According to the data of the Prospective Industry Research Institute, the market size of China's semiconductor industry is expected to grow at a growth rate of about 5% from 2024 to 2029, and the market size is expected to reach 256.4 billion US dollars by 2029. According to the analysis of Winsoul Capital, the inventory of the semiconductor industry will return to health in 2024, and the prosperity will rebound. Investment will tighten in 2023, and the number and amount of financing in Q1 2024 will decrease year-on-year. This market survey will analyze the market situation in the semiconductor field according to the "2024 China Semiconductor Investment In-depth Analysis and Outlook" released by Winsoul Capital.
AI models drive the rapid development of hardware infrastructure and advanced packaging
In 2023, the explosion of large language models represented by ChatGPT has made the saying "the first year of AI large models" widely circulated. The pre-training and generation capabilities of large models and the application capabilities of multi-modal scenarios have been greatly improved. Also in 2023, China's local manufacturers, technology giants, research institutes, and startups have poured into the field of AI large models, competing to deploy their own large language models. According to the statistics of Winsoul Capital, as of the end of November 2023, the number of large models released in the Chinese market has exceeded 300, of which about 40 are general-purpose large models, covering NLP, CV, modal large models and multiple customized large models. There are more than 250 manufacturers and universities in China with large models above the scale of 1 billion parameters.
With the development of large models and generative AI, the demand for computing power has increased dramatically. In the past four years, the number of large model parameters has grown at an average annual compound rate of 400%, and the demand for AI computing power has increased by more than 150,000 times, far exceeding Moore's Law. This means that the traditional CPU-centric computing infrastructure can no longer meet the demand, and requires larger-scale, higher-performance, and more stable computing resources. In addition, in order to further improve the efficiency of computing power, technologies such as distributed computing and heterogeneous computing are also widely used in the training and inference process of large models. Large models need to process massive amounts of data, including training data, model parameters, and output results. As a result, the requirements for data storage and processing power are very high. In order to meet the data requirements of large models, efficient data storage and processing technologies, such as distributed storage, data compression, and data preprocessing, are required. At the same time, it is also necessary to establish a sound data management system to ensure the quality, security and availability of data.
Figure: AI large models drive the rapid development of hardware infrastructure and advanced packaging
As AI puts forward higher requirements for basic hardware facilities, advanced packaging is also driving rapid development. AI chips require higher computing power, lower power consumption, and smaller size, which are driving innovation in advanced packaging technologies. According to the analysis of Winsoul Capital, advanced packaging technology can break through the blockade of the advanced manufacturing process of computing chips such as GPUs, and in order to meet the needs of AI chips for high performance and low power consumption, advanced packaging technology continues to develop in the direction of miniaturization and high integration. For example, fan-out packaging technologies (e.g., FOWLP, FOPLP) and 2.5D/3D packaging technologies improve chip performance and integration by improving packaging methods. In addition, with the continuous expansion of AI applications, advanced packaging technologies are also showing a trend of diversification. In addition to traditional packaging technologies, new packaging technologies such as through-silicon vias (TSV), micro-bumps, and RDLs have emerged, which provide more options and possibilities for AI chips. AI chips also generate a lot of heat during operation, so advanced packaging technology needs to have good heat dissipation performance. At the same time, in order to ensure the stable operation of AI chips, advanced packaging technology also needs to have high reliability.
Figure: Rapid development of advanced packaging
On the other hand, the scope of semiconductor regulation continues to expand, targeting semiconductor manufacturing equipment used in more advanced nodes. Since the U.S. government has escalated and increased its semiconductor export controls to China on the grounds of "national security" in an attempt to prevent China from acquiring advanced semiconductor technology and equipment, other Western countries have also followed the United States and strengthened export controls on semiconductor manufacturing equipment. Under external pressure, China has accelerated the localization of semiconductor equipment and promoted the development of the local semiconductor industry through policy support and capital investment. Domestic manufacturers are also rapidly promoting product upgrades, aiming at more advanced process nodes to develop supporting equipment and materials. Although the semiconductor control restricts China's access to advanced semiconductor technology and equipment, and brings certain challenges to the development of China's semiconductor industry, it also prompts China to accelerate the process of semiconductor equipment localization and improve its independent and controllable capabilities.
