It takes 10 DUV immersion scanners to process 45K wpm at 28nm node, 16 at 22nm, 19 at 10nm. It takes 15 DUVi at 7nm plus 6 EUV. I don't know where tphuang got my quote, and I may have been referring to SMIC reaching 7nm without EUV and substituted DUV for EUV. You're statement that my 10 DUVi demand at 28nm is not correct is wrong. It is correct.
@tinrobert, thanks for the clarification that you are talking
only about the 193nm immersion system requirement for 28nm & 7nm.
By the way, I didn't say you are wrong, I stated what
@tphuang referenced off you (what he provided was 10 DUV for 50K) is incorrect. It's clear now that what you actually said was 10 DUV
i for 45K 28nm.
Actually, I make no determination whether your statement or tinrobert's statement was correct. Since I don't work in this industry, I consider both to be plausible. It seemed like your estimation for the # of required DUVi for 28nm was about twice as his. That's fine, I can consider one as an upper bound and the other as lower bound.
Bro, I think you still don't see it. It takes more than just immersion systems to make 28nm wafers. I gave an estimate on
TOTAL number of scanners needed.
@tinrobert clarified his estimate is
ONLY on
immersion systems. My numbers are actually more conservative than
@tinrobert. If I use his assumptions on the immersion tools, we'll end up with an even higher number than what I had shared earlier.
I had a problem with you making an assertion of how quickly SMIC can scale up based on immersion scanners they were buying and Capex them were scanning. it's quite clear based on their stated Capex and how many immersion scanners they could purchase. Based on my calculation, they are allocation about $5.5 to 6 billion for equipment purchases at both Beijing and Tianjin and a little more at the Shanghai Lingang plant. As such, there is no indication they are just building 28 nm wafers or even 45/55 nm wafers. I think they are building a wide range of mature node wafer due to the tremendous industrial demand in China.
And again, they are not the only one footing the bills here. The local government are sharing the Capex in order to attract fabs on their property. I guarantee you that SMIC will be spending less Capex on its new fabs and lower annual operating cost than any other chip makers in the world.
That's how they are able to announce 340k wpm of 12-inch wafers in 3 years and possibly even more than that. They are looking at adding about 15 to 20% 8-inch equivalent wafer capacity per year. A large chunk of that will not be 28 nm and lower or even 45/55 nm.
I am aware money is not an issue for SMIC, never question it so don't know why people keep bringing this up.
Sum up what I said before:
1. You get out (wafer out) what you put in ($$ spent)
- fab announcement doesn't translate to actual wafer output
- CAPEx is $$ actually spent or budgeted to spend
- actual $ invested regardless where the money come from is recorded as CAPEx; this is not how much money SMIC has or has access to
- Faster the spending spree, the quicker the wafer capacity scale up.
- the current CAPEx pace is not enough for SMIC to complete all the announced fab by 2024*
- SMIC needs to spend more and faster than their current (2021/22) CAPEx to if
@ansy1968 claim that >400K wpm added by 2024 is to happen.
- Since SMIC is not money limited, CAPEx is to be taken as actual money they are able to spend (what WFE suppliers limits SMIC to spend)
2. Got money, but not enough scanners available for purchase
- SMIC has money to spend
- ASML/Nikon/Canon not able to meet SMIC's and global demand. ASML said they're not able to increase their supply before 2025
- Supply limited, China as a whole currently receives ~90systems/year.
- Assuming SMIC could buy 30 out of 90 per year, this is not enough if >400Kwpm added by 2024 is to happen
In summation, CAPEx measures actual money spent (not how cash rich SMIC is). SMIC needs to increase the pace of CAPEx spending, but actual CAPEx SMIC could spend will mostly be limited by the supply of scanners.
I feel I had repeated myself too many times trying to explain what was/wasn't said, so this will be last of it on my end. If you'd like to continue this, I propose we do it in private chat.
----------------------------------
* To bring the original context back into focus, below is what my analysis was responding to:
"Topline ASML are used in SN1 and SN2 FAB (100,000 wpm scalable to 150,000), expansion FAB in Beijing, Tianjin, Shenzhen, Shanghai Lingang and Chongqing will be using mix with majority domestic DUVL with IOC next year with full operation in 2024. Can SMEE deliver? with 4 months before 2023, yes it can!
"
