Chinese semiconductor thread II

[tokenanalyst]

Breaking through 12 inches! Keyou Semiconductor successfully produces 12-inch silicon carbide crystals​

In early September, Keyou Semiconductor successfully produced a 12-inch silicon carbide ingot based on its independently developed 12-inch silicon carbide crystal growth furnace and thermal field technology. This achievement marks Keyou as one of the few semiconductor companies in China to master the complete set of core technologies for 12-inch silicon carbide crystal growth equipment and processes, achieving a comprehensive independent breakthrough from equipment to materials.

12-inch wafers first appeared
This breakthrough not only reflects Keyou's continuous innovation capabilities in silicon carbide crystal growth technology and thermal field design, but also the company's self-developed equipment and independent process are important proof of its solid technical strength in the field of wide bandgap semiconductor materials.
As a third-generation semiconductor material, silicon carbide is a key foundational material for emerging national strategic industries such as new energy vehicles, 5G communications, and rail transit. In recent years, with the rapid development of emerging fields such as artificial intelligence and wearable devices (such as AR glasses), silicon carbide's performance advantages, such as high frequency, high power, and high temperature resistance, have become even more prominent.
The area of a 12-inch substrate is about 2.25 times that of an 8-inch substrate, which means that the number of chips for AR glasses that can be produced by a single 12-inch substrate is much higher than that of an 8-inch substrate. In large-scale production, this can significantly reduce the unit cost of crystal growth, processing, polishing and other links, and is a key path to reducing the cost of core components of AR glasses.

Please, Log in or Register to view URLs content!

 

[tokenanalyst]

Five semiconductor industrialization projects with an investment of nearly 2 billion yuan were signed and settled in Shuangliu, Chengdu​

10 companies including Chengdu University of Information Technology, Hongdian Technology, and China Radio and Television Metrology reached technical cooperation on the spot. Shuangliu District also signed industrialization projects with 5 companies including Chengdu Ultrapure Applied Materials Co., Ltd. and Nanjing Jinlianyi Information Technology Co., Ltd., with an agreed investment of nearly 2 billion yuan.

On the afternoon of September 5, at a semiconductor industry salon event called "Core Gathering in Shuangliu, Intelligent Creation and Progress Together" held in Shuangliu District, 10 companies including Chengdu University of Information Technology, Hongdian Technology, and China Radio and Television Metrology reached technical cooperation on the spot. Shuangliu District also signed industrialization project agreements with 5 companies including Chengdu Ultrapure Applied Materials Co., Ltd. and Nanjing Jinlianyi Information Technology Co., Ltd., with an agreed investment of nearly 2 billion yuan.

It is understood that this event focuses on the integration of semiconductor industry resources and ecological co-construction, aiming to promote win-win cooperation among enterprises, guide enterprises in the chain to take root and develop in Shuangliu, solidly promote the Liyuan Manyuan Action, and assist in the construction of "Smart Manufacturing Airport and Happy Park City".

"We mainly value Shuangliu's cluster advantages in the semiconductor industry. At the same time, it is also a gathering place for universities and research institutes, with obvious advantages in talent. In addition, the business environment is also very good. These factors are very important for the development of enterprises." Chai Jie, chairman of Chengdu Ultrapure Applied Materials Co., Ltd., one of the companies participating in the signing of the industrialization project, said that the company's contract amount this time is 500 million yuan. With the expansion of the company's investment in factory buildings in Shuangliu and the increase in production capacity, the company's output value is expected to exceed 700 million yuan in the next five years.

Please, Log in or Register to view URLs content!

 

[PopularScience]

The first domestic EUV photoresist with MOR core technology in China was launched

Please, Log in or Register to view URLs content!

 

[tokenanalyst]

Jingfei Semiconductor: A new breakthrough in 12-inch silicon carbide wafer laser lift-off technology!​

Beijing Jingfei Semiconductor Technology Co., Ltd. recently achieved a major breakthrough in silicon carbide wafer processing technology, successfully debonding 12-inch silicon carbide wafers using its independently developed laser debonding equipment. This breakthrough marks a significant step forward for China in key third-generation semiconductor manufacturing equipment and provides a new solution for reducing costs and increasing efficiency in the global silicon carbide industry. This technology has previously been validated by multiple customers in the 6- and 8-inch silicon carbide field, with equipment performance reaching internationally advanced levels.


This technological breakthrough has multiple implications for the development of the silicon carbide industry, including:

1. Significantly reduce production costs: 12-inch silicon carbide wafers have a usable area that is approximately four times larger than the current mainstream 6-inch wafers, and the unit chip cost is reduced by 30%-40%.
2. Improve industry supply capacity: Solve the technical bottleneck of large-size silicon carbide wafer processing and provide equipment guarantee for global silicon carbide production capacity expansion
3. Accelerate the process of domestic substitution: breaking the technological monopoly of foreign manufacturers in the field of large-size silicon carbide processing equipment, providing important support for the independent control of my country's semiconductor equipment
4. Promote the popularization of downstream applications: Cost reduction will accelerate the application of silicon carbide devices in new energy vehicles, renewable energy and other fields

Beijing Jingfei Semiconductor Technology Co., Ltd., a technology transfer enterprise of the Institute of Semiconductors, Chinese Academy of Sciences, specializes in the research, development, production, and sales of specialized semiconductor equipment. Focusing on laser application technology, the company has developed a series of semiconductor processing equipment with independent intellectual property rights, serving major domestic semiconductor manufacturers.

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

 

[tokenanalyst]

Sugon releases China's first open-architecture AI super cluster system capable of supporting millions of cards​

On September 5th, at the 2025 Chongqing World Intelligent Industry Expo, Sugon released the Sugon AI Super Cluster System, China's first product based on an open architecture for AI computing. With GPUs at its core, the system implements an integrated, tightly coupled design encompassing computing, storage, networking, power, cooling, management, and software. Featuring ultra-high performance, ultra-high efficiency, ultra-high reliability, and comprehensive openness, it provides an efficient computing power foundation for training and inference of large models with trillions of parameters, fine-tuning large industry models, developing large multimodal models, and AI4S scenarios.

Compared with closed systems, the Sugon AI super cluster system not only works as efficiently as a single computer through its tightly coupled design, but also supports multi-brand AI acceleration cards and is compatible with mainstream software ecosystems such as CUDA, providing users with more open choices and significantly reducing hardware costs and software development and adaptation costs, thus protecting initial investments.


As a model of open, standardized, and high-efficiency computing clusters, the Sugon AI super cluster system has the following technical features:
Ultra-high performance: A single cabinet can carry 96 GPU cards, with a computing power of hundreds of petabytes and a total memory access bandwidth exceeding 180TB/s; it supports multi-precision and mixed-precision computing; and it supports the expansion of ultra-large clusters with millions of cards.
Ultra-high efficiency: The Qianka cluster's large-model training and inference performance reaches 2.3 times the industry mainstream level, development efficiency is increased by 4 times, and man-day investment is reduced by 70%; storage, computing, and transmission are coordinated to improve GPU computing efficiency by 55%; advanced cold plate liquid cooling, 394 energy-saving design PUE is less than 1.12.
Ultra-high reliability: 121 equipment and link RAS reliability designs; mean time between failures (MTBF) increased by 2.1 times; mean time to repair (MTTR) reduced by 47%; cluster reliability testing for over 30 days of stable operation; and automatic analysis and second-level isolation of millions of component failures.
Fully open: Designed based on an open AI computing architecture; hardware compatible with AI accelerator cards from multiple brands; software compatible with the mainstream AI computing ecosystem; and multiple technical capabilities open and shared.

Please, Log in or Register to view URLs content!

 

[tokenanalyst]

Changcun Phase III Integrated Circuit Company was established with registered capital of RMB 20.72 billion.​

According to public information from Tianyancha, on September 5, Changcun III (Wuhan) Integrated Circuit Co., Ltd. was established with a registered capital of 20.72 billion yuan, and Yangtze Memory Chairman served as the legal representative.

Its shareholders are Yangtze Memory Technologies (50.19% stake, subscribed to 10.4 billion yuan) and Hubei Changsheng Phase III Investment Development Co., Ltd. (49.81% stake, subscribed to 10.32 billion yuan). Tianyancha shows that Hubei Changsheng Phase III Investment Development Co., Ltd. was established on September 2, 2025, with a registered capital of 15.12 billion yuan. Its shareholders are Wuhan Optics Valley Financial Holding Group Co., Ltd. (40% stake, subscribed to 6.048 billion yuan), Jiangcheng Industrial Investment Fund (Wuhan) Co., Ltd. (40% stake, subscribed to 6.048 billion yuan), and Yangtze River Industrial Investment Group Co., Ltd. (20% stake, subscribed to 3.024 billion yuan).

The business scope includes: integrated circuit manufacturing; integrated circuit sales; integrated circuit design; integrated circuit chip and product sales; integrated circuit chip and product manufacturing; technical services, technical development, technical consulting, technical exchanges, technology transfer, and technology promotion; import and export of goods; import and export of technology; and import and export agency services.

Please, Log in or Register to view URLs content!

 

[jx191]

I read through this stuff.

There is some good stuff there, but I think a lot of his AI related stuff are just horribly misguided.

I think everyone has now arrived at the conclusion that China's 7nm capacity is expanding quite a bit, because it's there for all to see. You can't really deny it anymore.

the problem with this is that Morgan Stanley doesn't seem to understand yield for die depends on die size. So, TSMC's 90% yield for 5nm process isn't going to be 90% for a 800 mm^2 AI die (I don't recall the size of 910B/C die, so just putting 800 here).

30-40% yield on AI chip is not end of the world. you get 30 die per wafer instead of 60 die. With 10k wpm, you still get 150k 910C per month or 1.8m a year. Again, China is swimming in AI chips. If you cannot arrive at this conclusion, then you are pretty stupid.

The more I read into things like this the more I realize that there are very few circles of people who actually have the knowledge to know what's going on with china's semiconductor industry. Even more so for western outlets and even big organizations like morgan stanley just have no clue what they are talking about.

 

[tokenanalyst]

Core Wisdom's Hybrid Bonding Equipment Showcased at CSEAC​

Core Wisdom (Jiangsu) Technology Co., Ltd. exhibited advanced bonding equipment (W2W/D2W Hybrid Bonder) and related measurement and debonding equipment. The company will continue to be committed to the research and development of semiconductor bonding technology, provide customers with competitive solutions, and promote the coordinated development of the industry.

W2W hybrid bonding equipment

The key performance indicators of the CANOPUS series hybrid bonding equipment are among the best in the industry, supporting the integrated integration of multiple modules such as pre-alignment, real-time alignment, activation, cleaning, detection, and debonding.

D2W hybrid bonding equipment

The key performance indicators of the SIRIUS series hybrid bonding equipment are among the best in the industry , supporting the integrated integration of multiple modules such as pre-alignment, real-time alignment, activation, cleaning, debonding, and detection.​

Please, Log in or Register to view URLs content!

 

[tokenanalyst]

Quintiles' fully magnetic levitation molecular pumps make their debut at Wuxi CSEAC.​

Quintiles showcased nine semiconductor-grade, fully magnetic molecular pumps with pumping speeds ranging from 90L/s to 4500L/s. This series of products, with their cutting-edge technology and unique design, attracted significant attention and in-depth consultation and discussion from exhibitors and visitors . These pumps not only feature five-degree-of-freedom, fully magnetic levitation control, extensive interface compatibility, standardized dimensions, and a compact, integrated design, but also achieve excellent ultimate vacuum performance. They can be flexibly paired with either integrated or split controllers, support any-angle mounting, enable multi-functional remote communication, and operate at ultra-low vibration.

​

 

[tokenanalyst]

​

The Institute of Microelectronics has made progress in the field of POSIT floating-point SRAM in-memory computing macro chips​

POSIT is an emerging floating-point number format that more efficiently allocates precision across different numerical ranges. Data close to zero achieves higher computational accuracy, while extremely large or small values can be reduced in precision in exchange for a wider data representation range. This dynamic precision allocation is well-suited for AI algorithms, enabling better algorithmic performance at the same data bit width (e.g., POSIT8 compared to INT8). Previously, researchers designed efficient computing hardware tailored to the specific POSIT format, but these designs failed to fully exploit the variable length of POSIT's mantissa. A strategy that dynamically reduces the cost of mantissa multiplication calculations, tailored to the variable length of POSIT's mantissa, could provide an approximate calculation method with minimal loss of algorithmic accuracy to reduce the power consumption of mantissa multiplication, which dominates POSIT multiplication calculations.

Recently, a team led by Researcher Zhang Feng from the National Key Laboratory of Integrated Circuit Manufacturing Technology, Institute of Microelectronics, Chinese Academy of Sciences, proposed an in-SRAM computation macro based on dynamic prediction of the mantissa length of POSIT results. This approach improves both throughput and energy efficiency with minimal loss of algorithmic accuracy. Taking advantage of the variable-length mantissa of POSIT, a new digital CIM macro was proposed that determines which parts of the mantissa can be discarded in the multiplication calculation before executing the POSIT floating-point MAC calculation. Using unsigned Radix8 Booth multiplication logic and final cycle fusion (FCF) technology, the CIM macro's computation cycle count was further reduced, improving energy efficiency by 2.19-2.79 times. To address the CIM structure's inability to support flexible cycle skipping, a CIM macro-level approximate computation control (MACC) unit was designed to dynamically adjust the computation cycle count of individual CIM macros, achieving a 1.89x speedup while maintaining a mean absolute percentage error (MAPE) loss of only 3.24% for the approximate computation. For the sparse bits in mantissa multiplication, an activation mantissa alignment skip (AMAS) unit is proposed to help reduce the number of input cycles of sparse activation data, thereby achieving a 1.41x performance improvement.

​

Please, Log in or Register to view URLs content!