On the morning of December 4, a number of chip concepts such as automotive chips, semiconductors, MCU chips, and memory chips led the rise. Among the constituent stocks, Dawei shares (002213. SZ, share price of 13.13 yuan, market value of 3.114 billion yuan), Huangting International (000056. SZ, share price 3.66 yuan, market value 4.429 billion yuan), Chengdu Huawei (688709.SH, stock price 27.3 yuan, market value 17.386 billion yuan), Fullhan Micro (300613. SZ, share price 49.53 yuan, market value 11.435 billion yuan), NOVOSENSE (688052. SH, share price of 128.09 yuan, market value of 18.256 billion yuan) rose more than 5%.

On the news side, on the evening of the 3rd, the China Semiconductor Industry Association, the Internet Society of China, the China Association of Automobile Manufacturers and the China Association of Communications Enterprises successively issued statements, saying that the arbitrariness of the United States' control measures against China has affected the stable supply of American chip products, and the trust and confidence of many industries in purchasing chip products from American companies have been shaken, and American chips are no longer reliable and safe.

In the eyes of the outside world, this will strengthen the market's attention to the localization of semiconductors.

Semiconductor export controls to China have been upgraded

On December 2, local time, the Bureau of Industry and Security (BIS) of the U.S. Department of Commerce issued the latest export control measures for semiconductors to China and added 136 Chinese entities to the list of export control entities.

It is worth noting that the depth and breadth of the domestic industrial chain coverage of this list are stricter than those of previous rounds, and the focus has shifted to restricting equipment companies, and restricting HBM (high-bandwidth memory) and software tools.

Guo Zuorong, Senior Vice President of Research at TrendForce, said at a recent seminar on memory industry trends that the relevant measures of the United States will further fragment the global semiconductor market. "In the future, everyone will become independent, everyone will build their own semiconductor factories, and the governments of each region will also attract some companies to build local semiconductor factories, which we think is irreversible, and the subsequent situation may be that the global market no longer exists."

For the materials and some parts companies began to be included in the entity list, many brokerages believe that this will force the semiconductor industry to accelerate localization.

Among them, CITIC Securities believes that the restriction of the market has been expected, because the relevant enterprises have been prepared, has carried out long-term stockpiling and supply chain switching to the United States in advance, the agency believes that the short-term actual impact is limited, and does not constitute a significant impact on the business continuity of the enterprise, in the long run, self-reliance and self-improvement, is expected to further accelerate the localization process of the whole industry chain.

China Merchants Electronics said in the research report that this round of new regulations and regulations are upgraded, and the localization of advanced manufacturing, packaging and other fields is expected to be further accelerated, and it is recommended to pay attention to the localization process in the fields of advanced manufacturing, HBM and advanced packaging, upstream equipment, parts, materials, and domestic computing power.

Localization is accelerating

In the past few years, the United States has repeatedly issued an "entity list" to restrict China's semiconductor development. This time, a new round of semiconductor control in the United States intends to cut off the supply of AI (artificial intelligence) underlying computing power, thereby limiting the development of Chinese artificial intelligence.

According to the White Paper on the Global Digital Economy recently released by the China Academy of Information and Communications Technology, China accounts for more than one-third of the world's AI models, reaching 36%, second only to the United States at 44%. In terms of the number of global AI companies, China ranks second after the United States with a 15% share.

For a long time, large model training has relied on the supply of massive AI computing power. For the construction of domestic computing power, Huawei, Alibaba, Baidu, and Tencent among the giants have self-developed AI chips, and there are also Cambrian (688256. SH, share price 552.6 yuan, market value 230.687 billion yuan), Haiguang Information (688041. SH, share price of 128.89 yuan, market value of 299.584 billion yuan), Loongson Zhongke (688047. SH, share price of 157.12 yuan, market value of 63.005 billion yuan) and other enterprises.

At the same time, to meet the demand for massive computing power in the era of large models, HUAWEI CLOUD's Ascend AI cloud service integrates large-scale computing power clusters, computing engine CANN, and AI development framework MindSpore to provide full-stack computing power for large model training, inference, and AI application development and operation. It is reported that the Ascend AI cloud service has been adapted to more than 100 large models in the industry.

Xu Zhijun, Vice Chairman and Rotating Chairman of Huawei, said at HUAWEI CONNECT 2024: "Computing power depends on semiconductor processes, but we must face the reality that the U.S. sanctions against China in the field of AI chips will not be lifted for a long time, and China's semiconductor manufacturing process is also subject to U.S. sanctions, which means that the advanced nature of the chips we can manufacture will be restricted." This is the challenge we must face to build a computing solution. ”

In addition, in terms of advanced manufacturing processes, Guo Zuorong said that it is necessary to start with domestic equipment. "At present, many companies are well aware of this, and they are moving towards semiconductor equipment, and it is difficult to go further in the process without tools."

In terms of HBM, the key memory unit of AI chips, Guo Zuorong believes that some domestic HBMs should be seen in the next 2 to 3 years. "HBM is a must-do product, without which it is difficult to succeed without AI chips, and at present, domestic memory factories are more active in developing HBM."

According to public information, HBM belongs to the DRAM category. It is vertically stacked by multiple layers of DRAM dies, and each layer of dies is connected to the logic dies through through-silicon via (TSV) technology, so that the 8-layer and 12-layer dies are packaged in a small volume space, so as to have the advantages of high bandwidth, high capacity, low latency and low power consumption. In fact, a number of chips launched by NVIDIA for AI training, such as V100, A100, H100, etc., all use HBM as video memory.

At present, with the explosive growth of generative AI, the market demand for high-performance AI chips has increased dramatically, which has also given rise to the industry's demand for HBM. According to TrendForce, the HBM market is expected to reach US$25 billion by 2025, a six-fold year-on-year increase, driven by the rapid growth of AI applications and the surge in demand for related storage.

At present, there are only three HBM suppliers in the world, namely SK hynix, Samsung Electronics and Micron of the United States. Now that HBM is included in the Export Control Classification Number (ECCN) 3A090.c, the new regulations restrict the export and re-export of HBM products with a memory bandwidth density of 2GB/s or more per square millimeter and above, basically covering HBM2 and more advanced products.

Domestic HBM is in a breakthrough period of 0 to 1. Guo Zuorong said, "From the perspective of factories and equipment, some domestic manufacturers are not inferior to international manufacturers, and now they mainly look at R&D resources, and the hardware conditions are sufficient."

Cover image source: Visual China-VCG211433816736