Chinese semiconductor thread II

[gelgoog]

Is Ymtc's Market Share Falling Due To U.S. Semiconductor Equipment Regulatory Tongue?

That article is bogus. There is AMEC which is a Chinese company which produces plasma etch tools suitable for 3D NAND. AMEC increased revenue by 50% last year.
What could have happened is that YMTC is taking time changing its processes to Chinese equipment suppliers. But I think YMTC is not a publicly traded company, they don't need to publish revenue numbers, so this is all speculative.

There are also reports that Phase III development of YMTC's Wuhan fab is underway.

Please, Log in or Register to view URLs content!

Last I looked at Sentinel-2 pictures they seen to be preparing the ground for its construction already.

 

[gelgoog]

Please, Log in or Register to view URLs content!

Yangtze Memory Breaks Ground on Phase-III Project, Aims to Capture 15% Global NAND Market Share with Pricing Power in Sight
China’s domestic memory chip industry has achieved a significant milestone. According to Tianyancha APP and official corporate information, Yangtze Memory Phase-III (Wuhan) Integrated Circuit Co., Ltd. was officially established on September 5, 2025, with a registered capital of RMB 20.72 billion. It is owned 50.19% by its parent company, Yangtze Memory Technologies Co., Ltd. (YMTC), and 49.81% by Hubei Changsheng Phase-III Investment Development Co., Ltd., an entity backed by Hubei state-owned capital. The legal representative is Chen Nanxiang, Chairman of Yangtze Memory.

This move is widely regarded as a key signal of YMTC's Phase-III capacity expansion. Public data indicates that YMTC's current monthly production capacity is close to 130,000 wafers, accounting for approximately 8% of the global 3D NAND market. The company plans to increase its monthly capacity to 150,000 wafers by 2025 and aims to capture 15% of the global NAND flash supply by the end of 2026. According to earlier environmental impact assessment information, its overall planned capacity is 300,000 wafers per month, with each phase targeting 100,000 wafers. The establishment of this Phase-III entity signifies that YMTC is steadily advancing toward doubling its total production capacity, further addressing the urgent domestic demand for independently controllable storage products.

Since its inception in 2016, YMTC has consistently focused on the research, development, and industrialization of 3D NAND flash technology. Even after being added to the U.S. Entity List in late 2022 and facing restrictions on equipment imports, the company has persisted in pursuing independent innovation. Its self-developed Xtacking® architecture has been iterated to version 4.0, successfully achieving mass production of 3D NAND Flash with over 232 layers. Currently, YMTC has begun shipping 232-layer TLC chips (X4-9070), which utilize a double-stacked structure to achieve a total of 294 layers and an interface speed of 3600 MT/s. The product performance has reached international first-tier standards.

From a market perspective, global NAND flash demand is gradually recovering, with some international brands already initiating price hikes. Additionally, China accounts for over 30% of global NAND market sales, providing solid market momentum and strategic significance for YMTC’s capacity expansion. If the monthly production capacity reaches 200,000 wafers, YMTC will possess the pricing power to influence global NAND flash market trends.

It is worth noting that while advancing capacity expansion, YMTC is accelerating the localization of its supply chain. According to reports, the company plans to initiate trial production on its first fully domestically equipped production line in the second half of 2025. This move aims to reduce reliance on foreign equipment and is a critical step in responding to U.S. export controls. Currently, the localization rate of YMTC’s equipment has reached 45%, significantly higher than that of other major domestic wafer fabs.

 

[gelgoog]

This chart is from last year. It shows how Nexchip gobbled up the LCD DDIC market.


LCD driver IC (HV55-HV90nm) and AMOLED driver IC (HV28-HV40nm)wafer input market share in 3Q24

Please, Log in or Register to view URLs content!

Nexchip and SMIC from China had like 2/3rds of the LCD DDIC market. And the rest was Taiwan.

Nexchip started producing AMOLED DDICs this year. It will be interesting.

 

[LanceD23]

Any one heard the news of China freezing the imports of foreign semiconductor equipments and chips.
17 of ASML scanners got freezed in shanghai port as well many US chips valued at 2550 Billion Yuan

中國突然按下芯片進口暫停鍵，2550 億元貿易一夜停擺，全球半導體產業瞬間墜入寒冬 ——ASML 股價暴跌 14% 創 17 年最大跌幅，高通庫存暴增、英特爾工廠停產，連德國研發中心都被迫停工。這從非偶然，而是中美歐科技博弈的關鍵信號。 本影片將拆解這場產業震盪的核心邏輯，看中國為何敢逆勢停購，自主化產能又如何撐起市場，揭開全球半導體格局重組的真相。 在這部影片中，你將看到： 中國為何突然暫停 2550 億芯片進口？是被動應對還是主動佈局？ ASML 暴跌、高通庫存暴增，全球半導體產業的「蝴蝶效應」有多恐怖？ 中國自主化產能能否頂住？中芯 28 納米良率 92%、長江存儲量產，夠不夠用？ 這場博弈的終局是什麼？全球半導體供應鏈會如何重組？

 

[huemens]

Any one heard the news of China freezing the imports of foreign semiconductor equipments and chips.
17 of ASML scanners got freezed in shanghai port as well many US chips valued at 2550 Billion Yuan

中國突然按下芯片進口暫停鍵，2550 億元貿易一夜停擺，全球半導體產業瞬間墜入寒冬 ——ASML 股價暴跌 14% 創 17 年最大跌幅，高通庫存暴增、英特爾工廠停產，連德國研發中心都被迫停工。這從非偶然，而是中美歐科技博弈的關鍵信號。 本影片將拆解這場產業震盪的核心邏輯，看中國為何敢逆勢停購，自主化產能又如何撐起市場，揭開全球半導體格局重組的真相。 在這部影片中，你將看到： 中國為何突然暫停 2550 億芯片進口？是被動應對還是主動佈局？ ASML 暴跌、高通庫存暴增，全球半導體產業的「蝴蝶效應」有多恐怖？ 中國自主化產能能否頂住？中芯 28 納米良率 92%、長江存儲量產，夠不夠用？ 這場博弈的終局是什麼？全球半導體供應鏈會如何重組？

This is garbage content that gets rehashed by a bunch of AI generated youtube channels every few weeks, that has been going on for years.

 

[tokenanalyst]

This machine does 300 wafers per hour. It is 50% faster than older machines. And right now China still cannot produce enough lithographers to meet its own demand.

20% more througthput
Nobody knows how many machines China is making right now, like I said they have a entire supply chain for it.

Dry ArF has not a lot use outside mature nodes, an area that everyone in the industry in China wants to localize badly. So there is push for local KrF and Dry ArF lithography machines.

Either way if Chinese companies have to choose between Nikon and ASML they will choose ASML, their holistic approach to lithography is much better than Nikon, even China metrology tools are made to work better with ASML scanners than Nikon in a Holistic approach. China domestic lithography scanners are made to more compatible with ASML than Nikon. Mix and Match overlay will be probably be easier between Chinese and ASML tools than Nikon. And in top of that is not that Nikon tools are less susceptible to export controls than ASML tools.

 

[tokenanalyst]

DeepSeek and Cambricon have joined forces to release the DeepSeek-V3.2 model architecture and source code for the vLLM-based model adaptation.​

DeepSeek released its next-generation model architecture, DeepSeek-V3.2, garnering widespread industry attention. Notably, at the same time as the release of this large-scale model, Cambricon also officially announced its adaptation to DeepSeek-V3.2 and open-sourced the source code for its large-scale model inference engine, vLLM.

DeepSeek has officially released the DeepSeek-V3.2-Exp model, an experimental version. As an intermediate step toward a new generation of architecture, V3.2-Exp builds on V3.1-Terminus by introducing DeepSeek Sparse Attention (a sparse attention mechanism) to exploratoryly optimize and validate the training and inference efficiency of long texts.

Currently, the DeepSeek official app, website, and mini-program have all been updated to DeepSeek-V3.2-Exp. At the same time, the API price has been significantly reduced. We welcome users to test their experience and provide feedback.

Cambricon has also adapted DeepSeek's latest model, DeepSeek-V3.2-Exp, and has open-sourced the source code for its large-scale model inference engine, vLLM-MLU. The code address and testing steps are at the end of this article. Developers can experience the highlights of DeepSeek-V3.2-Exp firsthand on Cambricon's software and hardware platforms.

In the view of industry experts, this "high-energy" operation of simultaneous release and adaptation carries an important signal of deep collaboration among China's top technology companies.

From a technical perspective, this new large model is quite substantial, reaching 671GB. Industry estimates suggest that, under ideal conditions with ample bandwidth, downloading this large model alone would take approximately 8-10 hours. Furthermore, adapting the chip to this large model involves a complex set of tasks, including optimizing underlying architectures, matching computing resources, and debugging compatibility. This is by no means something that can be accomplished in a short period of time.

A senior expert in the AI industry pointed out: "Such a rapid adaptation response fully demonstrates that Cambricon had already started adaptation research and development before the release of DeepSeek-V3.2, and the two sides had already quietly started communication and collaboration on the technical level. Both parties are low-key and pragmatic companies. They kept it secret in the early stage, and it is estimated that they have already moved towards each other." Rejecting hype and focusing on technology is a typical embodiment of the current style of China's top technology companies.

Please, Log in or Register to view URLs content!

 

[tokenanalyst]

Zhaoxin's new generation KH-50000 processor releases more specifications and details​

Zhaoxin's official website announced more specifications and details of the Kaisheng KH-50000 series server processors, presenting a "technological gift" on the eve of the 76th anniversary of the founding of the People's Republic of China. Let's interpret the five core highlights.

Kaisheng's KH-50000 series processors use the industry's advanced chiplet packaging technology. A single processor can integrate 12 CPU dies and provide two products: 96-core and 72-core. The base frequency is 2.2GHz (96-core version) / 2.6GHz (72-core version), and the maximum operating frequency is 3.0GHz. Both have up to 384MB of L3 cache, and the computing density has been significantly improved compared to the previous generation product, the Kaisheng KH-40000 series (up to 32 cores per socket).


*Chiplets.

IO upgrade to support mainstream DDR5/PCIe5 interface

Kaisheng's KH-50000 series server processors offer comprehensive memory upgrades, supporting 12 channels of DDR5 5200 ECC memory with a maximum capacity of 3TB. They also offer 128 PCIe 5.0 lanes (supporting CXL and proprietary ZPI interconnects), 16 PCIe 4.0 lanes, four USB 3.2 Gen2 lanes, and 12 SATA 3.2 high-speed interfaces, bringing I/O count and standards to the forefront of international standards.

Efficient interconnect technology provides higher bandwidth, lower latency and power consumption

Kaisheng's KH-50000 series server processors support dual-channel or quad-channel efficient expansion interconnection. A single server can be equipped with up to four KH-50000 processors. With the help of the new generation of self-developed ZPI 5.0 technology, it supports the construction of a high-performance computing platform with up to 384 cores with higher data bandwidth, lower latency and power consumption, greatly expanding the application space of the server.

Continuously deepen security efforts to safeguard information security in key areas

In terms of security and reliability, Kaisheng's KH-50000 series processors support self-developed SM2, SM3 and SM4 national encryption acceleration instructions, as well as security technologies such as secure boot and key management, and further enhance the server RAS (reliability, availability and maintainability) functions.

Please, Log in or Register to view URLs content!

 

[tokenanalyst]

Jingsheng leads the way: the first 12-inch silicon carbide substrate processing pilot line is officially put into operation.​

The first 12-inch silicon carbide substrate processing pilot line was officially put into operation at Zhejiang Jingrui SuperSiC, a subsidiary of Jingsheng Machinery Electronics Co., Ltd. This marks that Zhejiang Jingrui SuperSiC has truly achieved independent research and development of a full range of equipment from crystal growth, processing to testing, with 100% localization. This marks that Jingsheng has moved from running side by side to leading in global SiC substrate technology, and entered a new stage of efficient intelligent manufacturing.



SiC, a core representative of third-generation semiconductor materials, is widely used in key industries such as new energy vehicles, smart grids, and 5G communications due to its high-voltage resistance, high-frequency, and high-efficiency properties. Simultaneously, SiC is becoming a key material driving technological breakthroughs in emerging applications such as AR devices and CoWoS advanced packaging intermediate substrates. Compared to 8-inch products, 12-inch products increase single-wafer chip output by approximately 2.5 times, significantly reducing the unit cost of crystal growth, processing, and polishing in large-scale production, and are a key path to significantly reducing downstream application costs.


End to End SiC manufacturing tools

Zhejiang Jingrui SuperSiC's pilot line covers the entire process of crystal processing, cutting, thinning, chamfering, grinding, polishing, cleaning, and testing. All processes utilize domestically produced equipment and proprietary technologies. Core processing equipment such as high-precision thinning machines, chamfering machines, and double-sided precision grinders are the result of years of in-house research and development, achieving industry-leading performance. With this achievement, Jingsheng Electromechanical has established a complete closed-loop system for 12-inch SiC substrates, from equipment to materials. This has completely eliminated the risk of bottlenecks in critical equipment and set a new benchmark for cost and efficiency for downstream industries.

Zhejiang Jingrui SuperSiC, a subsidiary of Jingsheng Electromechanical, specializes in the research, development, and production of compound semiconductor materials, including silicon carbide and sapphire polishing wafers. The official launch of its 12-inch SiC pilot line is a significant achievement in Jingsheng Electromechanical's dual-engine development strategy of "advanced materials and advanced equipment," and a comprehensive demonstration of my country's high-end semiconductor equipment manufacturing capabilities.

Please, Log in or Register to view URLs content!

 

[tokenanalyst]

BEUV(Beyond EUV Lithography)