Chinese semiconductor industry

Status
Not open for further replies.

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
AFAIK this one-year reprieve refers to Samsung and TSMC fabs in China that still can buy US tools, but it does not mean Samsung can sell advanced foundry services for Chinese AI customers. Samsung in China has a Dram fab.



In the original
Please, Log in or Register to view URLs content!
article is written Samsung will be ready "from as early as 2024", so not clear what is going on here.



Here Baidu refers to current Kunlun chip that has performances below the specification threshold for the ban to enter in effect. Future, more powerful iterations of Kunlun chip is not clear if will be sanction-free and where will be manufactured.


The article is a kind of collage of different pieces of info, that belong to different contexts, that have been assembled in a somewhat optimistic way by the author. I would wait for confirmation before celebrating.

Of course, if would be confirmed that Samsung can manufacture advanced nodes for Baidu for the entire 2023, it is clear that (1) it would be a huge news, (2) Baidu would submit such a big order that the Kunlun chips out of Samsung would be enough for the next 10 years (3) also TSMC would have this possibility: there is no reason why Samsung yes and TSMC no.
I agree with most of what you are saying here except I would caution you to equivocating Samsung and tsmc. They are not in the same position.

Tsmc dominates the advanced node market and Samsung is losing customer after customer. Tsmc can pick and choose it's customers whereas Samsung really cannot. Tsmc also has far more at stake in America than samsung. It is far more easily swayable by American pressure. Taiwanese govt basically does American govt bidding. Koreans at least have some leg to stand on. As such, Koreans can take on certain project that tsmc would be fearful of doing.

I think it's more likely that Samsung creates a de americanized production line with Chinese tools than tsmc. Which is basically getting cut up to pieces and sent to America as we speak.

Kunlun 3 chip is coming out in 2023/2024 based on previous announcements. Given that Kunlun 2 uses 7 nm process, next gen is likely to use 5 nm process or better. I think it's quite optimistic of them to operate under the assumption that America won't adjust ai chip criteria to be even more stringent but we will see. Unless Samsung can create de americanized production line, I don't see how baidu can trust their next generation ai chip at their hand. They need to work with smic like everyone else.

But one thing I do sense is that Japan and South Korea are going to resist further pressure to decouple from china. As I predicted before, these two countries are too interconnected with china economically. They would definitely suffer if they voluntarily impose new restrictions. As such, Dutch govt have to think about whether or not they are going to put their national champion in a bad position by pushing them to both local Chinese competition and Japanese competition. If Japan doesn't fold, the likelihood of other countries folding is significantly reduced.
 

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
bid info from this week
Please, Log in or Register to view URLs content!
quit a lot of bid request from Shanghia Jita. Including one for Arfi scanner from what I can see. Would be interesting to see who wins that one.

Looks like 8/12 winning bid went to domestic firms, including 5 for NAURA. The remaining went to Japanese firms and Nova. Looks like Shanghai Jita is still favoring foreign products. Given that these products shouldn't be for mature system, one would imagine international firms are still fighting very hard to win these bids.
 

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
Apparently CCTV had a report about a hydroelectric station turning to FT2000/4 CPU, NeoKylin OS and other Chinese chips/tech in order to use 100% domestic technology in controlling the power station. From what I can recall, the requirement of computing power in power station is not high, but they need to be very reliable. So it makes sense for them to move these critical infrastructure that don't have high processing requirements to 100% domestic technology to prevent possible tampering or hacking. Or maybe they are doing this to just supporting local chipmakers.

I also see phytium and hygon chips being adapted in more processor intensive industries. But I think they need to come out with more powerful chips if they want to encourage more non government organizations to start using their desktop and server CPUs.
 

FairAndUnbiased

Brigadier
Registered Member
I agree with most of what you are saying here except I would caution you to equivocating Samsung and tsmc. They are not in the same position.

Tsmc dominates the advanced node market and Samsung is losing customer after customer. Tsmc can pick and choose it's customers whereas Samsung really cannot. Tsmc also has far more at stake in America than samsung. It is far more easily swayable by American pressure. Taiwanese govt basically does American govt bidding. Koreans at least have some leg to stand on. As such, Koreans can take on certain project that tsmc would be fearful of doing.

I think it's more likely that Samsung creates a de americanized production line with Chinese tools than tsmc. Which is basically getting cut up to pieces and sent to America as we speak.

Kunlun 3 chip is coming out in 2023/2024 based on previous announcements. Given that Kunlun 2 uses 7 nm process, next gen is likely to use 5 nm process or better. I think it's quite optimistic of them to operate under the assumption that America won't adjust ai chip criteria to be even more stringent but we will see. Unless Samsung can create de americanized production line, I don't see how baidu can trust their next generation ai chip at their hand. They need to work with smic like everyone else.

But one thing I do sense is that Japan and South Korea are going to resist further pressure to decouple from china. As I predicted before, these two countries are too interconnected with china economically. They would definitely suffer if they voluntarily impose new restrictions. As such, Dutch govt have to think about whether or not they are going to put their national champion in a bad position by pushing them to both local Chinese competition and Japanese competition. If Japan doesn't fold, the likelihood of other countries folding is significantly reduced.
Samsung has a substantial advantage over TSMC -
Please, Log in or Register to view URLs content!
Please, Log in or Register to view URLs content!
. In addition South Korean companies look like they're more integrated into the component level of equipment supply i.e. ultrahigh purity cleaned components, stainless steel machining, corrosion resistant coatings, etc than Taiwanese. But they still have some ways to go.
 

gelgoog

Lieutenant General
Registered Member
You assume too much. Samsung collaborates with IBM for process development. Their 3nm process for example is supposed to be based on IBM's work on GAA transistors. US government put the screws on sales of EDA software to China which can work with GAA transistors. And FinFET is having horrible issues with power draw at lower process geometries. I mean just look at the power draw of the latest generation graphics cards and CPUs. It is insane. For example AMD Ryzen 9 went from 105W to 170W. Nvidia released a high end graphics card with 450W power draw. GAA transistors are expected to reduce power draw by up to 50%. Shrinking the transistors no longer leads to significant improvements in reduced power draw like it used to in the past without redesigning the transistors. If anything design of transistors and better EDA tools will become more important not less.
 
Last edited:

WTAN

Junior Member
Registered Member
Do you have a timeframe for when the 90W - 120W excimer laser will be ready? Also, do you have any updates on the EUVL? Are things still on track for 2025 or have there been setbacks/delays?
RSLaser was only established in 2016 and is doing quite well developing both the 40W and 60W Arf Lasers so far.
They are definitely working on a more powerful Arf Laser in the 90W - 120W Class.
It shouldnt take too long for them as it is just an upgrade to their current range of Arf Lasers.
Gigaphoton developed their 90W Arf Laser back in 2007, so this technology has been around for a while and is nothing new.
Huawei is also providing alot of funding to RSLaser these days, so a upgraded SMEE DUVL is just round the corner.

It is also good to hear confirmation that Huawei is now getting involved in EUVL R&D as shown by the patent filing and this will speed things up quite a bit.
I believe the Govt now realises the strategic importance of Lithography equipment especially EUVL and will allocate quite a bit of funding to it.
This reminds me of the Chinese AWACS program with many different radars developed.

Good progress is being made with the EUVL LPP Light Source which is the most critical component and i believe we will see something in 2025.
 

Zichan

Junior Member
Registered Member
No surprises here.. Military hardware comes 4-5 generations behind because reliability and durability is paramount. It is same for western and Chinese equipment. This why sanctions to countries like China and Russia to prevent military development is a joke.
It’s not entirely a joke. State-of-the-art radars use 7nm and 5nm chips. Future applications of AI in weapon systems will require large computing and storage resources for training of deep learning models. AI at the edge often requires high power efficiency: nVidia recently demonstrated a 5nm inference chip for such applications.
 

european_guy

Junior Member
Registered Member
bid info from this week
Please, Log in or Register to view URLs content!
quit a lot of bid request from Shanghia Jita. Including one for Arfi scanner from what I can see. Would be interesting to see who wins that one.

Looks like 8/12 winning bid went to domestic firms, including 5 for NAURA. The remaining went to Japanese firms and Nova. Looks like Shanghai Jita is still favoring foreign products. Given that these products shouldn't be for mature system, one would imagine international firms are still fighting very hard to win these bids.

8 out of 12 is a lot for domestic firms, current localization ratio in recent bids is around 35-40% so this is a big jump. If it will be confirmed in the next months it means tool localization capacity has reached scale (60%-70% means market needs are practically covered).

This Nova firm is interesting, actually is an Israeli company specialized in process measurement, clearly an answer to KLA.

Please, Log in or Register to view URLs content!

BTW no US toolmaker did win even a single bid!

Considering we are not talking of advanced banned nodes here, this confirms our expectations that for US tool makers is game over in China: a ban on advanced node's equipment can be eventually tolerated, but to abruptly interrupt after-sales support and spare parts service cannot be forgiven in any industrial environment. These US hawks clearly never worked in a production company otherwise they would have known this is really a big no no...and IMO they even didn't bother to ask people with real-world production experience.

I don't know how it will end up with ASML but I'm very confident that, no matter what, ASML after-sales service will not be touched.
 
Last edited:

tonyget

Senior Member
Registered Member
bid info from this week
Please, Log in or Register to view URLs content!
quit a lot of bid request from Shanghia Jita. Including one for Arfi scanner from what I can see. Would be interesting to see who wins that one.

Looks like 8/12 winning bid went to domestic firms, including 5 for NAURA. The remaining went to Japanese firms and Nova. Looks like Shanghai Jita is still favoring foreign products. Given that these products shouldn't be for mature system, one would imagine international firms are still fighting very hard to win these bids.

You'd think SMEE's lithography machine is good enough for mature node producer like Shanghai Jita,but somehow I think Shanghai Jita will stick with the Big3
 
Status
Not open for further replies.
Top