Chinese semiconductor industry

hvpc said:

You are correct. Currently, domestic WFE other than litho has been qualified only up to 28nm. Even then, these domestic equipment are still not at parity with industry's tools of record. We still have ways to go to improve domestic equipment's tool reliability, throughput, and yield.

Click to expand...

You mean that SMIC currently does volume production on a 28nm line with all domestic WFE but ASML?

Well, if this is the case I would not use the expression "only up to 28nm". This is a huge achievement from any point of view.

This was considered pretty impossible only 2/3 years ago, not 10 years ago!

european_guy said:

You mean that SMIC currently does volume production on a 28nm line with all domestic WFE but ASML?

Well, if this is the case I would not use the expression "only up to 28nm". This is a huge achievement from any point of view.

This was considered pretty impossible only 2/3 years ago, not 10 years ago!

Click to expand...

No, I didn't say that SMIC is currently using non-litho domestic equipment for 28nm. I'm only saying, if push comes to shove, non-litho domestic equipment could support 28nm node. BUT not all equipment is at the same performance level of current tool or records.

hvpc said:

You referenced deceptive marketing message.

For any given node, there is a spectrum of easy to complex process steps. Just because a tool is used for one of plethora of process steps for a given node doesn't make it a "5nm etcher" that could handle all etch steps.

I can't reveal too much, but let me ask you, have you any clue which process step tsmc uses AMEC's etcher for? Do you really believe its for the most critical layer or process step?

Only Chinese WFE suppliers go out of their way to attach a process node designation to their equipment. They do this to create buzz to hype up their true capabilities. We don't typically do that in our industry. Only the process engineers, we know which model is good enough for which process step, application, node combination.

In our industry, tools are advertised more so by application and its performance spec. We don't typically use a "node" as an adjective/description for the equipment.

Click to expand...

But it wasn't really AMEC claiming that, it was revealed by a third party. AMEC itself didn't make that press release. And I had even discussed this exact point before, on this exact thread, that a given semiconductor tool can be used for a wide variety of processes, and even include the exaggerated case of 1990s refurbished AMAT and Lam tools.


As for where TSMC uses AMEC etchers, AMEC makes dry metal and dielectric etch. These are front end processes, and I don't think there's anything in the front end that's not critical.

"The Hikvision HikSemi CC700 is a solid-state drive in the M.2 2280 form factor, launched on October 31st, 2022. It is available in capacities ranging from 512 GB to 2 TB. This page reports specifications for the 2 TB variant. With the rest of the system, the Hikvision HikSemi CC700 interfaces using a PCI-Express 4.0 x4 connection. The SSD controller is the MAP1602A from Maxiotech, a DRAM cache is not available. Hikvision has installed 232-layer TLC NAND flash on the HikSemi CC700, the flash chips are made by YMTC. To improve write speeds, a pseudo-SLC cache is used, so bursts of incoming writes are soaked up more quickly. Thanks to support for the fast PCI-Express 4.0 interface, performance is excellent. The HikSemi CC700 is rated for sequential read speeds of up to 7,450 MB/s and 6,750 MB/s write; random IOPS reach up to 860K for reads and 690K for writes.
At its launch, the SSD was priced at 190 USD. The warranty length is set to five years, which is an excellent warranty period. Hikvision guarantees an endurance rating of 3600 TBW, a good value."

FairAndUnbiased said:

But it wasn't really AMEC claiming that, it was revealed by a third party. AMEC itself didn't make that press release. And I had even discussed this exact point before, on this exact thread, that a given semiconductor tool can be used for a wide variety of processes, and even include the exaggerated case of 1990s refurbished AMAT and Lam tools.

Click to expand...

Alright, let me rephrase from "deceptive marketing message" to "deceptive/inaccurate/misleading media report"

FairAndUnbiased said:

Chinese semiconductor industry

I guess this is the presentation you refer to: https://www.asml.com/-/media/asml/files/investors/investor-days/2021/asml-investor-day-2021_business-line-duv---ron-kool.pdf?rev=eea4dec5c9384d48a21bb12f913aca90 It was part of last year's investor days at ASML...

www.sinodefenceforum.com

As for where TSMC uses AMEC etchers, AMEC makes dry metal and dielectric etch. These are front end processes, and I don't think there's anything in the front end that's not critical.

Click to expand...

Over the past few months, I have gain better understanding of and developed respect of you and your expertise. But the question of how AMEC is really used at tsmc 5nm process is not something that can be debated via logic. I also have my own expertise in semiconductor process flow, but even I can not logic my way through the how's and why's behind why a fab may chose to utilize any fab equipment.

I trust my iterative investigation process and how I derived at the bits of info I shared about how AMEC etchers are utilized for 5nm process flow. It's your prerogative to believe your logical reasoning, media report over what I claimed. Or you could check with your contacts at tsmc, AMEC, AMAT/LAM, or other that may have better and more direct knowledge to fact check some media report.

To tell you the truth, I'm already surprised AMEC is even part of tsmc 5nm process flow. I understand AMEC is a low risk, more economical solution for the non-critical process step. If I am making tool-purchase decision, I personally wouldn't have introduced the need to manage yet an extra vendor/supplier over a little bit of CAPEX saving.

hvpc said:

Alright, let me rephrase from "deceptive marketing message" to "deceptive/inaccurate/misleading media report"

Over the past few months, I have gain better understanding of and developed respect of you and your expertise. But the question of how AMEC is really used at tsmc 5nm process is not something that can be debated via logic. I also have my own expertise in semiconductor process flow, but even I can not logic my way through the how's and why's behind why a fab may chose to utilize any fab equipment.

I trust my iterative investigation process and how I derived at the bits of info I shared about how AMEC etchers are utilized for 5nm process flow. It's your prerogative to believe your logical reasoning, media report over what I claimed. Or you could check with your contacts at tsmc, AMEC, AMAT/LAM, or other that may have better and more direct knowledge to fact check some media report.

To tell you the truth, I'm already surprised AMEC is even part of tsmc 5nm process flow. I understand AMEC is a low risk, more economical solution for the non-critical process step. If I am making tool-purchase decision, I personally wouldn't have introduced the need to manage yet an extra vendor/supplier over a little bit of CAPEX saving.

Click to expand...

Ok, that's fair. I don't know enough about the process at TSMC to know where AMEC equipment could be used. I am several steps removed from the fab, so those who are closer to the fab should know better. But, I also know that even 1990s AMAT and LAM equipment can be used in certain modernized R&D processes with quite small CD, so it is not unimaginable that, even if not at TSMC, even if not at 5 nm, AMEC equipment made 30 years later could be used in FinFET processes.

FairAndUnbiased said:

Ok, that's fair. I don't know enough about the process at TSMC to know where AMEC equipment could be used. I am several steps removed from the fab, so those who are closer to the fab should know better. But, I also know that even 1990s AMAT and LAM equipment can be used in certain modernized R&D processes with quite small CD, so it is not unimaginable that, even if not at TSMC, even if not at 5 nm, AMEC equipment made 30 years later could be used in FinFET processes.

Click to expand...

hvpc said:

Alright, let me rephrase from "deceptive marketing message" to "deceptive/inaccurate/misleading media report"

Over the past few months, I have gain better understanding of and developed respect of you and your expertise. But the question of how AMEC is really used at tsmc 5nm process is not something that can be debated via logic. I also have my own expertise in semiconductor process flow, but even I can not logic my way through the how's and why's behind why a fab may chose to utilize any fab equipment.

I trust my iterative investigation process and how I derived at the bits of info I shared about how AMEC etchers are utilized for 5nm process flow. It's your prerogative to believe your logical reasoning, media report over what I claimed. Or you could check with your contacts at tsmc, AMEC, AMAT/LAM, or other that may have better and more direct knowledge to fact check some media report.

To tell you the truth, I'm already surprised AMEC is even part of tsmc 5nm process flow. I understand AMEC is a low risk, more economical solution for the non-critical process step. If I am making tool-purchase decision, I personally wouldn't have introduced the need to manage yet an extra vendor/supplier over a little bit of CAPEX saving.

Click to expand...

I think the consistent difference between you two is one is judging adoption and progress from the standpoint of “sufficiency” and the other is judging from the standpoint of “competitiveness”. Speaking personally I think it’s pretty clear which standard I’m preferential to wrt how the Chinese semis industry should be observed, but maybe explicitly laying out these standard of judgment will help clarify the analysis either of you are presenting.

FairAndUnbiased said:

Ok, that's fair. I don't know enough about the process at TSMC to know where AMEC equipment could be used. I am several steps removed from the fab, so those who are closer to the fab should know better. But, I also know that even 1990s AMAT and LAM equipment can be used in certain modernized R&D processes with quite small CD, so it is not unimaginable that, even if not at TSMC, even if not at 5 nm, AMEC equipment made 30 years later could be used in FinFET processes.

Click to expand...

Oh, I'm certain AMEC etcher is much more capable than what tsmc is using it for. I didn't hear from domestic fab specifically domestic etcher is already qualified for 14nm FinFET. But I'm sure AMEC etchers are in the lead in terms of primary domestic etch solution to support fab's effort to qualify indigenous equipment for 14nm

I suspect the problem AMEC is facing is less the capability variety, but more the reliability (uptime), defectivity, etch chamber-chamber fingerprint matching, wafer-to-wafer, lot-to-lot, day-to-day drift varieties. Basically, issues associated with managing a large fleet of AMEC etch chambers in an HVM setting.

hvpc said:

No. SMIC accounts for 5% of foundry revenue NOT 5% of global chip sales.

Click to expand...

So, are you saying non-SMIC customers mark up their chip prices so much more than SMIC customers that 5% global foundry revenue leads to end products only been like 1% global chip sales?

I have no idea how they measure these things, but there are a whole lot of new Chinese designed chips that are not actually made by Chinese fabs. do they only count chips that are designed and made by Chinese firms in that 6.6%?

hvpc said:

You are correct. Currently, domestic WFE other than litho has been qualified only up to 28nm. Even then, these domestic equipment are still not at parity with industry's tools of record. We still have ways to go to improve domestic equipment's tool reliability, throughput, and yield.

Click to expand...

That's not what I'm saying at all.
I don't understand why they need to have same reliability, throughput and yield as their foreign counterpart? If there is no alternative to them, then Chinese fabs have to use them.

hvpc said:

SN1 expansion is basically completed. No more scanners needed and can't get access to American process equipment. Without American equipment, SN1 is unlikely to expand its wafer output; the bottleneck is limited by amount of American process equipment already purchased.

Click to expand...

Q：Lam在中芯南方的support团队后面也全撤？但是中芯南方本来也拿不到Lam的设备？

A：对。中芯南方做比较多7nm、14nm。虽然中芯南方拿不到Lam的设备，但还有一小部分人在中芯南方做服务。

Click to expand...

From the Lam interview. They have very few employees serving at SMSC. It took 3 years for SMIC to find the tools to ramp up production at SN1 to probably 30k wpm. That's really slow. Seems like they were really hard pressed to get any American tools there.

hvpc said:

Your previous "predictions" on what SMIC's extra 1.6B CAPEX is for was not correct.

Click to expand...

SMIC said there were 2 things they were for. 1) ASML scanners 2) low production items that require long lead time. Which one do you think is taking bulk of the $1.6 billion Capex?

Let me ask you something. Huahong Capex for 2023 is like $1 billion (let's say 2022 is around the same) and it's adding probably 30k wpm 12-inch wafers next year. SMIC is maybe adding 50k wpm 12-inch wafers next year. It's Capex is $6.6 billion. How is SMIC spending 6 times as much Capex when its capacity expansion is just 60% more? It's basically spending 4x as much per wafer capacity added as Huahong. Is SMIC just throwing money around and wasting them? Huahong fab 7 was mostly adding 55-90 nm chips. Is 28 to 65 nm Capex supposed to be 4x of that? Seems unlikely based on the charts that everyone have been showcasing.

hvpc said:

Besides, like I said, the bottleneck at SN1 is not scanners. By the way, your prediction on NXT2050/2100 is too bullish. There aren't that many NXT2050 and there is no NXT2100 in China.

SN2 will not ramp for another year.

Click to expand...

ASML said they delivered their first 2100i in Q3. Are you telling them there is a fab outside of China that needs 2100i? I'm not saying they bought that many NXT2050/2100i. Based on what others and ASML said, you don't need 2050/2100i outside of a few steps. I'm saying they've spent way more money per year on ASML scanners than they need to do if they were just adding 50k wpm of mature process capacity.

We know for example, they spent $1.2 billion for the year before March 2021 on ASML scanners. We know they spend around $4.3 billion in Capex in both 2020 and 2021. Their original Capex plan was $5 billion for this year. So based on that, they were originally planning to spend around $1.5 billion in ASML scanners this year. Now since they added $1.6 billion to that for both early deposit and probably early deliveries. Let's say that will lead to $4.5 to 5 billion Capex with ASML in 2022/2023. Based on the cost of these machines. That seems way more than what they need for adding capacity at the new Beijing plant. They could be stocking up some machines or they could also be adding more capacity at SMSC fab.

hvpc said:

You think wrong, bud. Since SMIC could still buy ASML scanners, and you are aware and agree domestic non-litho equipment are at 28nm level.......American WFE is indeed the limiting factor for SMIC's SN1 & SN2.

Click to expand...

I don't agree with that at all. I think their slow pace of expansion at SN1 after 2020 was due to not being able to get the American equipment they needed, hence they had to get less reliable/efficient domestic tools and take their time to ramp up with them.

And I think they've been really quiet about what they've been doing there, because it does them no good to talk about their advanced node achievement.

hvpc said:

With a full year of financial data, we will obviously see a great Y/Y revenue uptick. But, the revenue will definitely not reflect the full 35K wafer per month worth of revenue... Capacity in the beginning of the year is obviously lower than the second half of the year, so my guess we are more likely to see revenue reflecting 120K 14nm & 60K 7nm wafers for 2022.

Click to expand...

I think we can agree that they were not at 35k wpm for the entire year. I'd be surprised if they are at 35k wpm right now actually.

tonyget said:

Please, Log in or Register to view URLs content!

Chinese 128-Core CPU World Record Expelled From Rankings: No Availability​

View attachment 102736

The SPEC Committee has expelled the Chinese-produced Yitian 710's world record result in the CPU 2017 benchmark, an industry benchmark for CPUs, from its ranking, citing availability issues. We were impressed when Alibaba introduced its 128-core processor made on TSMC's N5 node about a year ago. We were even more impressed when this CPU broke some records about five months ago. However, we felt a bit troubled due to the limited availability of preview samples, even in Alibaba's cloud.

"SPEC has determined that this result does not comply with the SPEC OSG Guidelines for General Availability and the SPEC CPU 2017 run and reporting rules," a statement by SPEC reads. "Specifically, the submitter has notified SPEC that General Availability requirements were not met."

We don't know whether the lack of availability occurred because of the United States' curbs against the Chinese semiconductor sector, but that could be the source of the issue.

Alibaba said the Yitian 710 can hit a 440 integer score in SPECint2017, which is comparable to the performance of a two-socket machine powered by Intel's 32-core Xeon Platinum 8362 (64 cores at 2.80 GHz). This remarkable score was allegedly obtained on hardware that was barely available even to users of Alibaba's cloud. The problem for Alibaba is that this level of performance is enough to fall under the new U.S. export rules.

(Image credit: Alibaba)
The Yitian 710 server SoC contains about 60 billion transistors, making it one of the most complex processors ever developed. TSMC makes it on one of its N5 (5nm-class) production nodes, which falls under the U.S. sanctions against Chinese semiconductor and supercomputer sectors.

And while American sanctions against the Chinese semiconductor industry are well-known, there is another wrinkle in this story — yields. A 60-billion transistor chip would be prone to poor yields, meaning that a higher percentage than normal of the chips produced will end up in the waste bin. Alibaba indirectly confirmed that its SoC doesn't perform up to the desired spec when it offered a selected few to try an underspecc'd SoC in its cloud. Perhaps yield issues are why it did not procure enough chips to deploy. The lack of availability appears to be why the Yitian 710 was expelled from the SPEC rankings.

Click to expand...

Alibaba just came out a few weeks ago and said 20% of their new cloud computing capacity will be using their own chips like Yitian-710. They've also talked extensively about how much better Yitian-710 is over their competition. I think toms hardware might be overthinking things a little bit.