Chinese semiconductor thread II

leibowitz

Junior Member
I only joined this forum last year and haven't seen earlier posts.

But I've been aware of what you mentioned for some time—similar perspectives exist in China. The real question is: What’s the current resolution status? What’s today’s actual progress metric?

As an external observer, I consider this source exceptionally precise in characterizing China's semiconductor industrial status—concretely defining completion milestones versus remaining gaps (and where those gaps persist).

For instance, amid rampant EUV speculation, his detailed specification of one fact stands out: China achieved 5nm through extreme DUVi enhancements, not premature EUV adoption.

He acknowledged significant EUV breakthroughs but also underscored that the U.S. decisively severed China's EUV access in 2019, immediately enforcing comprehensive and effective embargoes across all EUV-related domains. Thus, expecting near-term EUV production deployment is unrealistic—there are foundational gaps to address first.

I can't vouch for his absolute accuracy, but his demonstrable work convinces me he engages regularly with China’s core semiconductor institutions/teams—though not at operational-level understanding. Crucially, his sourcing is highly credible, and he willingly clarifies technical principles in depth.

I don’t regard his technical expertise as superior to mine or others here. Yet his video revealed startling insights:

Mastery of cutting-edge semiconductor manufacturing workflows
Profound understanding of systemic challenges
Structured problem-solving frameworks
Having tracked microelectronics/semiconductor/lithography literature for over a decade, I grasp most core concepts. While perhaps less technically adept than many here, I'm far from credulous.

Countless domestic EUV/DUV claims surfaced recently—I remained silent, skeptical of all. But I trust his account. Why?

Not merely by revealing outcomes, but by:

➤ Exposing root constraints
➤ Detailing problem-solving frameworks
➤ Honestly stating: “We’re bottlenecked here—resolving this takes time”

Ultimately, that’s what ultimately builds credibility.
1. Who is this source?
2. Stop using AI to write your posts
 

european_guy

Junior Member
Registered Member
Big if true..

as per South Korean Media, China's CXMT Begins Supplying HBM3 to Huawei, Threatening to Reshape the Memory Market into a "Big 4"

Please, Log in or Register to view URLs content!

Hidden in the above article there was this little gem

Huawei plans to procure the graphics processing units (GPUs) that comprise the AI semiconductors, along with HBM3, from a Chinese fabless semiconductor design company. Huawei previously unveiled its own AI semiconductor, the "Ascend 910C."

It means that Huawei will buy AI chips from fabless companies, like for instance Cambricon, Moore Threads, etc. and then integrate those chips in their systems. It is the first time I read Huawei to not use own Ascend chips for their AI hardware solutions.

If confirmed, I would read the above together with the very recent open sourcing of CANN, their CUDA alternative AI software. Technically those non Huawei chips will very probably run on CANN too because they have to be integrated in the HW stack.

But apart from the techncal meaning, there is a clear strategic decision behind all this. A new strategic decision materializing only in recent months, a decision from China I guess, not from HW alone.

On a wider scope, the decision to open source AI models with related papers from mainly all Chinese big actors (Qwen-Alibaba, Bytedance, Tencent, Huawei,etc) goes along the same direction.

It seems that, to overcome the current unbalance in hardware access between China and US, 众人拾柴火焰高 is the way to go.
 
Last edited:

european_guy

Junior Member
Registered Member
Micron will stop developing mobile NAND products worldwide

Please, Log in or Register to view URLs content!

Goodbye Micron!

If confirmed Micorn will exit smartphone NAND market, leaving it to Korea (Samsung, SK) and China YMTC

After Chinese smartphone firms ditched Micron, the only potential customers remaining are Samsung and Apple, but Samsung uses its own NAND.

So Apple alone is not a good enough reason to keep the mobile NAND branch...we will see if this is finalized or Trump will make them change their mind (very possible).
 

european_guy

Junior Member
Registered Member
New Cambricon chip MLU 690


It seems Cambricon is currently in production with their new chip called MLU 690

- Current production is of 1000 wpm (wafers per month) at SMIC (probably on 7nm)

- According to rumors monthly volume is 20,000 units

It means 20 good chips per wafer and assuming a yield of 50% the tweet author extrapolated MLU 690 is nearly 900 mm2.

I have verified myself:

1755002522319.png

With yield of 50% it should be around 30 good pieces per wafer, so it can be MLU 690 is even bigger or yield is a bit worse.

Assuming 900 mm2 per chip and 20 good dies per wafer -> yield 35%

But apart from the yield, a new AI chip of 900 mm2 or even larger is quite big....very big indeed!

A 900mm2 NPU on 7nm node it means a lot of TOPS, a lot of computation power!
 

tphuang

General
Staff member
Super Moderator
VIP Professional
Registered Member
New Cambricon chip MLU 690


It seems Cambricon is currently in production with their new chip called MLU 690

- Current production is of 1000 wpm (wafers per month) at SMIC (probably on 7nm)

- According to rumors monthly volume is 20,000 units

It means 20 good chips per wafer and assuming a yield of 50% the tweet author extrapolated MLU 690 is nearly 900 mm2.
I find the original tweet to be quite suspect. His rumors are quite wild in general.


SMEE spin off AMIES announces that it has hit 500 deliveries of advanced packaging scanner. 90% domestic market share.
 
Top