Chinese semiconductor industry

ZeEa5KPul said:

That's an interesting point of view. Why do you think this commercialization will necessarily precede the assembly of a prototype?

Click to expand...

Well they tested a prototype machine for key EUV technologies in 2017, achieving 32nm single exposure vs 14 nm ASML, no sure how many WPH but probably limited, I already posted about that one. So is possible that they will built a non commercial prototype before starting commercialization but I think if they want to build machines for volume commercial (even prototype) production is another deal because the companies involved need to built warehouses, clean rooms, buy production machinery and so on. Companies like Raycus, high vaccum companies, sputtering coating machines companies, EUV sensors companies, hydrogen recycling companies, they all need to be more or less ready. That is not counting the numbers of companies that need to be created and reporpurse. The good news is that we already seeing signs of that happening.

tokenanalyst said:

EUV technology testing platform from 2017, kinda like the concept. Looks suitable to test SSMB as a light source.

View attachment 97418View attachment 97419

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

General Research Laboratory of Microelectronics Equipment of the Institute of Optoelectronics has made breakthroughs in key unit technologies such as vacuum, film transport, workpiece table, vibration reduction and focus detection: the vacuum system achieves a stable high vacuum environment of 5.8 × 10 -7 Torr , ensuring that The stability of EUV light source propagation is improved; the frame and vibration reduction system of the whole machine meet the VC-F vibration reduction standard, and the vibration isolation effect is good in the frequency range of 2.2Hz - 500Hz in the horizontal direction and 2Hz- 500Hz in the vertical direction . The frame, workpiece table and mask table The stability meets the requirements of 32nm lithography and sub-nanometer wave aberration detection; the leveling and focusing detection system achieves nanometer-level detection accuracy, which provides a guarantee for the continuous multiple repetition of 32nm lithography resolution. At the same time, in the development of the workpiece table, the collaborative control technology of multiple workpiece tables is studied on the basis of the existing ones, which solves the problems of simultaneous control and position switching of the wafer table and the image table, the mask table and the object table; the automatic film feeding system Realize automatic loading and unloading in a vacuum environment ; the software system completes multi-module complex control, and coordinates each sub-system to complete the exposure process.

On the basis of completing the development of the key unit system, it has played a great role in cooperating with the overall project unit Changchun Institute of Optics and Mechanics to achieve 32nm repeatable exposure and with Shanghai Institute of Optics and Mechanics to achieve sub-nanometer wave aberration detection, which has been greatly affected by the overall project unit and each subject unit. And the acceptance group experts praised. During the research and development of the subject, the research group has achieved a series of original achievements in the fields of workpiece table control and precision measurement technology, and has applied for 31 invention patents , including 28 domestic patents and 3 international patents .

Click to expand...

ZeEa5KPul said:

I would think assembling the prototype and commercializing the component would proceed independently? Besides, what use do these components have outside an EUV lithography tool?

Click to expand...

Some of the technologies has dual use, of the knowledge can be used in other areas.

TK3600 said:

So from what I see Longsoon is a fully independent design not even risc-v related?

Click to expand...

LoongArch is an extension of MIPS64. Supposedly there is some ancient history between MIPS and RISC ... and since RISC-V is also related to RISC... Loongsoon mention they will support RISC-V in the future. Whether that is from some add instructions to loongson arch to help binary emulation or a new RISC-V cpu is unknown

In the 1980s there were two successful RISC CPU designs in the US, one out of Berkeley University (David Patterson's team) which was later redesigned into the Sun SPARC CPU family, and another out of Stanford University (John Hennessy's team) and that was MIPS which got adopted by Silicon Graphics. Both CPU designs were originally funded by the US government DARPA VLSI Project.

RISC-V is a much more recent ISA designed by a new team which also had David Patterson in it.

Alot of experts agreed that semiconductor ban doesnt affect China. Somewhere in China's lab, they are able to produce 5 nm semiconductor.

The only problem is they cannot produce it in a massive commercial scale to make it cheap and profitable.

Produce a chip in a lab is do-able but very expensive. I am sure they can produce a few 5nm chips per day.

I read this from some article but I cannot find it. Many experts agreed that China is capable of producing 5nm in a lab.

daifo said:

Hygon is x86 based and Loongson is MIPS/Loongson Arch. Supposely some chinese entity owns the MIPS arch now but nearly no one uses MIPS/Loongson. I believe Playstation or Nintendo a few generation back was the last global hurrah. No one seems to completely understand how and what type of agreement China has with x86 license/patents. Some have expire so its free to use, some are cross license from AMD/VIA etc. Either way, the benchmark for both seem to indicate that they are about 5-10 years behind current Intel/Amd chips in performance.

Click to expand...

Hygon and Phytium say they have full license to do what they want with x86/ARMS in their case. I've posted this in several places. Not the case with other chinese chip designers. The most recent Chinese CPUs are probably 3 to 5 years behind and they are hoping to narrow that to 2 years over the next couple of years.

tokenanalyst said:

You are correct, they have not done greatly recently. They may counting in the more recent controllable domestic ISA LoongArch to incite more goverment purchases on the grounds that is more secure for the goverment than this two processors companies that depend on ARM and X64-86 ISAs that are not controlled by China. Also LoongArch claims to be pretty compatible with X86 using translation so it could minimize the cost of porting domestic software.

Click to expand...

I really do get confused by the idea that ARM and x86 not being controlled by Phytium and Hygon. They claim to have full license and developed their own IP for the instruction set. How they still get sanctioned at this point? It's not like ARM can prevent Phytium from using that instruction set for future CPUs in the event of a conflict. The domestic OS and compilers/software makers should all support the instruction set Phytium and Hygon use by this point.

tokenanalyst said:

They also have been developing GPUs and IGPUs with their ISA.

View attachment 103377

Click to expand...

I do know that they have their own GPUs, but it's just not competitive to GPUs that other Chinese design shops have developed.

pecopls said:

Please, Log in or Register to view URLs content!

Reuters reporting that the China government is planning to introduce US$143bn in subsidies for domestic semiconductor production, mainly to purchase domestic equipment, potentially by 1Q23. If it comes to fruition, it makes the US Chips Act look like peanuts

Click to expand...

The math here does not really work out.

$143 billion over 5 years and a 20% subsidy would mean that Chinese fabs need to invest $143 billion in domestic tools a year for the next 5 years. That's clearly not possible.

This is why you can't take these MSM articles too seriously. They can't get basic facts straight. So, I think we can assume that there will be a large package coming that will be more than 1 trillion RMB over 5 years. That money will probably subsidize both construction of fabs, purchase of domestic tools and expansion for domestic tool makers. My guess is that it will also include a bunch of buy China clause for chip purchases. Intel/AMD/Nvidia/Lam/AMAT/KLA can permanently kiss 25 to 30% of their revenue away. The Chinese market is gone for them completely.

And if this causes TI and Onsemi to collapse in a few years, then the market share loss will be even greater for them.

KYli said:

Many Americans are having China syndrome. Just don't ever mention anything China.

Please, Log in or Register to view URLs content!

Click to expand...

Yeah, I tweeted about this. These politicians never think about the consequences of their actions. What a bunch of wackos. They don't seem to understand that 28 nm is not all that advanced and China is well capable of having de-americanized line. Even if the Japanese/Dutch sanction 14nm node, those 28nm node tools are still available. And it's just a matter of a year or two before China is fully ramped up on 14 to 28 nm node with fully domestic process.

With China winning these WTO lawsuits against US sanctions, it is fully able to now put in buy China clauses without legal blowback. You say we are engaged in non-market behaviour? Well, you made us do it because you refused to sell us anything.

Of course China is going to dominate advanced nodes. What do they think was going to happen. If they are not suddenly so obsessed about this, they would have noticed China's advances in other tech spaces too.

tphuang said:

Hygon and Phytium say they have full license to do what they want with x86/ARMS in their case. I've posted this in several places. Not the case with other chinese chip designers. The most recent Chinese CPUs are probably 3 to 5 years behind and they are hoping to narrow that to 2 years over the next couple of years.

I really do get confused by the idea that ARM and x86 not being controlled by Phytium and Hygon. They claim to have full license and developed their own IP for the instruction set. How they still get sanctioned at this point? It's not like ARM can prevent Phytium from using that instruction set for future CPUs in the event of a conflict. The domestic OS and compilers/software makers should all support the instruction set Phytium and Hygon use by this point.

I do know that they have their own GPUs, but it's just not competitive to GPUs that other Chinese design shops have developed.

Click to expand...

ARM have become more nasty with their ISA, see Qualcomm and Nuvia, I think customs extensions are not allowed anymore, if Phytium wants a GPU will have to use ARM ones.
I think Hygon do not have a license to modify the ISA or make custom extensions, they only can manufacture the processors.
I guess Zhaoxin can add custom extensions for to their version of the X86 ISA called the ZX-D archtecture.
Loongson can do whatever they want because the own the ISA, they can add extensions instructions for whatever purpose they want.
And RISC-V users also can do what ever they want, but Instruction extensions that are not supported have to be mantained by the users themselves not the RISC-V community and foundation.

tphuang said:

$143 billion over 5 years and a 20% subsidy would mean that Chinese fabs need to invest $143 billion in domestic tools a year for the next 5 years. That's clearly not possible.

Click to expand...

Yeas, my guess is they mean 143 billion for the entire industry, that include research and development of tools, materials and software. Also subsidies for new fabs and also to domestic tools suppliers.
Western journos just make it see as if is just for buying equipment.

KYli said:

It would be better if China could have accessed to chips equipment from Japan and Dutch for a few months. But I think the West might have overplayed their cards and now there is only one way out for China that is to build its own chips industry from materials, equipment designs to manufacturing.

Please, Log in or Register to view URLs content!

“There’s no way China can build a leading-edge industry on their own. No chance,” said Sanford C. Bernstein analyst Stacy Rasgon.

Click to expand...

Again, without knowing the details of what the Dutch or Japanese want to do here, it's impossible to know how this might affect China. As just discussed with this recent 1 trillion RMB news, Chinese gov't response is to give so much monetary support that it will overwhelm foreign tool makers (first at mature end but will spread to advanced nodes soon enough).

At this point, only SMIC is really affected if 14 nm tools get cut off. And as we discussed, they probably have received or will receive all the ASML machinese they need. Depending on how a future sanction gets implemented, this may or may not actually affect SMIC. I have no reason to believe that they would not be able to work with domestic tools makers to get rest of their 14nm process worked out.

Even if Japanese/Dutch implement a 16 nm process sanction, it still would not affect 28nm process expansion that's going on in China right now. And frankly in a couple of years, this won't make any difference once the Chinese tool makers ramp up.

But the Japanese are really going full stupid. It doesn't matter how much pressure Washington is putting on them. The reality is that China has. looking for ways to just bury Japanese industries. They now have an excuse to do so by cutting off rare earth or whatever else Japanese industries need. It's one thing for the Dutch to impose sanctions on China. It's quite another thing for the Japanese to impose sanctions on China.

tokenanalyst said:

Yeas, my guess is they mean 143 billion for the entire industry, that include research and development of tools, materials and software. Also subsidies for new fabs and also to domestic tools suppliers.
Western journos just make it see as if is just for buying equipment.

Click to expand...

Well, America did the $52 billion subsidy which ended up just being a large tax break for new fabs. China already offers low taxes, low interest loans and cheap power/water for all the fabs. Many local gov't also already offer to buy tools for new fabs in JVs. This is just putting things on steroid.

Even if we ignore China's other cost/tax/subsidy advantages, you can do a lot more with $150 billion in mature process vs advanced process. I mean Huahong's entire year capex (at least non-HLMC portion) is just over $1 billion. With this kind of subsidy, Huahong can just keep building new fabs, because the capex risk will now be significantly lower. The chips act basically brought America 2 TSMC fabs + possible Samsung/Intel fab that are very advanced but also quite limited in capacity while doing next to nothing for mature nodes at all. American politicians have no one to blame but themselves and advisors like Eric Schmidt who is overly obsessed about AI and just care about the shiny advanced nodes.

tphuang said:

I really do get confused by the idea that ARM and x86 not being controlled by Phytium and Hygon. They claim to have full license and developed their own IP for the instruction set. How they still get sanctioned at this point? It's not like ARM can prevent Phytium from using that instruction set for future CPUs in the event of a conflict. The domestic OS and compilers/software makers should all support the instruction set Phytium and Hygon use by this point.

Click to expand...

If Phytium or Hygon are found to be infringing on patents on ARM or x86 then the owners of those patents can basically sue to prevent Phytium or Hygon from exporting their chips.

tokenanalyst said:

ARM have become more nasty with their ISA, see Qualcomm and Nuvia, I think customs extensions are not allowed anymore, if Phytium wants a GPU will have to use ARM ones.

Click to expand...

The lawsuit with regards to Nuvia has its reasons. Nuvia was supposed to be developing a server CPU, but now Qualcomm wants to use in smartphone SoCs, the license is supposed to depend on use case. ARM SoCs do not need to use ARM GPUs. At least that has not been the case until today.