China Unicom CPU bid for 8B RMB resulted in domestic chipmakers getting 90% of orders and Intel getting remaining 10%
Chinese semiconductor thread II
- Thread starter vincent
- Start date
China Unicom CPU bid for 8B RMB resulted in domestic chipmakers getting 90% of orders and Intel getting remaining 10%
Kirin 9610A/X90 die shot. Let me know what you guys think, from what I can see, you have the 4+4+2 cores here.
The 4 large and 4 medium cores have 1 MB L2 cache each and the 2 small cores have 512kB L2 cache -> 9MB L2 cache in total
There is also 16MB of L3 cache + 8MB of SLC cache.
Maleoon 916 looks to have 6CUs + 1664KB L2 cache
NPU has 4 cluster tensor cores + 2 Ascend Lite + 1MB L2 Cache
To put things in perspective, comparing it to the latest Kirin 9020 design
8 cores, 5MB L2 cache and 10MB of L3 cache and 8MB SLC? Maleoon 920 is 4CUs
9000S die shot showed Ascend Lite tenor + Ascend Tiny. I'd imagine 9020 is something similar.
So basically with a larger die, you can load in larger cores, more L2/L3 cache, more powerful GPU and NPU. pretty much what one would expect
I've always been a bit pessimistic and cautious here but server market is really a different story. For Chinese companies buying servers, ordering domestic chips isn't just de-risking or helping domestic supply chain, but it is the better choice in terms of compute power. Often these servers with hundreds of cores prioritize compute density rather than single-core performance, spawning thousands of conainers and possibly tens of thousands of threads on one machine. Huawei already sells dense cloud instances like kc2 with 160 cores, and there are rumors of new chips like Kunpeng 920F that slashes single-core performance in exchange for even denser compute. If you don't care about max single-core performance, Intel really doesn't look like the better choice here.
Interestingly, AMD was missing in the bidding. AMD server chips offer much better compute density than Kunpeng because of TSMC. Not sure if they are banned or simply not included in the bidding.
Interestingly, AMD was missing in the bidding. AMD server chips offer much better compute density than Kunpeng because of TSMC. Not sure if they are banned or simply not included in the bidding.
GPU looks tiny, but otherwise a solid configuration. Probably because there aren't many games on HarmonyOS laptops anyway, and transistor budget is tight on N+2.
IMHO they would be better off making a larger GPU by also using the ISP and NPU regions.
In theory you can do that sort of compute with a GPGPU as well. Might be less area optimized, but the GPGPU is more multi-purpose.
Having separate units might make more sense in power sensitive smartphones but I do not think it is the right choice for laptops.
In theory you can do that sort of compute with a GPGPU as well. Might be less area optimized, but the GPGPU is more multi-purpose.
Having separate units might make more sense in power sensitive smartphones but I do not think it is the right choice for laptops.
Huawei HiSilicon enters the SiC field and launches two 1200V industrial-grade SiC devices
HiSilicon Technologies Co., Ltd., a subsidiary of Huawei, has entered the field of silicon carbide power devices and launched the first two TO-247 packaged 1200V SiC single tubes, targeting industrial high-temperature and high-voltage scenarios.
HiSilicon 1200V SiC device:
1. HiSilicon ASO1K2H020M1T4
HiSilicon ASO1K2H020M1T4 is a 1200V silicon carbide MOSFET in TO-247-4 package. It has a resistance of 20mΩ at 25°C and can maintain an ultra-low resistance of 30mΩ at a high temperature of 175°C. The gate charge is 185nC and can pass a maximum current of 100A (25°C).
2. HiSilicon ASO1K2H035M1T4
HiSilicon ASO1K2H020M1T4 is a 1200V silicon carbide MOSFET in TO-247-4 package . It has a resistance of 35mΩ at 25°C and can maintain a resistance of 57mΩ at 175°C. It has a gate charge of 108nC and can pass a maximum current of 100A (25°C).HiSilicon's launch of 1200V SiC devices stems from Huawei's long-term strategic investment in the field of silicon carbide and its layout of the entire industry chain. Since the establishment of Hubble Technology Investment Co., Ltd. on April 23, 2019 , Huawei has been committed to accelerating its ecological construction in the field of semiconductors and related cutting-edge technologies through strategic investment.Since August 2019, Hubble Technology Investment has continued to make efforts to comprehensively deploy all links of the silicon carbide industry chain. In the substrate segment, it invested in Shandong Tianyue Advanced (holding 10% of the shares) and Beijing Tianke Heda; in the epitaxial wafer segment, it invested in Dongguan Tianyu Semiconductor; in the device design and manufacturing and power electronics solutions segment, it invested in Dongwei Semiconductor, Hanxin Technology, Yingfeiyuan, Jiehuate Microelectronics, Yichong Semiconductor, etc.The 1200V SiC single tube launched by HiSilicon has enriched its product line in the field of power devices with its performance in high temperature and high voltage scenarios, and has provided efficient and reliable solutions for industries such as industrial and automotive electronics. This move demonstrates HiSilicon's technical strength and market layout in the field of wide bandgap semiconductors, and is expected to promote the popularization of silicon carbide power devices in more high-end applications, accelerate industrial upgrading and technological innovation, and is of great significance to improving my country's self-sufficiency and competitiveness in key semiconductor fields.
Zhaoxin's new server chip KH-50000 released: 3GHz, 96 cores!
On July 27, during the 2025 World Artificial Intelligence Conference, Zhaoxin, a domestic X86 processor manufacturer, announced a new generation of server CPU, the KH-50000, which is a significant upgrade over the previous generation KH-40000. At the same time, Zhaoxin also released the KX-7000N CPU for PCs.
According to the leaked information, KH-50000 is based on the Chiplet architecture and adopts a new microkernel design. Its IPC performance is also further improved compared to the previous generation. The top-level SKU has up to 96 cores, a base frequency of 2.2GHz, an acceleration frequency of 3.0GHz, and a maximum L3 cache capacity of 384MB (KH-40000 is 64MB).
In terms of I/O, KH-50000 has also been significantly upgraded: supports 12-channel DDR5 (KH-40000 only supports 8-channel DDR4), supports ECC; supports 128-channel PCIe 5.0/ZPI 5.0/CXL; supports 16-channel PCIe 4.0/SATA/USB.
In terms of expansion interconnection, KH-50000 supports dual- and quad-socket systems, which means that four CPUs can be integrated on a single motherboard to provide up to 384 cores. The new ZPI 5.0 interconnection also further enhances data bandwidth while providing lower latency and power consumption.
Zhaoxin said that in response to the rapidly growing application needs of AI training and inference, Zhaoxin focuses on the new generation of server processors KH-50000, taking advantage of the product's higher computing density, fully upgraded high-speed IO, new autonomous interconnection, and stronger multi-channel expansion capabilities. It provides comprehensive, efficient and reliable support in parallel computing, data processing, heterogeneous hardware expansion, AI algorithm support and other aspects for AI computing infrastructure products of various forms and specifications, including AI workstations, general AI servers, high-density AI servers, etc.
Previous data showed that the KX-7000 series processors "Century Avenue" independent core micro-architecture and advanced Chiplet interconnection architecture have 8-core CPUs with a maximum main frequency of 3.7GHz, integrated 4MB L2 cache, 32MB L3 cache, and computing performance is doubled compared to the previous generation. At the same time, the IO interface has been upgraded to the international mainstream DDR5, PCIe 4.0, and USB4. In addition, the KX-7000 series processors integrate high-performance graphics cards, and the graphics performance is 4 times higher than the previous generation. It supports H.265 hardware encoding and decoding and supports up to dual-channel 4K display.
The new KX-7000N is a heterogeneous integration of high-performance NPU based on KX-7000. It is based on high-performance architecture and has powerful computing power, which can significantly improve the end-side AI processing performance and user experience. At this exhibition, Zhaoxin AIPC product solutions based on KX-7000N processors were also unveiled simultaneously.
Zhaoxin said that the AIPC solution based on the KX-7000N processor can provide users with efficient large-model inference capabilities on the end side, and provide excellent technical performance for AIPC, meeting the needs of scenarios and applications such as personal AI assistants, smart office for government and enterprises, smart education, and content creation. The solution supports mainstream AI frameworks and model deployment, and realizes local offline real-time operation of AI applications such as speech recognition, image generation, and large language models, while significantly reducing system latency and energy consumption. It has the advantages and characteristics of autonomous security, high performance, low cost, and flexible development, and is a key engine for promoting AI localization, privacy protection, and intelligent experience.
In addition, Zhaoxin also revealed that the next-generation Kaixian series AIPC processors will provide more CPU cores and stronger computing performance, upgraded to support PCIe 5.0, heterogeneously integrated high-performance NPU, support flexible expansion of computing power, and provide a variety of options for users' differentiated AI reasoning application needs.
Extreme ultraviolet radiation characteristics of dual-wavelength laser with adjustable time delay interacting with Sn target
Abstract: Laser-produced extreme plasma ultraviolet (LPP-EUV) source is one of the key technologies in advanced lithography systems. Recently, solid-state lasers have been proposed as an alternative drive laser for the next-generation LPP-EUV source. Compared with currently used CO 2 lasers, solid-state lasers have higher electrical-optical efficiency, more compact size, and better pulse shape tunability. Although limited to shorter operating wavelengths, the solid-state lasers have critical higher plasma density and optical depth. Consequently, re-absorption and spectral broadening cause lower conversion efficiency (CE). Therefore, to optimize EUV emission features and improve CE, a 0.532-μm pre-pulse laser is utilized in this work to modulate the plasma density distribution. The pre-pulse and a 1.064-μm Nd: YAG laser (the main pulse) are incident on an Sn target slab co-axially. The EUV energy and spectra of the Sn plasma are characterized at various delay times. It is demonstrated that compared with the 1.064-μm single pulse, the 0.532-μm pre-pulse laser with short delay times of 10 ns and 20 ns respectively results in a 4% increase in CE at 26° and 18% increase at 39°. The angular distribution of EUV energy is modulated by the 0.532-μm pre-pulse. An isotropic emission can be obtained within a certain delay time. The spectral feature near 13.5 nm is optimized, and a spectral purity of 12.2% is improved by 69%. The laser spot sizes of 0.3 mm and 1 mm for the pre-pulse are compared in the experiment. The results show that the 1-mm spot size has a better modulation effect on the EUV emission. Moreover, the time-resolved visible-band plasma profile is captured by an ICCD with 1.6-ns gate width. The plasma size and the distance to the target surface are increased by the 0.532-μm pre-pulse, which suggests that the energy of the main pulse is deposited in the low-density pre-plasma plume instead of in the plasma near the target surface. The lower plasma density leads to an increase in CE and spectral purity. The angular distribution of EUV energy is found to be closely related to the plasma morphology, and defined as the ratio of the longitudinal size to lateral size of the plasma. This indicates that the variation of plasma morphology can influence the Angular distribution of EUV energy, which is caused by the 0.532-μm pre-pulse. This work has guiding significance for optimizing the emission characteristics of solid-state laser driven EUV sources.
Silan Microelectronics' automotive semiconductor packaging phase II project foundation laying ceremony
Recently, the groundbreaking ceremony of the second phase of Silan Microelectronics' automotive semiconductor packaging project was held in Cheng'a Industrial Concentration Development Zone, Huaikou Town, Jintang County, marking the official entry of the project into the construction implementation stage.
The project is invested and constructed by Hangzhou Silan Microelectronics Co., Ltd. with a total investment of 1.5 billion yuan. The project will build a new plant of 79,000 square meters and supporting facilities and equipment, and expand the automotive-grade power module and power device packaging production line to create a modern industrial cluster covering 6 major units including production plants, power stations, and stereoscopic warehouses, with a construction area of 82,000 square meters.
The project is expected to complete equipment installation and commissioning and trial production in December 2026, and reach full production by the end of 2027. After reaching full production, it is expected to achieve annual sales revenue of more than 800 million yuan, annual tax payment of more than 40 million yuan, and provide more than 500 jobs.
The Chengdu base is the only packaging core among Silan Microelectronics' three major production bases, and its IPM modules have occupied more than 30% of the global market share. The implementation of the second phase project will further enhance Chengdu Silan's production capacity and technical level in the field of automotive semiconductor packaging, break the bottleneck of production capacity, build a guarantee system of "continuous production and safe delivery", strengthen Silan Microelectronics' competitiveness in the field of automotive semiconductors, and will also play a positive role in promoting the high-quality development of the industrial economy in Chengdu and even Sichuan Province.