performance required for China to be able to manufacture chips at a rate that enables whatever application demands (in defense, strategic AI applications) deemed to be critical by the Chinese establishment.

 

[count]

I had been wondering for a while how China could possibly get ASML EUV parts through sales. Now it seems there might be an answer: https://news.eeworld.com.cn/manufacture/ic511949.html

"We know that GlobalFoundries had two early EUV machines, one of which was installed, but ultimately sold both when it decided not to pursue the leading 7nm process. One of these machines was ordered by a domestic company, but as far as we know, it was not installed due to restrictions imposed by the US."

The Reuters story is still made bogus by this news, there's no need for second-hand markets or DUV parts from Nikon or Canon.

It does beg the question of whether the parts got to shipped to China and were just lying around in crates. Presumably, ASML should have reclaimed them.

I think it's really impossible to prevent China from getting their hands on the technology. At this point EUV has been commercial for a decade - there are enough current and former ASML engineers out there as well as engineers from TSMC who could be attracted by the right offer to go to China and support development (and bring stolen IP with them). There are enough machines and spare parts out there for to support a small black market for parts. China is determined and has the resources to pursue this, and at best you can slow them down.

Yes there are other hurdles they will face, but it's a mistake to assume they won't be able to surmount them.

 

[count]

I believe that we are making a mistake by focusing on when China catches up with the "best in class" technology available in the west. Instead, we should be looking at the minimum viable performance required for China to be able to manufacture chips at a rate that enables whatever application demands (in defense, strategic AI applications) deemed to be critical by the Chinese establishment.

100% agree with this take. China doesn't need to be leading edge. If they get to N-1 or even N-2, it can also break the economics of leading edge foundries who need to depreciate their investments over increasingly long timelines.