Basically a export tax. Nvidia will hide revenue and hide sales geography through accounting tricks. Sales will register in Taiwan and Singapore.
Chinese semiconductor thread II
- Thread starter vincent
- Start date
Yes, the source spent over ten to twenty minutes specifically addressing these two points you raised. For the domestic DUVi systems, a dedicated solution was implemented—reportedly mobilizing thousands of researchers who conducted massive mathematical studies and simulations on every microscopic detail of the production process. Only after confirming optimal equipment parameters and investing substantial resources did they achieve breakthroughs for entirely new DUVi systems and production processes.
Producing 5nm chips on DUVi equipment would be simply impossible with ASML’s standard systems.
The true cost lies in the yield rate sacrifices and extended tape-out cycles—with overall production time significantly increasing as process complexity intensifies.
What I’ve heard is that current 5nm production relies on modified ASML machines combined with domestic DUVi systems—specifically using quadruple patterning alignment technology.
While the EUV hardware is ready, it lacks mature supporting infrastructure. Solutions for metrology tools, photoresists, and ancillary components exist but remain unoptimized. Crucially, EUV process parameter libraries are being built from zero—doping parameters for EUV-compatible ion implantation/deposition, photoresist tuning coefficients, etc.—all require foundational development.
The system functions unstably. Every component is undergoing calibration, making immediate production line deployment impractical. This 1.5-2 year stabilization phase is unavoidable before pilot production can begin.
Though I too hope domestic EUV accelerates, current evidence makes his timeline credible. Personally, I’ve tracked semiconductor lithography literature for over a decade (albeit as an observer without operational experience). His 4-hour presentation—including ~90 minutes of chip fabrication analysis—demonstrates striking coherence with my technical understanding. His breakdown of inherent sub-7nm noise issues (citing three distinct noise sources) aligns with my accumulated knowledge and even deepened my perspective on certain details.
While he isn’t a semiconductor industry insider, his access is verified:
Engages directly with China’s semiconductor leadership
Attended at least one of six classified strategy meetings that reshaped China’s post-trade-war semiconductor response (2018-2022)
Thus, I find his analysis substantiated.
Last edited:
Any links to video?
I think it can both be the case that the presenter’s understanding of all the optimization details that need to be done for effective mass employment are correct *and* that EUV is such a capability uplift for critical steps towards future node shrinks that even a suboptimal process will be employed as best they’re able to just to continue pushing an aggressive product cadence. If for example Huawei can get to 3 nm with an expensive EUV process that is inefficient and has some yield penalties faster than with sextuple patterning that is what they will do. They may even run batches employing two different processes just to hedge their bets and see which will get them to the next node shrink faster. The ideal “stable” industry wide employment of EUV can take another 1.5-2 years *and* EUV can potentially be employed for those who are willing to pay up on risk and costs as soon as any process is minimally viable.
The video emphasized: don't circulate this widely, especially the video itself. Privately celebrate the information—but public sharing could complicate things for the presenter. Everything shared was cleared for discussion within appropriate circles.
Additionally, there were discrepancies between the earlier translation and my original intended meaning. The updated post has now corrected these specific sections.
Last edited:
They have to cut prices before selling to China, otherwise they will not buy due to security concerns.
Right, I think it’s great that you have now explained what his background is. I think his info is very useful but there are also multiple parties working on EUV systems in China. Depending on their requirements, they may have tolerance for more or less stable production using EUV. In fact, DUVi 7nm process is also not that stable. Hence you saw the pitch only change from 42nm in 2023 to 40nm last year iirc. It takes 3 to 6 months for each batch of chips to be produced so it really takes time to improve their process, especially when they had limited number of suitable equipment. Going from 40nm to 36nm will take additional effort that is likely to take time to become stable also.
again, that’s why @latenlazy and I think they are going to tolerate instability in EUV for the critical step. It’s because the process using DUVi using multi patterning is itself lower yield and take long time to stabilize.
Lmao this reads like AI slop. So many em dashes. And who is the "he" you are referencing and where is "his 4-hour presentation"
The rumor I’ve seen is that domestic gear’s advanced-node lines are yielding just fine.