Chinese semiconductor thread II

[tphuang]

I'd like to know how many Ascend GPU has Huawei sold last year

my base line assumption is minimum of 20k/month

Of course when you put together AI servers, you also need Kunpeng CPUs with that. So we need at least 1000 wpm for Ascend (probably more) and another 1000 wpm for kunpeng (probably more).

To put it this way, Huawei is building a new 16 EFLOPS data center and Shenzhen local govt is building a new 100k card data center. So that's 150k card just in these 2 projects and almost all of that would be using Ascend NPUs.

 

[european_guy]

Please, Log in or Register to view URLs content!

SMIC reported results

revenue finally up vs 2022 in Q4, but margins down to 16.4%

Expecting flat to 2% jump in revenue in Q1
just 9-11% margins

avg utilization of 75%
806k wpm of 8-inch capacity by end of 2023

Expecting capex in 2024 to be same as 2023

So SMIC continues to spend a lot for its size, about $7.5B last year so expecting $15B in two years

Here I add also the data point on

Please, Log in or Register to view URLs content!

An unexpected small bump in US share in last quarter, but the trend is downward as expected.

This is the full official results and presentation

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

Almost 10K wpm of new capacity added in the last quarter!

 

[tacoburger]

SMEE isn't responsible for the EUV effort, that's a collection of Chinese institutions.

SMEE has nothing to do with EUV
SMEE has nothing to do with EUV
SMEE has nothing to do with EUV
SMEE has nothing to do with EUV
SMEE has nothing to do with EUV

SMEE may not be as well funded or talented as the bunch of companies and institutions working on EUV but don't forget that SMEE has two decades of experience in lithography systems, has actual working lithography products and has been working on DUVi for more then a decade. The fact that they still can't mass produce cutting edge DUVi machines yet says a lot.

Don't forget that while EUV is nice, the mature nodes using DUV is still the backbone of the economy, it's more of priority for China to get domsetic DUVi then EUV.

Sure companies like Huawei is competent as hell, but it's still an area that they have no prior experince in and they only got roped in after 2018 after the sanctions dropped. The fact that DUVi is taking so long says a lot about how bumpy the road to EUV is. After all, SMEE depends on lots of vendors and suppliers that will no doubt also will contributing to the EUV project due to the similarity of some of the competents. How can you develop EUV so fast when the DUVi project and other lithography systems are so undeveloped? I bet that mountains of money and talent also poured into SMEE after 2018, surely the highers up in China recognised the threat of America getting ASML to stop DUV exports. And 6 years later, their SSA800 might finally go into mass production this year maybe? Not a good sign.

This things don't develop out of nowhere, you have to build the industry up. It's like a country trying to developing 5th gen fighters when their 4th gen fighter program can barely get off the ground. It's not impossible, but like DUVi machines, I think it's going to take a lot longer then what many people in this thread think.

 

[Blitzo]

Bigger die have lower yields compare to smaller die. The die size of Ascend GPU should be bigger than Kirin SOC

Yes, that is what I was alluding to in my reply

This things don't develop out of nowhere, you have to build the industry up. It's like a country trying to developing 5th gen fighters when their 4th gen fighter program can barely get off the ground. It's not impossible, but like DUVi machines, I think it's going to take a lot longer then what many people in this thread think.

The thing is that everything you described here is what most of the regular and long time contributors to the thread are aware of, and it is in context of that where the EUV estimates and projections are coming from.

We will know in the next two or so years just where their EUV trajectory will likely land, but my observation is that it's probably not useful to see DUV and EUV system development in the PRC lithography industry context as sequential but rather semi parallel.
Using the fighter aircraft analogy, current DUV efforts are like J-10C/J-16, while EUV is like J-20 -- the two different generations of aircraft ultimately entered service alongside one another within a year or so despite being fundamentally different, because they were also developed relatively in parallel. (The analogy of course can only be taken so far)

 

[olalavn]

Both IMEC and ASML have been verified, SAOP can achieve a spacing of 16nm, and LER is excellent. In theory, SMIC can catch up with the density using DUV

Since LER will cause increased resistance and uneven current distribution, EUV has a random LER effect, but DUV does not, which is a blessing in disguise.

 

[sunnymaxi]

SMEE may not be as well funded or talented as the bunch of companies and institutions working on EUV but don't forget that SMEE has two decades of experience in lithography systems, has actual working lithography products and has been working on DUVi for more then a decade. The fact that they still can't mass produce cutting edge DUVi machines yet says a lot.

CIOMP has more experience in lithography system than SMEE. they are leading from the front in EUV.

CIOMP infact has EUV prototype in their Lab since 2017. they have been working on this system since 2010..



CIOMP/SIOMP along with Harbin Institute of Technology and other players have successfully developed all components of EUV. first prototype assembling is underway or soon process will start.

-----------------------------------------------------------------------------------------

you said ''SMEE cannot mass produce of DUVi right now''

small scale production has begun. as i said earlier production will gradually increase. its because suppliers are unable to meet the current demand. they are in process to increase the production capacity. it will take time to move machine and skilled workers.

RSLaser , U-Precision , Changchun Guoke precision and Nanjing Mloptics are known suppliers of SMEE .. all these firms currently expanding the production just to accommodate the SMEE serial production plans. this year SMEE is all set to produce 20 sets of Lithography machines. dry+immersion..

---------------------------------------------------------------------------------------------
EUV components suppliers are different.

 

[tokenanalyst]

SMEE may not be as well funded or talented as the bunch of companies and institutions working on EUV but don't forget that SMEE has two decades of experience in lithography systems, has actual working lithography products and has been working on DUVi for more then a decade. The fact that they still can't mass produce cutting edge DUVi machines yet says a lot.

SMEE was a child of Chinese research institutions, the same institutions the are now developing china EUV lithography machine, Tsinghua -UPrecision, Harbin ( maglev wafer wages, grating interferometers, heterodyne interferometers, reticle stage ), Cheertech-Zhenjiang (immersion, ultrapure fluid systems), CIOMP-SIOM-CAS-CNEPO (highNA optics, projection system, Masks), RSLASER-BIT (ArFi excimer lasers)

These same institutions are now developing the EUV project separated from SMEE, because the main goal of SMEE is to get their DUV machines up to ASML 2000i machines.

Don't forget that while EUV is nice, the mature nodes using DUV is still the backbone of the economy, it's more of priority for China to get domsetic DUVi then EUV.

Sure companies like Huawei is competent as hell, but it's still an area that they have no prior experince in and they only got roped in after 2018 after the sanctions dropped. The fact that DUVi is taking so long says a lot about how bumpy the road to EUV is. After all, SMEE depends on lots of vendors and suppliers that will no doubt also will contributing to the EUV project due to the similarity of some of the competents. How can you develop EUV so fast when the DUVi project and other lithography systems are so undeveloped? I bet that mountains of money and talent also poured into SMEE after 2018, surely the highers up in China recognised the threat of America getting ASML to stop DUV exports. And 6 years later, their SSA800 might finally go into mass production this year maybe? Not a good sign.

This things don't develop out of nowhere, you have to build the industry up. It's like a country trying to developing 5th gen fighters when their 4th gen fighter program can barely get off the ground. It's not impossible, but like DUVi machines, I think it's going to take a lot longer then what many people in this thread think.

The main proof that Chinese research institutions are developing an EUV machine apart from SMEE "because lets be frank nobody likes to read the past post in this thread" is because the already tested an EUV machine in 2017.

EUV is an independent project from SMEE DUV machines and will be make by a consortium of companies that include Huawei, SMIC, Shanghai ICRD and others. I think it's going to take a lot less then what many people in this thread think.

 

[Blitzo]

The main proof that Chinese research institutions are developing an EUV machine apart from SMEE "because lets be frank nobody likes to read the past post in this thread" is because the already tested an EUV machine in 2017.

I must say, the forum itself is not very conducive to searching up and looking for past useful posts. The search function of this forum is legitimately broken, and one often has to use google to indirectly search SDF, and that is often imprecise as well.

For newcomers it is often a difficult topic to get caught up on (or indeed, other topics in the PLA watching parts of the forum either)

 

[tacoburger]

I already know that SMEE is separate from the EUV project. The problem is that SMEE has been a company since 2002, created for the sole purpose of developing lithography products, failed to develop any competitive products, and despite DUVi being a mature technology, still can't mass produce them despite the very real threat of ASML cutting China off of DUV exports, forcing China to stockpile tens of billions of ASML's DUVi machines.

This matters because it represents the state of the lithography development in China. It's a niche market and a large amount of the trained lithography experts and engineers in China would already be working in SMEE. Yeah, various different companies and institutions are working on EUV, but they're not magic. If they were, SMEE wouldn't be in this current state of limbo for the last 20 years. Don't forget that SMEE depends heavily on various suppliers, if the domestic chinese optics/mirrors/lasers suppliers are somehow good enough to suddenly make EUV systems, then SMEE won't be in such a wretched state for so many decades.

So what difference will all of this new companies/institutions make? If they are so good, why didn't they go help out SMEE in their development 10 years ago, or 6 years ago when the sanctions made the threat of even basic DUV machines being cut of very real? Producing a prototype is very different from mass production too. Again with the fighter analogy, if your 4th gen fighters can barely get off the ground, your entire aviation industry isn't magically gonna upgrade itself just because you got a bunch of different companies to develop your 5th gen fighter.

So what if they test a EUV machine in 2017? SMEE was in the market since 2002, trying it's best to develop a whole range of lithography products, failed to get market share or develop a competitive product and can't even get it's SSA800 into mass production when China needed it the most. Again, if all this chinese institutions like Tsinghua/Harbin had super geniuses who could develop the vastly more complicated EUV systems in a fraction of the time, SMEE wouldn't be struggling, they would have dominated the market from the get go.

 

[tokenanalyst]

The only way I can see of stopping China from creating not only a mere EUV and DUV immersion lithography machine but an entire EUV and DUVi lithography supply chain that includes the mass production of wafers-reticle stages, DUV and EUV optics and mirrors, ultrapure water systems, EUV and DUV light sources, advanced positioning sensors, lithography machines parts and components, overlay metrology machines, OPC software, CDSEM and others ADI and AEI machines, EUV-DUVi photoresist, DUVi and EUV mask manufacturing and so on and on, is:

If ASML sells then their EUV and DUVi machines with not restrictions or a nuclear war, beyond that I don't see any way.