5 ms·
The best China has is an internationally uncompetitive "7nm" fab and that's the best they'll have until they can manufacture EUV machines domestically. So the
by KK7NIL 11mo ago
The best China has is an internationally uncompetitive "7nm" fab and that's the best they'll have until they can manufacture EUV machines domestically.
So the EUV blockade has absolutely been effective and the fact that the PRC is paying so many shills to convince westerners otherwise just shows how behind they are.
- TrainedMonkey 11mo agoI noticed that people love pointing how far AI field has advanced in a few years and extrapolate next few years. While at the same time being dismissive of Chinese semiconductor manufacturing process. In similar vein I also remember claims that TSMC Fab in Arizona can never work, and yet it does. So I don't know man, I wouldn't underestimate what a billion of enterprising people can do. Especially when paired with the system that has a pipeline of funneling smart people into elite schools.
- op00to 11mo agoUnderestimating China seems like a really, really, really stupid thing to do.
- 11101010001100 11mo agoYes, we are doing a bad job of updating our priors.
- illiac786 11mo agoIs that sarcastic? Isn’t underestimating by definition a bad thing?
- alwa 11mo agoTaking this as an earnest question—no, I don’t get that sense from that word. To me it describes the direction of an error, not the error itself. It’s a thing you’d prefer to avoid, sure; but some degree of prognostic uncertainty is totally routine (in fact I would call that definitional: no predictions are truly certain until they’ve come to pass, and by the time that happens it’s usually too late to act). It’s not “bad” any more than mortality is “bad”—it just is, whether or not we wish it were; wisdom lies in managing it as best you can. In the sense that the gp used the word, I think they allude to a tradeoff: you can reduce the probability of an underestimate by increasing the probability of an overestimate. I took their comment to imply that it would be wiser to risk an overestimate than to risk an underestimate on questions of “can Chinese society achieve a massive goal on a tight timeframe if their leadership decides it’s important.”
- illiac786 11mo agoNo I get what the GP meant. Your comment sounded like a triviality from Lapalisse a bit, because I cannot think of any occurrence where underestimating something is a good thing. Bit like “15 min before my death I will be alive”. But Lapalisse too didn’t mean it literally, he just wanted it to sound like that, it seems it’s what you did. Much better than sarcasm then =)
- thfuran 11mo agoIt's definitionally non-ideal, but not definitionally really, really stupid.
- bell-cot 11mo agoPerhaps the USA feels that it has a reputation to downhold?
- tracker1 11mo agoI don't think the US is underestimating China... I do think that the US is preemptively shoring up a domestic posture against long term changes. It would be a pretty bad strategy to continue to outsource everything and continue to see a massive trade imbalance with the outside world for a prolonged period of time.
- judahmeek 11mo ago> It would be a pretty bad strategy to continue to outsource everything and continue to see a massive trade imbalance with the outside world for a prolonged period of time. It's not actually a strategy at all. It's the organic result of being the global reserve currency. Foreigners want American dollars so that they can trade with everyone else and are incentivized to do whatever it takes to get it. Also, the "massive trade imbalance" is only an imbalance in goods. When you take services & the flow of foreign investments/loans into consideration as well, things don't look anywhere near as uneven as Donald Trump would like you to believe.
- seanmcdirmid 11mo agoIt’s not even just the flow of services Trump is ignoring: an iPhone is made in China but the design and software is done in the USA, most of the parts come from other countries, most of an iPhone’s value isn’t originating from China. Trump wants us to give up high value jobs in designing hardware and software so we can make less working in factories again.
- tracker1 11mo agoNo... not everyone is capable of doing hardware or software design... There are over 350 million people in the US. There are nowhere near that many software/hardware jobs.. or other IP generating jobs. Not only that, but corporations are bringing in a lot of people to do those jobs as well, driving down those wages. It's not even just jobs, it's also the tax revenue itself.. the population is overburdened with taxes and increased prices combined with relatively lower wages due to excessive inflation the past few years. While tariffs can increase prices, they can also eat into the margins of foreign production leading to more insourcing of jobs. Beyond those aspects is being able to handle production of critical infrastructure in times of supply constraints... such as war or a global pandemic. You can increase from 50% production of medications to 100% of domestic needs pretty easily, but scaling from 0% is almost impossible in any reasonable time frame.
- NedF 11mo ago[dead]
- gmerc 11mo agoThey can just throw power at it, you're delusional if you think it's going to hamper them even mid term.
- beej71 11mo ago> that's the best they'll have until they can manufacture EUV machines domestically. And how far out is that?
- stickfigure 11mo agoSo far only one company in the world has successfully accomplished it, so the answer could be "a very very long time".
- KK7NIL 11mo agoIf you ask PRC shills, it's just around the corner because this one Chinese lab demonstrated a very small part of the system. And a surprising number of westerners fall for that crap. My guess is that it's at least 10 years away, but that could obviously change depending on what resources they're willing to commit. But even at that point they'll be 2 decades behind ASML's EUV tech so it probably won't be competitive.
- buran77 11mo ago> If you ask PRC shills GP must have been asking for the non-PRC shill opinion. > My guess is that it's at least 10 years away, That doesn't sound at all like a lot. China has a uniquely effective industrial espionage... industry, combined with a very thick geopolitical skin and disregard for international demands. This helps accelerate any process that others have already perfected. We'll start to see the real deal if/when China eventually catches up to the leaders in every field and the only way to pull ahead is to be entirely self propelled (you can't take advantage of someone else's draft when you're in front of the pack).
- tracker1 11mo agoI think you may underestimate the ability of China to abuse industrial espionage at scale.
- ruszki 11mo agoThere are things which needs time, even with all or almost all the information at hand, just like with atomic bomb. I’m not sure whether this case similar to that, but that ASML in front for so much time indicates that their moot is probably not just information.
- kakacik 11mo agoApart from gaming and llms, most of the chip applications including all of military and consumer electronics is more than happy with 7nm process, whatever that means (proper nanometers those ain't). I know some people live in the IT bubble and measure whole reality by it, but that's not so much true for the world out there. They have ie roughly F-35 equivalent, minus some secret sauces (which may not be so secret at the end since it seems they stole all of it). You are making a mistake of thinking of them as yet another russia, utterly corrupt, dysfunctional at every level and living off some 'glorious past', when reality is exactly the opposite.
- immibis 11mo agoOkay? There's a lot of chips you can make that aren't the cutting edge. You don't need a 4090 to do AI, as evidenced by all the AI we did before the 4090. You definitely don't need a (random Intel chip) 14900HX to do general-purpose computing, as evidenced by all the general-purpose computing we did before the 14900HX.
- tracker1 11mo agoFor that matter, the 14900hx was already based on a refined 7nm production process, which China already has started using, though maybe not as effectively yet. As you mention, prior to the 4090's 3090 was on an 8nm node, already behind current China capabilities.
- KK7NIL 11mo agoIf each node provides a 10-15% improvement in power, performance and area, how many of those need to compound until your already uncompetitive 7 nm is 10x less efficient, slower and more expensive?
- immibis 11mo agoBeing behind doesn't mean they're permanently stuck where they are today - but aren't our processes running into the wall of soon trying to make transistors smaller than an atom?
- KK7NIL 11mo ago> Being behind doesn't mean they're permanently stuck where they are today Without EUV, they very much are. > but aren't our processes running into the wall of soon trying to make transistors smaller than an atom? No, the finest pitches are still in the low double digit nanometers in 2 nm processes. The "2 nm" nomenclature hasn't denoted a physical dimension for decades.
- adrian_b 11mo agoOne should remember that EUV is necessary only for obtaining profit from the mass production of integrated circuits. For making a limited quantity of chips, for research purposes or for some special applications where the price is irrelevant, there would be no problem for China to make today ICs with e.g. a 2-nm CMOS process, by using electron-beam lithography. (Obviously, for developing a 2-nm process many other problems must be solved first, but lithography is not a roadblock, so the process can be developed before having EUV lithography, because test wafers can be made with e-beam lithography.) Moreover, they have enough money and people to ensure that an alternative EUV technology will be developed, eventually. I might take them 5 to 10 years, but not more than that. The attempts to sabotage China should have been started more than a decade earlier in order to have chances of success. Now it is too late and the cleverer way would have been to try to accelerate progress in USA, instead of trying to hinder progress in China, by using means that have totally discredited USA as a product supplier all over the world (i.e. by using the dubious legal theory that USA can dictate what to do to the owners of products that include components "made or designed in USA").
- ayewo 11mo agoYou are ignoring the possibility of technological disruption. Apple disrupted Nokia and Blackberry. ARM is currently disrupting Intel. What if someone lands on a break-through using a completely different tech: what if X-ray lithography [1] becomes viable enough that they don’t have to acquire state-of-art EUV machines from ASML? [1] X-ray lithography was abandoned in the 80s but it is being revisited by Substrate https://substrate.com/our-purpose https://substrate.com/our-purpose. They are an American company that hopes to make it commercially viable by being cheaper and far less complex than EUV.
- impossiblefork 11mo agoI think you're general point is completely true, but Substrate is a bad example, since the people running it don't appear to be semiconductor experts and it's probably a fraud.
- KK7NIL 11mo agoSubstrate is a scam; their marketing is misleading and they have yet to answer to the fundamental reason why X-ray and e-beam failed over 40 years ago (despite it being generally agreed they were the future of litho and optical would soon be dead): writing one line at a time is extremely slow compared to optical which can scan a whole reticle in a fraction of a second. E-beam is still used for making DUV/EUV masks where the low write speed can be tolerated but no one in the industry thinks it will replace EUV in the silicon litho steps any time soon. But lay people eat this crap up and journalists turn a blind eye either because they're literally paid PRC shills or because clicks are everything now a days.
- einpoklum 11mo agoSo, you're saying that China has chip fabrication capabilities which are on par with the world cutting edge as of 2018: https://en.wikipedia.org/wiki/7_nm_process https://en.wikipedia.org/wiki/7_nm_process not too shabby of a fall-back.
- KK7NIL 11mo agoNo, they don't. Their "7 nm" relied on multi patterning DUV which leads to restrictive design rules, more steps and masks and lower yields, which is why I put it in quotes and said it's uncompetitive. The last DUV node was 10 nm, that's the best logic node they have which is comparable to TSMC/Samsung/Intel's 10 nm.
- pyuser583 11mo agoMy understanding, which is not complete, is China has done some amazing things optimizing training on slower chips. Which is cool, but there are limits to the number of times you can do that. At the end of the day, the little man has to flip the switch.
- hearsathought 11mo ago> the fact that the PRC is paying so many shills to convince westerners otherwise just shows how behind they are. And yet, it's anti-PRC shills that are all over social media. Go figure.