Didn't know that the late elder Jiang wrote about semiconductor in China.
Chinese semiconductor industry
- Thread starter Hendrik_2000
- Start date
- Status
Didn't know that the late elder Jiang wrote about semiconductor in China.
MIC2025 was heavily misunderstood by Western analysts since 2015, as far as I know the 70% by market share has always been quoted by Western IC insiders, inside China all commentary has always been by units fabricated, in fact if you enter in the National Bureau of Statistics which is the guide that the Chinese goverment use for their policies, the output of Chips is measured in billion of units rather than in currency and makes sense because you just measure market share you are not seeing the full picture.
I don't think they are many fabs pursuing less than 14nm, except for SMIC and they are doing it on their own because is the logical step after mastering 28nm, they may have received subsides, apart from private money, but I can bet money with anyone that was a decision made by their own executives. Also given that fact that SMIC has a very heavy Taiwanese influence in their DNA, pursuing harder to make nodes comes natural to them. The other attempt I can think of was the delusional intentions of a mad guy who thought that he could make 14nm logic chips in a first attempt, of course he burn out money at match 25 before he could build anything.
The goverment has been investing heavily in HSR networks, Ultra High Voltage Networks, NEV networks, so a lot of "mature" semiconductors are needed to sustain those efforts. The critical infrastructure for China development. The goverment has invested heavily power semiconductors, optoelectronics, MEMS, Silicon Carbide, GaN and so on. Probably even more money than they have put in memory. They don't want to be dependent on this critical infrastructure chips on hostile countries.
Taking out the fact that most American politicians don't know even the ass is located, is possible that they could ban even mature nodes.
Ok, China as country do not made revenue making chips, companies like SMIC do and they will find a way, be sure about that.
28nm is often quoted in China because mark the boundary between advanced nodes and mature nodes, so will allow China equipment industry to evolve into more advanced node, create a positive feedback loop, that is probably much better than making 7nm borrowing tools subject to the will of hostile politicians. That is why SMEE immersion dual stage scanner is marketed for 28nm.
For all the talk about "national security" different from chips needed for the infrastructure, that need to be made in high volume, the Chinese military and for Chinese supercomputing efforts they dont need a lot of chips and those high end chips will be made even if the have to pattern them at 1WPH with a lab made EUV machine.
"Commerce Secretary Gina Raimondo says China has prioritized its national security over economic growth and trade with other nations."
A bit of a correction:
"Commerce Secretary Gina Raimondo says United States of America has prioritized its national security over economic growth and trade with other nations."
His quote.
That still sounds iffy right? 10% total global foundary revenue is going to be a lot more when just focused on the China market. Even NAND, I remember recently reading that YMTC is at 5% global revenue. 6.6% of domestic demand seems really under counting things.
SMIC revenue has been about $1.9 billion a quarter this year. Huahong without counting HLMC is at around $600 million a quarter. With HLMC, it's probably at a $1 billion a quarter. With those 2 alone, you are going to hit $12 billion this year. The thing is that China has so many IDMs that make a few hundred million here or a billion there a year. This is significantly undercounting things.
Nobody said this is ideal. The cost were using domestic tools for some of these process is probably quite high. But SMIC has no choice in this case.
btw, I agree with all of what you are saying here. You are looking at this from a commercial point of view. The rest of us are looking at it from a national security point of view. When Shanghai gov't is ponying up the money for capex and giving SMIC free land at Lingang and Pudong, SMIC can run an operation that is far from optimal since much of its risk has already been taken away by the local gov't.
SMSC itself rarely gets talked about. SMIC just mysterious have $1.6 billion additional capex from somewhere. SMIC doesn't talk about where the money comes from and makes up unbelievable tale about how it gets used. But geopolitical climate is such that SMIC will be expected to run lower yielding and more wasteful production lines.
That's a good thing. The fact that the domestic fabs are working with domestic tools means they will be able to improve faster.
btw, I also see the current American strong arming of TSMC/Samsung as great for Chinese tools makers, since these 2 companies will be more willing to give Chinese tool makers a look. Nobody wants to get strong armed by monopolies.
And I definitely don't expect you to know everything about the one fab that SMIC is trying to keep secret, lol.
American laws didn't allow export of sub-14nm tools to China. Lam employee stated they had very few staff at SMSC. I think they've been struggling for 2 years in trying to raise production at SN1 without American tools.
That's where this gets interesting. I agree that 35k worth of 7 nm probably only needs a few NXT2050/2100. I also agree that SN1 doesn't need anymore scanners. Yet, ASML has invested a bunch of money into developing them with China seeming to be the primary market. As such, it probably only looking to sell more than 2 or 3 of these machines to SMIC and HLMC. We also know that SMIC should be concerned about getting cut off, so it would fast tracked orders to stock up on machines.
Here is some math for you. SN1 capex for 35k was supposed to be $9 billion. Let's say it spent most of that from 2018 to 2021, since we both agree that they should be close to 35k wpm by this year. We also know that they averaged probably $4 billion a year in total capex during those years. As such, it would reason that half of the SMIC capex from 2018 to 2021 went to its SMSC fab! We also know that SN2 fab was in the midst of construction for this year. As such, tools they take in this year should be going to SN2 fab. Now, if they stopped taking tools for SMSC, then we should see their capex drop to about $3 billion in 2022 (which would still be a lot more than what they spent on other fabs in previous years), but it has actually gone up to $6.6 billion. That means, they are probably spending over $3 billion on SN2 tools this year + some accelerated deliveries for their new production lines in Beijing and Lingang.
That doesn't mean SN2 production will necessarily start in 1H of next year. Their SN1 production increase has been very slow. I'm not convinced they are actually at peak capacity at SN1. As they will inevitably be looking to increase 7 nm production, they will need more time to work out non-American tools to make sure they can raise yield to acceptable levels. That might take a while. It doesn't change the overall plan/trajectory they have here.
What I got out of this is that SN2 will require more capex and generate more revenue than SN1 (whether from higher wpm or from more advanced processes). Other projects simply do not justify this level of capex. Think about it, total capex between Beijing/Tianjian/Shenzhen/Lingang is $26 billion. that's to mostly spent over 7 years from 2021 until 2027, which is under $4 billion a year. And that number is likely to be higher when we get to 2024 to 2026 timeframe when Beijing/Shenzhen/Lingang are all adding capacity at the same time. If we continue to see total capex in 2023 to 2025 at $6 to 8 billion a year, then 2 to 3 billion of that are going to SN2.
ASML's own words
I'm sure you know more than I do, but it sounds like something you'd really need in order to have advanced process without EUV. Does TSMC really need 2100i for 5 to 7 nm process when its most critical layers are already done by EUVs?
We've gone through this several times already. Other fabs cut orders, so slots opened up. Now, SMIC can take earlier slots than it otherwise had access to. We've already seen October sales double from ASML.
And it's quite likely that any possible Dutch action would provide a grace period. Dutch gov't have already made it clear that they don't consider EUV and DUVs to be the same case.
Nicely done Bro, without mature chip, those EUVL and DUVL won't function, from what I read one of the reason for the ASML supply chain problem.
Q&A with Hygon
On if future US sanctions could affect them. They answered that they've mastered x86 architecture and independently developed 3rd generation chips. Seems like they are confident here
On whether or not they will be affected by the sanctions in Q4. They said minimally.
When asked about status of Hygon 3 (海光3). They said it is for sale right now and will be their main product in 2023.
Asked in the event of AMD stop licensing technology to them. They said that the company has completed the digestion of licensed technology and possess the ability to independently complete subsequent technical iterations and product upgrades
Here is a product guide for Hygon as of Nov 2022
You will notice that Hygon 4 is coming out in 2024 and uses 7nm process. Not clear to me if Hygon 3 is produced by TSMC and SMIC. I think I read that it's supposed to be TSMC, but they may have a SMIC version also.
On if future US sanctions could affect them. They answered that they've mastered x86 architecture and independently developed 3rd generation chips. Seems like they are confident here
On whether or not they will be affected by the sanctions in Q4. They said minimally.
When asked about status of Hygon 3 (海光3). They said it is for sale right now and will be their main product in 2023.
Asked in the event of AMD stop licensing technology to them. They said that the company has completed the digestion of licensed technology and possess the ability to independently complete subsequent technical iterations and product upgrades
Here is a product guide for Hygon as of Nov 2022
You will notice that Hygon 4 is coming out in 2024 and uses 7nm process. Not clear to me if Hygon 3 is produced by TSMC and SMIC. I think I read that it's supposed to be TSMC, but they may have a SMIC version also.
- Status