Companies have to pare back their plans because they don’t have reliable access to smaller nodes. Your plans can’t get ahead of production capabilities. Those would be meaningless plans. That doesn’t mean they don’t want to get to those advanced nodes ASAP.
Chinese semiconductor thread II
- Thread starter vincent
- Start date
Chinese companies don't even have access to smaller nodes until 2023, when SMIC released China's first commercial 7nm chip. in fact SMIC still increasing the production capacity of 7nm.
before that, no sane company could even dare to design chip for advance nodes coz of fear of sanctions. it wasn't feasible.
In 2024, China accounted for 34.1% of global automobile sales.. when your home market is 1/3 of entire global market you barely need foreign sales.. Xiaomi doesn't even need to expand overseas. first fulfil the demand at home.. they have exceed sales in 2024 and planning to produce 300,000 cars this year.
this has to be a joke.. right
the demand of advance nodes is too sacred in China that SMIC is unable to fulfil the Huawei's demand alone. let alone to supply entire market.. from Huawei to Oppo to Vivo and Xiaomi and many more all will jump to SMIC 5nm/3nm once chip is available. home made chip is now must at any cost. its like do and die situation. commercialization comes second.. Loongson is Intel competitor for CPUs .. they have 7nm process next.
I'm surprised anyone is even asking who the customers will be if China gets EUV. Even if we completely ignore the entirety of China, which many companies are desperately waiting for EUV like Buren, Loongson, and others, there's still lines of international customers waiting.
The most obvious one is Russia. Baikal has recently started producing 14nm chips again. No one seems to know how. If you ask me, I would assume that China is completely independent on 14nm with their equipment. Whether Russia bought the equipment themselves and made it or Russia contracted to a Chinese fab I do not know. I would assume contracted to a Chinese fab. So there's at least 1 international customer.
Then you have Iran who is now starting up a chip design firm. They'll obviously buy anything China can sell. More advanced the better.
Next, Taiwan. Taiwan already buys a lot of the packaging equipment from China. If the lithography machines are price competitive, I can't see Taiwan saying no.
Korea. They already have fabs in China and are unable to leave. If China gives them a legal EUV to use, you can bet they would expand.
I could go on, but why? I would even argue that it's possible none of the international sales would occur. China needs these machines and will give them to domestic firms well before any of them could even touch the machines. The demand is insanely high.
The most obvious one is Russia. Baikal has recently started producing 14nm chips again. No one seems to know how. If you ask me, I would assume that China is completely independent on 14nm with their equipment. Whether Russia bought the equipment themselves and made it or Russia contracted to a Chinese fab I do not know. I would assume contracted to a Chinese fab. So there's at least 1 international customer.
Then you have Iran who is now starting up a chip design firm. They'll obviously buy anything China can sell. More advanced the better.
Next, Taiwan. Taiwan already buys a lot of the packaging equipment from China. If the lithography machines are price competitive, I can't see Taiwan saying no.
Korea. They already have fabs in China and are unable to leave. If China gives them a legal EUV to use, you can bet they would expand.
I could go on, but why? I would even argue that it's possible none of the international sales would occur. China needs these machines and will give them to domestic firms well before any of them could even touch the machines. The demand is insanely high.
The science behind a Tipler machine is well known, so basically all you need to do is take a neutron star and stretch it into an infinite cylinder. And then you can time travel. Easy! I'm sure it will be "ready very soon."
"Industrial maturity" is literally the entire problem. Ideas are nothing; implementation is everything.
This is going beyond the idea, way beyond the idea. Like a deepseek moment. 7nm from the start.
For 1.35 immersion novelty will be about 40nm planar, 28nm is in the middle and peak maturity by forcing things, about 5nm. SMEE claims 7nm.
For this one, novelty start around 14nm from the get go and not much after, 7nm.
For this one, novelty start around 14nm from the get go and not much after, 7nm.
One thing that I have been seeing a lot is how insanely complex the global semiconductor industry and supply chain is. It's basically a Jenga tower that is barely standing due to the efforts of some really talented people, and it's so complex and expensive, that it's basically impossible to sort it out and make it more efficient. It's so complex that literally nobody even understand the entire picture in detail. That's how you get situations where a single obscure Japanese company that nobody has even heard about that makes some obscure chemical or machine that is crucial to 90% of the semiconductor supply chain or how there's a lot of the chip making equipment that's still running on decades old code because people are fearful of updating it in case it screws up the yield or and other really outlandish examples. The supply chain is a tangled mess, that nobody can really sort out.
With a supply chain this large, complex, long and with plenty of middlemen inbetween, there has to be some insane inefficiency in the system, that still continue to exist because it's so complex that people aren't even aware that the inefficiency even exists. Combine that with the insane amount of money flowing in and there is probably a lot of bloat floating around somewhere. Especially for older more mature nodes, since there's not much of a drive to make their manufacturing process more efficent, as compared to the cutting edge nodes.
I do hope that since the American sanctions have pushed said Jenga tower over and forced China to basically remake the global semiconductor industry largely from scratch inside of the chinese market, that China rebuilds it in since a way that is a lot more efficient. Maybe that one Japanese company that rests on it's rest on it's laurels because it sits on monopoly in a niche market that most people don't even know exist, is actually charging a 100% markup and hasn't actually innovated in decades due to their monopoly. Maybe there's dozens of legacy companies across a dozen countries that all sell variations of the same core product, that aren't challenged because not wanting to rock the boat and the geopolitical issues, that can be all squeezed into a single company with 1/20th of the workforce.
I always find it funny whenever people keep quoting just how insanely complex the semiconductor industry is, just how many companies and people are working in it, how many countries are involved, how specialized everything is, how siloed it is, how nobody can barely understand the supply chain in depth. And how this means that China will never be able to replicate it due to it's complexity and be self sufficient, especially in the high end nodes. But to me when I hear all those things, all I can think about is how insanely inefficient the entire thing must be. It's not gonna be easy, but since China is starting basically from scratch in a lot of this fields, they hopefully don't have to deal with existing supply chains, existing contracts, technical debt and outdated institutional knowledge and can start with a clean slate.
As Deepseek and plenty of other examples have proven, working with constraints and starting from zero can sometimes have it's advantages. Hopefully China comes out of this with a much more efficient, cost effective, innovative and centralized semiconductor industry that can do a better job with less people, companies, materials and money. Also, all located within hours of travel times from each other instead of having to travel across the entire globe.
With a supply chain this large, complex, long and with plenty of middlemen inbetween, there has to be some insane inefficiency in the system, that still continue to exist because it's so complex that people aren't even aware that the inefficiency even exists. Combine that with the insane amount of money flowing in and there is probably a lot of bloat floating around somewhere. Especially for older more mature nodes, since there's not much of a drive to make their manufacturing process more efficent, as compared to the cutting edge nodes.
I do hope that since the American sanctions have pushed said Jenga tower over and forced China to basically remake the global semiconductor industry largely from scratch inside of the chinese market, that China rebuilds it in since a way that is a lot more efficient. Maybe that one Japanese company that rests on it's rest on it's laurels because it sits on monopoly in a niche market that most people don't even know exist, is actually charging a 100% markup and hasn't actually innovated in decades due to their monopoly. Maybe there's dozens of legacy companies across a dozen countries that all sell variations of the same core product, that aren't challenged because not wanting to rock the boat and the geopolitical issues, that can be all squeezed into a single company with 1/20th of the workforce.
I always find it funny whenever people keep quoting just how insanely complex the semiconductor industry is, just how many companies and people are working in it, how many countries are involved, how specialized everything is, how siloed it is, how nobody can barely understand the supply chain in depth. And how this means that China will never be able to replicate it due to it's complexity and be self sufficient, especially in the high end nodes. But to me when I hear all those things, all I can think about is how insanely inefficient the entire thing must be. It's not gonna be easy, but since China is starting basically from scratch in a lot of this fields, they hopefully don't have to deal with existing supply chains, existing contracts, technical debt and outdated institutional knowledge and can start with a clean slate.
As Deepseek and plenty of other examples have proven, working with constraints and starting from zero can sometimes have it's advantages. Hopefully China comes out of this with a much more efficient, cost effective, innovative and centralized semiconductor industry that can do a better job with less people, companies, materials and money. Also, all located within hours of travel times from each other instead of having to travel across the entire globe.
Last edited:
lol, I don’t mind the criticism, just throwing out ideas since we saw such a huge market correction this week in the US. Now people far better qualified than me (not really qualified for anything…) seem to think the demand for cutting edge chips might be greatly exaggerated.
To address some of the points brought up.
- We’ve seen actually many companies continue work despite being at risk of sanctions which ended up coming down, the latest example being Li Auto. Prior to this Moore Threads and Biren, those are off the top of my head (All were still working on TSMC process past 2023, they must have known sanctions were coming and continued work anyway).
- Cars are using mostly 28nm and above except for advanced UIs and Drive assist functions. They are only on higher end cars, so the market is not huge yet. The cost of the self driving/assistance systems as a whole is not yet sufficiently low to drive volumes in the millions. Even in the large China market this holds true.
- Handset makers cannot just go to SMIC because they don’t have their own SOC design. Only Huawei/HiSilicon has a high end SOC and they are not known for sharing. China doesn’t have a Qualcomm equivalent. UNISOC is low end for now. Of course, this could be a chicken and egg situation.
- Having the machines is not a genie in a lamp situation. As I pointed out, Intel and Samsung both have EUV equipment, and are in trouble. Samsung is reported to have extremely low yields at 3nm. They have not had any big clients on the level of nVidia/Qualcomm since their 8nm process.
- @tokenanalyst is capturing the essence of my thinking. The starting point might be stronger than we know. Going back to Samsung, they are reported to have only 20% yield on the 3nm GAA process, yet they were trying to sell this as commercially viable. Also reportedly Intel has below 50% for 18A (similar to 3nm). Chinese companies developing EUV equipment must be aware of the issues those companies are having so it makes sense (especially in light of current market conditions) to take time to go beyond what our current expectations might be.
To address some of the points brought up.
- We’ve seen actually many companies continue work despite being at risk of sanctions which ended up coming down, the latest example being Li Auto. Prior to this Moore Threads and Biren, those are off the top of my head (All were still working on TSMC process past 2023, they must have known sanctions were coming and continued work anyway).
- Cars are using mostly 28nm and above except for advanced UIs and Drive assist functions. They are only on higher end cars, so the market is not huge yet. The cost of the self driving/assistance systems as a whole is not yet sufficiently low to drive volumes in the millions. Even in the large China market this holds true.
- Handset makers cannot just go to SMIC because they don’t have their own SOC design. Only Huawei/HiSilicon has a high end SOC and they are not known for sharing. China doesn’t have a Qualcomm equivalent. UNISOC is low end for now. Of course, this could be a chicken and egg situation.
- Having the machines is not a genie in a lamp situation. As I pointed out, Intel and Samsung both have EUV equipment, and are in trouble. Samsung is reported to have extremely low yields at 3nm. They have not had any big clients on the level of nVidia/Qualcomm since their 8nm process.
- @tokenanalyst is capturing the essence of my thinking. The starting point might be stronger than we know. Going back to Samsung, they are reported to have only 20% yield on the 3nm GAA process, yet they were trying to sell this as commercially viable. Also reportedly Intel has below 50% for 18A (similar to 3nm). Chinese companies developing EUV equipment must be aware of the issues those companies are having so it makes sense (especially in light of current market conditions) to take time to go beyond what our current expectations might be.
The DeepSeek breakthrough show that sometimes you have to make things differently to what the others are doing to get results, in most of the times the financial guys it will always limit you, that is why DeepSeek treated their project as academic with long term rewards. Shanghai ICRD and others are suppose to be that, the academic branch of China semiconductor industry and they come with breakthroughs but they always faced the doubts of semiconductors managers and force most of the time to get along.
There is the mentally specially along the old guard if you don't do things exactly as TSMC for example you can't, you need ASML, LAM and JSR photoresist, instead of pushing SMEE scanners to become better this people lead SMEE to stagnate. Knowing very well that China was susceptible to export controls. It took the treat of death for Huawei and SMIC to innovate their way in semiconductors, the same goes to YMTC, by using domestics tools they are learning the hard way after wasting years of research collaborating with US toolmakers instead of domestic tool companies.
I think the work that China is doing EUV lithography is interesting, the pressure of broad export controls and not having an EUV scanner is pushing to do things that in software, hardware and exascale compute machine learning it could prove to be innovative, they already getting results, I think it could be another DeepSeek moment. But once the machine is ready, is up to China semiconductor industry managers to decide whatever they want to push this domestic technology to make advanced cutting edge semiconductors or they want to be stuck making refrigerator chips, ASML EUV scanners are not coming to China ever.
Another technology that going against the industry ASML bias is nanoimprint lithography, I do think the technology has potential to be cost effective and could give better results that even EUV. But good luck telling the "decade experience" China semiconductor guard break their ASML bias to setup semiconductor labs with Tenren Nano and others to push the technology forward for its use in advance logic.