Qixin Chenguang Chen Wenchao: Semiconductor data does not leave the domain, how do 38 agents break through the "black box" of chip design

Zhidongxi Editor | Zhidongxi Editorial Department Zhidongxi reported on July 9 that from July 2 to 3, the 2026 China AI Intelligence Conference with the theme of "Paradigm Leap Reshaping the World" was successfully held in Hangzhou. The conference gathered 64 heavyweight guests. Through an opening ceremony, two forums, and seven closed seminars, the core topics of the Agent track were fully analyzed. The topics covered Harness, self-evolving Agents, Coding Agents, multi-Agent collaboration, and Skills. At the meeting, Chen Wenchao, vice president of Qixin Chenguang and chief architect of the EDA intelligent computing platform, gave a keynote speech titled "Tenfold Improvement in AI-driven Chip Design Efficiency: The Evolution of the DeepChip Platform", sharing Qixin Chenguang's thoughts, core layout and existing implementation results on the implementation of AI in the semiconductor industry. Chen Wenchao introduced how the DeepChip chip design collaborative innovation platform introduces AI technology into four key semiconductor scenarios: design environment (DeepEDA), chip design (DeepIP), chip testing (DeepATE) and talent training (DeepEDU). He said that DeepEDA has been adapted to mainstream EDA tools and supports offline deployment of local models such as Llama and DeepSeek. DeepIP combines AI Agent with tool chains and knowledge bases, and has packaged 38 agents. Together with a data knowledge base of more than 500G, it can achieve full AI intervention from system design to debugging optimization, increasing per capita efficiency by 20 times and increasing the average automation rate by 91.5%. DeepATE can improve the efficiency of automated chip testing and shorten mass production time. With the help of this platform, a well-known domestic test machine manufacturer has shortened the test program writing cycle by about 29 days. DeepEDU has built a "one base and four centers" teaching and research service platform, providing more than 20,000 courses. The following is the transcript of the speech: 1. Semiconductor data does not go out of the domain, and there is a threshold for the full implementation of general-purpose large models. Dear experts and colleagues, good afternoon. I am honored to be able to