I think they’re also doing 5nm using DUV. Low yields but probably OK when you’re not on the hook to make chunks for nameless shareholders.
Comment on China's first 6nm GPU boots up, targets performance parity with RTX 4060
Buffalox@lemmy.world 1 week ago
WTF!? OK they had the 7nm process for more than a year now, but 6nm without EUV?
I’m not an expert, but it seems to me that when China cracks EUV, which they allegedly are getting close to, then SMIC will progress fast to even better processes.
China is making very good progress, especially considering they cannot use any western equipment. Everything has to be home grown!
avidamoeba@lemmy.ca 1 week ago
emergencyfood@sh.itjust.works 6 days ago
but 6nm without EUV?
I read elsewhere that 5nm was already done (in the lab, not on industrial levels), and they’re planning on 3nm by 2026.
The workaround seems to be to use particle accelerators instead of EUV machines. More expensive and with lower yield, but it’ll do the job.
Buffalox@lemmy.world 6 days ago
Yes 5nm should be ready this year, and allegedly Huawei is planning a 3nm product to be produced by SMIC already next year!
That’s a very bold schedule, but maybe they expect to have something close to ASML EUV soon?
golli@lemm.ee 1 week ago
Well there are claims that Huawei is aiming for 3nm with GAA with tape out next year See Here.
I think we shouldn’t forget that the nm numbers really are just that: Numbers. They don’t correspond to any specific measurements and can be chosen more or less arbitrarily. So 6nm for example might just be a slightly refined 7nm node.
Another thing is power efficiency and yields. If they get 4060 performance at terrible yields and with massive power draw then it is very different to getting there at similar parameters as Nvidia.
If China does end up cracking EUV by themselves it would indeed be massive. It’s arguably one of the most complex things mankind has ever done. But there are so many factors to get right that tbh I don’t see it happening any time soon.
Buffalox@lemmy.world 1 week ago
Isn’t GAA one of the most advanced types?
If they can really do this next year, there is no doubt China must have their own EUV lights.
Absolutely true but AFAIK SMIC usually make their numbers comparable to TSMC. Even Intel changed their numbering to follow TSMC, and SAMSUNG and TSMC have always been close in their numbering.
golli@lemm.ee 6 days ago
GAA is the next evolution of transistor architecture from FinFET, but as far as I know has no direct link to smaller process nodes. In that (to my understanding) it doesn’t require small nodes and could be used just as easily in larger ones. It’s just that it is more difficult so until now there were other easier ways to make progress. However with new nodes getting more expensive and giving less scaling gaa and other things like backside power delivery are being pursued.
So not only do you expect China to have a working domestically produced EUV machine within 5 years, but a competitive one? Or possibly even next year?
Next year is just pure fantasy that I don’t think even the most optimistic would assume. If they were anywhere close to that we would already know. They’d have shown a working prototype by now.
Euv is crazy difficult and you not only the result of a single company ASML, but many highly specialized companies that are leaders in their respective fields and all over the world like e.g. Zeiss for for the lenses. So for China to replicate it domestically they’d need to copy the whole supply chain. Which is orders of magnitude more difficult than what they’ve done in other industries like electric vehicles or solar panels.
Imo if they have a working prototype of a complete EUV machine within this decade it would already be impressive. But that would still be far off from mass production or wherever the industry is by then (Intel is currently trialing high na EUV). Also for reference Wikipedia says ASML had their first prototype in 2006 and we know how long it took to being that to mass production. China as a second mover might have an edge that speed things up, but just knowing how it works in theory isn’t enough and there are o shortcuts.
But maybe they also pursue another technique such as nano imprint (like canon) to achieve smaller nodes. Maybe that would be easier to replicate without existing global supply chains.
Buffalox@lemmy.world 6 days ago
AFAIK Canon has a process (nano imprint) that competes with EUV too, maybe not as advanced as ASML.
When I say China will have a competitive EUV system within 5 years, I don’t mean competitive in how advanced it is compared to the best ASML has to offer like the new NA process TSMC hasn’t even decided to buy yet.
But an EUV that allows a production process that will be competitive down to just below 2nm. Meaning competitive with the EUV process TSMC is currently using.
Depending on how much China is investing in this, they could be working on different processes in parallel.
Obviously anything as hard as this with a 5 year time span is very uncertain, we recently saw how bad things went for Intel with their 10nm process, being both multiple years delayed, and on top of that pretty bad when they finally announced it was ready.
As I stated next year would be very fast, but in reality we don’t know how close they are. Your claim that they would advertise it is speculation. What would be the purpose of that?
China is already competitive in 8 out of 10 key industries, and since China is prevented from even using western parts for chip making or even just buying the chips, they are pulling ressources to catch up. China will do this because they have massive talent mass and ressources, and because they have to.