The demand for computing power is urgent! Beijing's plan directly points to the pain point of AI, which is called "Chiplet" to make up for the technology gap

Science and Technology Innovation Board Daily, May 20th (Editor Zheng Yuanyuan)According to the official website of the Beijing Municipal Bureau of Economy and Information Technology on the 19th,Beijing plans to organize and implement the "Beijing General Artificial Intelligence Industry Innovation Partnership Plan"20251010The Plan states that,The current problems include insufficient intelligent computing power, insufficient supply of high-quality data, etcIn response, the Plan proposes eight key tasks:Accelerate meeting the urgent demand for computing power in the near future and enhance the supply capacity of computing power in the medium to long termLaunch a batch of high-quality training data, plan to build a national level data training base, implement a pilot project for innovative application of big models, promote the empowerment of big models in various industries, cultivate new paradigms for software developmentImplement the foundational engineering of supporting technology at the bottom of the large modelAs you can see,Multiple sub tasks in the eight tasks are related to computing power, including short-term and medium to long-term computing power supply and demand in terms of time, and links involving aggregation, scheduling, supply, subsidies, etc/5-10At the same time, Beijing has also expanded its focus fromUnderlying supporting technologyIt is worth noting that in the "Promotion of Breakthroughs in Chip Manufacturing Processes", the "Plan" proposesMake up for the gap in advanced process technology with the advancement of Chiplet technology, and advance the layout of advanced computing chips, new technologies, and new architecturesThe computing power base may become the "most important issue in the next stage", and Chiplet is expected to play a crucial roleZeng Yi, Chairman of China Electronics, pointed out that among the three elements of artificial intelligence technology (computing power, algorithms, and models), algorithmic problems are expected to be properly solved. Although models are difficult, they can still be solved, and the most difficult problem is the problem of computing power base,The computing power base may be the most significant issue we need to address in the next stageDue to the increasing demand for GPT computing power, higher requirements have been placed on chip speed and capacity,Advanced packaging technology led by Chiplet is considered the best solution and key technology at presentChipletOn the one hand, it can break through the area constraints of SoC single chips, which is also a key support for system computing power; On the other hand, Chiplet can enhance communication bandwidth between computing and storage, as well as between computing and computing, alleviating the problem of "storage walls"AIGPUChipletGPUHBMAIChipletIDM|ChipletTianfeng Securities believes that from the perspective of the entire industry chain, as a new design concept, Chiplet,Elevated the leading position in design, IP, and EDA processes, with the potential to bring value increment to midstream manufacturing and downstream testing