Chinese semiconductor industry

SleepyStudent said:

Not really. EUV is only used in smartphones and demand side issues would resolve themselves in a total decoupling scenario

Click to expand...

All of AMD is on TSMC 7nm I think. some major chips on F35 fighter are on TSMC 7nm (I am sure this is even more relevant from drones, especially as we move to smaller swarming drones)

Nvidia uses samsung 10nm (not sure if EUV used) for its 30x0 GPUs and the result is almost 2 generation jump. Why? most CPU tasks are single threaded which means clock dependent and we are close to the ceiling. But GPU and AI chips are parallel which means its performance is highly dependent on advanced fab.

anyway I didn't want to get into a debate on how big a problem it is for China. It is a huge challenge to tackle and terrible situation no matter what. I want to focus on solutions. And I trying to point out that chips and software are the two side of the same coin.

CHINA SHOULD BE WORKING ON ADVANCED FABS, BUT ALSO LIGHTER SOFTWARE AND BETTER HARD/SOFTWARE INTEGRATION (LIKE APPLE) IN ORDER TO FULLY ADDRESS THIS CHALLENGE.

victoon said:

All of AMD is on TSMC 7nm I think. some major chips on F35 fighter are on TSMC 7nm (I am sure this is even more relevant from drones, especially as we move to smaller swarming drones)

Nvidia uses samsung 10nm (not sure if EUV used) for its 30x0 GPUs and the result is almost 2 generation jump. Why? most CPU tasks are single threaded which means clock dependent and we are close to the ceiling. But GPU and AI chips are parallel which means its performance is highly dependent on advanced fab.

anyway I didn't want to get into a debate on how big a problem it is for China. It is a huge challenge to tackle and terrible situation no matter what. I want to focus on solutions. And I trying to point out that chips and software are the two side of the same coin.

CHINA SHOULD BE WORKING ON ADVANCED FABS, BUT ALSO LIGHTER SOFTWARE AND BETTER HARD/SOFTWARE INTEGRATION (LIKE APPLE) IN ORDER TO FULLY ADDRESS THIS CHALLENGE.

Click to expand...

Am aware. I wasn't trying to be hostile with you. Just pointing out that at advanced nodes, it becomes a purely logic-based debate and once your at logic, heterogeneous computing and design can mitigate significant amounts of the node-shrink debate

victoon said:

CHINA SHOULD BE WORKING ON ADVANCED FABS

Click to expand...

Just a minor correction. First priority is IC equioment and then build Fabs utilising these machines to produce advanced chips.

This is why I never talk about TSMC. Whats important is ASML (and its upstream suppliers)

ougoah said:

Fair enough if that's what you're referring to as scientific metrics. I would consider Japan to have more high technology exports than China though but that's very quickly closing.

Different principles of physics as in the EUV method involves different physics compared to DUV for example. No matter how good China is at DUV lithography, it isn't the equivalent of EUV is what I mean. It's like training very hard and being very good at high jump when you need to compete in a long jump competition. Hence different. The evidence it's different is self apparent in that China has yet no solution to EUV.

It's as simple as bouncing light off mirrors just like a turbofan is as simple as compression air and lighting some fuel with it. Much easier said than understood and even harder to actually do.

Click to expand...

...the only major difference between EUV and DUV is the photon generation mechanism, and the specific mechanisms used are from research that’s three decades old, which Chinese research institutes have already demonstrated the capability to reproduce. Everything else follows the same physics principles as other lithography mechanisms. You’ll need some different materials for your masks and what not but even the selection of those follow the same physical principles, just fitted for a different set of parameters. The sports analogy makes no sense here. The engineering can be challenging, but once you have the science down all engineering comes down to an function of man hours and resources. I’ve said this before and I’ll say it again. We shouldn’t treat these technologies as inscrutable magic. These kinds of technologies are things that take time because they require a lot of tweaking and optimization to hit a set of performance parameters, not because people are dumbstruck trying to learn an entire universe of new knowledge.

latenlazy said:

...the only major difference between EUV and DUV is the photon generation mechanism, and the specific mechanisms used are from research that’s three decades old, which Chinese research institutes have already demonstrated the capability to reproduce. Everything else follows the same physics principles as other lithography mechanisms. You’ll need some different materials for your masks and what not but even the selection of those follow the same physical principles, just fitted for a different set of parameters. The sports analogy makes no sense here. The engineering can be challenging, but once you have the science down all engineering comes down to an function of man hours and resources. I’ve said this before and I’ll say it again. We shouldn’t treat these technologies as inscrutable magic. These kinds of technologies are things that take time because they require a lot of tweaking and optimization to hit a set of performance parameters, not because people are dumbstruck trying to learn an entire universe of new knowledge.

Click to expand...

I'm not doubting that China has managed to master EUV physics. I'm saying no matter how experienced and proficient Chinese industries are with DUVL, that doesn't translate into mastering EUVL and using EUVL to fab anything. Until they do, that is. This is in response to a comment that said there's hardly a difference between them so it shouldn't take any time.

hkbc said:

I think you've missed my point, calling it illegal is just going along with the US worldview that its laws are the world's laws, hence propaganda! If people start allowing the perversion of a language, you'll get oxymorons like 'peacekeeping in Iraq' and 'war on terror'. Just because governments do crap and get away with it doesn't mean it should be minimised by the use of words to diminish its obvious lack of truth. Go down that slippery slope and next you'll just legitimise all the other crap.
...
Nonetheless those bits of US law is never extended there, however, put a US dollar bill in an Iranian beggar's bowl in the streets of Tehran and you've committed an International money laundering crime! American Justice is blind but, apparently, only when its convenient!

Click to expand...

I see what you mean now, we are actually talking about the same thing in different ways.

ougoah said:

Unfortunately, EUV uses different principles of physics than other lithography techniques. There is a world of difference between 14nm and 7nm and beyond. Those equivalent products may be silicon based and applying many similar or the same principles but there are enough differences in their manufacture where it can be considered a totally different kind of product.

Click to expand...

The EUV light source is a huge barrier but most of the other technologies in EUV lithography are similar enough to DUV that the transition to EUV is "relatively" straightforward if you can solve the light source and photoresist problem. There are of course the increased tolerances needed for photoresist, CMP, wafer cleaning and you now have to deal with constant mirror decontamination/replacement, but otherwise, it's actually easier in some ways vs. DUV when you're dealing with fewer exposures to get the same relative transistor density along with higher yields. This is all only true if the light source and photoresist issue can be solved for EUV.

victoon said:

I am afraid EUV is not the only thing China need to tackle to address the chip vulnerability in a sanction or worst case total decoupling scenario.

But all of these are supply side challenges. There are also demand side problems. Because China can’t make a competitive SOC for smartphone for a long time, China should try to catch the train started by raspberry pi and now solidified by Apple’s move to linux (variant) on ARM for desk/laptops.

Click to expand...

Imo, I think China's hopes of a competitive domestic SoC are dead until it has EUV lithography. Even though Xiaomi and Oppo/Vivo have their own inhouse chip divisions working on their own SoC, it will be years before they become relatively competitive and assuming they do, I am sure they will be deemed "national security" threats by the US hegemon and sanctioned like Huawei. In the meantime, Huawei must finally offer HarmonyOS to have any remote chance of retaining their smartphone division. That's assuming that ARM access doesn't become a problem.

ougoah said:

I'm not doubting that China has managed to master EUV physics. I'm saying no matter how experienced and proficient Chinese industries are with DUVL, that doesn't translate into mastering EUVL and using EUVL to fab anything. Until they do, that is. This is in response to a comment that said there's hardly a difference between them so it shouldn't take any time

Click to expand...

That’s...not really correct. Most of the principles that apply to doing DUV lithography apply to EUV. One is a progression of the other, just with higher energy photonics. The transition here is about tweaking the parameters of the process for higher energy photonics, not about relearning an entirely new set of disciplines. This is more like relearning the best strategies for basketball when you add a three point line or 3 second rules than it is learning a whole new sport. If anything, it’s much easier to transition into EUV from a DUV dry process than it is to transition from DUV dry to DUV immersion and multipatterning. EUV adoption was much faster and smoother than adoption of multipatterning or immersion for a reason. EUV is desired because it *simplifies* workflows, and takes out extra components to the lithography process that had to be added to do immersion optics and multipatterning. The reason EUV took so long wasn’t because they had to relearn a bunch of new stuff to make EUV lithography work, but almost exclusively because the light source output was insufficient to make the process fast enough to be economical for industry. The light source was the bottleneck. It’s the one thing that was actually technically challenging, but it’s a very tractable problem.

I agree with the general point that China won’t just get to industrial use EUV with the snap of the finger, but it’s not for the reasons you’re suggesting here. The time it takes to tweak everything to perform reliability and optimally, not the technical knowledge barrier, is what takes so long.

ansy1968 said:

@Skywatcher Sir as confirm by @foofy.

According to Havok, 28nm immersion duv machine can do double patterning, will participate in 16nm domestic equipment trial production line, and will be used in big production line this year end.

Sir about the statement above, need your clarification, by using the SMEE 28NM DUVL for the 14nm models means double patterning? or a new kind of DUVL capable of single exposure at 14nm?

Click to expand...

Not long ago Havok did mention that the new SMEE 28nm DUVL was developed with the goal of achieving a Resolution of around 38nm (resolution of ASML 1980i/2000i is 38nm).
SMEE was benchmarking the ASML product and was developing something with similar resolution to the 1980i.
So basically the news that the SMEE DUVL can produce 14nm Chips is probably not too surprising.
The Nikkei News report about the SMEE DUVL being able to achieve 14nm is probably another confirmation through their own sources.
In any case a unit of the SMEE DUVL will be shipped to ICRD for the 16nm Domestic Equipment trial production line later this year. We will then have confirmation of the true capability of this DUVL.
Also note that TSMC used the ASML 1980i to produce its 1st Generation 7nm Chips made entirely using DUVL.
Another interesting news from the Nikkei article is that Chinese Semiconductor Equipment Makers are already producing a large range of 28nm and 14nm Semiconductor equipment. This indicates that the Localised 14nm Production line will be coming soon shortly after the 28nm Line is established.