Feb. 27

01Market Trend

Why Chiplet is favored by all?

In recent years, Chiplet (small chip) has emerged rapidly in the global semiconductor industry, attracting the attention of many tech giants and industry experts. A Chiplet combines multiple small chips (or chip cores) into one in a heterogeneous way and breaks the design bottleneck of traditional single chips, showing tremendous technological advantages and market potential.

As Moore's Law approaches physical limits, traditional single-chip design faces many challenges in terms of high-performance computing, such as rising costs, extended design cycles, and performance bottlenecks. With Chiplet, complex functions can be broken down into small chips through modular design, reducing design complexity and cost while improving performance. In the fields of AI,data center, and high performance computing (HPC), there are increasing demands on chip performance and energy efficiency. A Chiplet can provide flexibility to meet diverse computing needs by integrating different functional modules such as CPU, GPU, AI accelerators. For example, companies such as AMD and Nvidia have adopted the Chiplet technology extensively in their high performance computing platforms. Chiplet allows integration of small chips of different process nodes, which can reduce costs while maintain performance. For example, in terms of data center applications, cost and power consumption can be optimized by integrating high computing capacity cores with high-capacity storage cores.

According to market research institutions, Chiplet's market is expected to grow rapidly in the coming years. The global Chiplet market is expected to reach US$411bn by 2035. In the future, the Chiplet technology will be used more widely in the fields of data center, automotive electronics, consumer electronics, IoT and more.

Comments: The rise of Chiplet is the semiconductor industry's inevitable choice to cope with Moore's Law limits and high performance computing needs. It not only addresses the bottleneck in traditional chip designs but also shows greater flexibility and cost-effectiveness through modular design and heterogeneous integration. From the perspective of industry development, the success of Chiplet depends on standardization and ecosystem building. By driving the implementation of industry standards such as UCIe and CSA, cores from different fabs are interoperable, thus accelerating the population of Chiplet. For the development of domestic semiconductor industry, Chiplet offers a new development opportunity. With the measures like strengthening packaging technology and reuse of IP, the domestic semiconductor industry is expected to seize a more important position in the global semiconductor industry. However, there are still some challenges to be coped with to achieve the goal of the comprehensive population of Chiplet, such as the reliability of interchip connectivity, optimization of manufacturing processes, and so forth.

Feb. 28

02Policy Trend

US tech giant calls to modify chip export restrictions

Recently, US tech giants have called on the Trump administration to reconsider and modify the chip export ban imposed during the Biden administration period. Companies such as Microsoft, Amazon and Nvidia have expressed concerns about the existing export restriction policies, arguing that these restrictions not only adversely affect US allies but also could undermine US competitiveness in the global AI space.

The President of Microsoft pointed out that current export control rules limit the expansion of US technology companies in strategically important markets such as Israel, Saudi Arabia and the UAE. Amazon CEO also said that these restrictions could cause US allies to turn to other countries to acquire chips, which may deprive the US of important markets and partnerships. Nvidia CEO stressed that the long-term implementation of chip export controls against China may force American industries to lose the competitive opportunities in one of the world's largest markets. In addition, the US Information Technology Industry Council (ITI) has expressed concern about the potentially far-reaching impact of export restrictions on the global technology industry chain, and has called on the US government to consider industry opinions fully when formulating policies.

Comments: The call of the US tech giants reflects the complexity and potential risks of the current chip export restriction policy. On the one hand, these restrictions are designed to maintain the technological advantages and national security of the USA; on the other hand, they have a negative impact on the global market share and competitiveness off American companies. The conflict between the export policies and the interests of companies shows the limitations of the "small yard and high fence" strategy of the USA in the field of science and technology.

From the perspective of global industry chain, the export restrictions of the USA not only affect its allies' AI infrastructure but may also drive other countries to accelerate the development of alternative technologies, thus weakening the US dominance in the global semiconductor market. Moreover, the uncertainty of such policies of the USA also threatens the steady development of the global science and technology industry and undermines the international economic and trade order. In the long run, the US export restriction policy may accelerate the restructuring of the global semiconductor industry chain and promote other countries and regions to strengthen the R&D and application of their independently developed technologies.

Feb. 28

03Company Trend

Nvidia stock price plunged

Nvidia stock price experienced a historic plunge, down 17% a day, and its market value decreased by around US$589bn. This decline is not only the largest single-day decline since the early days of COVID19 pandemic in March 2020 but also the highest single-day market value loss in global history. The direct trigger for the stock price plunge of Nvidia was the large language model R1 released by Chinese AI start-up DeepSeek. The model features similar performances as the products of tech giants like OpenAI and Google of the USA, but it cost only US$5.6mn.

The development cost of DeepSeek's model is far below the market expectations, which raised concerns about AI hardware needs. Although the DeepSeek model uses Nvidia's H800 chips, it claims that it took just 2,048 chips and 2.7 million hours of training to complete the development. The news made the market to question the need to continue to invest heavily in expensive high-end chips to develop AI models.

In addition, the model of DeepSeek is open sourced and allows commercial use, which further increased market uncertainty. Investors are concerned that if other AI companies follow DeepSeek and reduce their reliance on Nvidia's high-end chips, it will negatively impact Nvidia's future revenue.

Comments: The plunge of Nvidia’s stock price reflects the market high sensitivity to the changed AI competition landscape. The rise of DeepSeek not only shows the rapid progress of AI technology but also shows that Nvidia’s dominance in the market may be challenged. Despite Nvidia's dominance in the AI chip market, DeepSeek's success indicates that the development of AI in the future may no longer rely entirely on high-end hardware. In the long run, the stock price plunge of Nvidia could be a turning point to drive Nvidia and other AI-related companies to re-examine their technology paths and market strategies. On the one hand, Nvidia needs to continue to innovate to maintain its advantages in HPC market, and on the other hand, companies need to pay attention to changing trends of cost efficiency and market demands to adapt to more competitive markets.

Feb. 26

04Company Trend

Arm launches the smallest CPU

Arm officially launches the Cortex-A320, the smallest CPU based on Armv9 architecture, which is specially designed for edge AI and Internet of Things (IoT) applications. The birth of Cortex-A320 represents the further advancement of Arm in the field of ultra-efficient processors and provides more robust support for edge computing and AI applications.

Cortex-A320 is based on the Armv9.2-A architecture and is a CPU featuring single-transmit, sequentially executed 32-bit instruct Ion fetch, with an optimized eight-stage pipeline. Its energy efficiency is 50% more energy efficient than that of its predecessor, Cortex-A520. In addition, Cortex-A320 offers a 30% improvement in scalar performance compared with Cortex-A35 and a 10-fold improvement in machine learning (ML) performance. Cortex-A320 supports 64KB L1 cache and even 512KB L2 cache and is also equipped with a 256-bit AMBA5 AXI interface. It can also be configured as a single-core or quad-core cluster, supporting flexible memory management.

Cortex-A320 integrates NEON and SVE2 SIMD, supports new data types such as BF16, and is capable of handling large AI models with more than 1 billion parameters. In addition, it features enhanced security functions such as memory tagging extension (MTE), pointer authentication (PAC), and branch target identification(BTI). Cortex-A320 can be used together with Arm's Ethos-U85 neural processing unit (NPU) to form a complete edge AI platform. The platform supports OS such as Linux, Android, and real-time OS such as FreeRTOS and Zephyr OS.

Comments: The released Cortex-A320 is an important strategic layout for Arm in the field of edge AI and IoT. This CPU not only achieves significant improvements in energy efficiency and performance but also provides stronger processing power and security for edge devices by integrating advanced AI and security functions.

From an industry perspective, Cortex-A320 will further drive the application of AI technology in edge devices. Its combination with Ethos-U85 NPU provides developers with more flexible hardware options and can support more complex AI models and application scenarios. In addition, the low power consumption design of Cortex-A320 makes it an ideal upgrade option for traditional Cortex-M series of microcontrollers. For Arm, Cortex-A320 not only solidifies its market position in IoT and edge computing but also provides a solid foundation for future AI applications. With ever increasing software complexity and AI demands, Cortex-A320 is expected to be a key force to drive the development of edge AI.

Feb. 25

05DomesticNews

Samsung and YMTC cooperate to develop next-gen NAND by using Chinese company’s patent

Samsung has confirmed that it will use the patented technology of Yangtze Memory Technologies (YMTC) from V10 (the 10th generation NAND), especially in “hybrid bonding” of the new advanced packaging technology, according to South Korean media. Both parties have signed a license agreement on the patent of 3D NAND hybrid bonding and have reached cooperation.

It is learned that V10 is the next-gen NAND that Samsung Electronics plans to start production in H2 this year as early as possible and the product is expected to have about 420 to 430 layers. Several new technologies, especially W2W (Wafer-to-Wafer) hybrid bonding technology, will be adopted for the product.

It is learned that YMTC is the first company to apply hybrid bonding to 3D NAND, and named this technology “Xtacking”. With this technology, the peripheral circuits responsible for data I/O and memory unit operation can be independently processed on a wafer. This processing method facilitates the selection of the right advanced logic processes to enable NAND to achieve higher I/O interface speeds and more operational features.

In the early days of its business, YMTC obtained the original patent for hybrid bonding by signing a licensing agreement with the US tech company Xperi. Then, YMTC also established a large number of independent patents in the field of NAND packaging technology and has gained a strong patent reserve in terms of related technologies.

Feb. 27

06DomesticNews

Chongqing Sanan Optoelectronics & STMicroelectronics’8-inch SiC project powered on officially

According to the official WeChat account of Sanan Optoelectronics, the SiC wafer plant in Chongqing, jointly ventured by Sanan Optoelectronics and STMicroelectronics, have been officially powered on.

It is reported that the project is expected to achieve mass production in Q4 2025, which will become the first domestic automotive-grade 8-inch SiC power chip mass production line, with a planned capacity of about 10,000 units of wafers per week.

It is learned that Chongqing Sanan Optoelectronics and STMicroelectronics was established in August 2023 , jointly ventured by Sanan Optoelectronics (holding 51% of the equity) and STMicroelectronics (China) Investment Co., Ltd. (holding 49% of the equity). The total investment of the project is about RMB23bn and will produce about 480,000 units of 8-inch SiC wafers per year, and these wafers are mainly used to produce automotive-grade electronic control chips. To support this project, Sanan Optoelectronics established its wholly-owned Chongqing Sanan Semiconductor Co., Ltd. in July 2023, and invested RMB7bn in the same park to build a production line with a capacity of 480,000 units of 8-inch SiC substrates, and the production line was powered on and put into production in September 2024.

Feb. 24

07DomesticNews

AMEC’s R&D and production base & southwest headquarters settled in Chengdu, with total investment of RMB3.05bn

According to the news of the official WeChat account of Chengdu High-tech Zone, Advanced Micro-Fabrication Equipment Inc. China (AMEC) and Chengdu High-tech Zone signed an investment cooperation agreement, by which AMEC will set up a wholly-owned subsidiary in Chengdu High-tech Zone and build its R&D and production base and southwest headquarters project.

It is reported that the total investment of the project is about RMB3.05, and the registered capital of the project is RMB100mn. In addition, it plans to purchase more than 50 mu of land (about 33,333.33 sq meters) for the construction of its R&D center, production base, office buildings and ancillary facilities to meet the demand for mass production. The project is scheduled to start construction in 2025 and will be put into production in 2027.

According to the agreement with Chengdu High-Tech Zone, AMEC will set up a wholly-owned subsidiary -AMEC Semiconductor (Chengdu) Co., Ltd., which is committed to developing high-end logistics and storage chips and carrying out R&D and production of related equipment, including chemical vapor deposition equipment, atomic layer deposition equipment and other key equipment. It will also actively promote the companies in the upstream and downstream supply chains of AMEC to settle in Chengdu High-tech Zone and drive the formation of a high-end semiconductor equipment industry chain cluster to provide a strong support for the upgrading of the semiconductor industry chain in Chengyu and Chongqing.