AI Begins to "Design Chips": Chen Yunji Discusses AI for IC and IC for AI, Scientists Stress That Asking Good Questions Is Most Critical

As the dividends of Moore's Law fade, demand for computing power is still surging. At the 2026 World Laureates Forum, multiple scientists and industry figures noted that AI is no longer merely an auxiliary tool for the chip industry; it is reaching into the deepest layers of chip design, architectural innovation, and even wafer fabrication equipment. And in the next decade, how the chip industry will develop and how innovative research talent will be cultivated have also become topics facing the scientific and industrial communities alike.

AI for IC

"In the AI era, the two major cornerstones of AI and IC (integrated circuits) will merge and converge," Chen Yunji, deputy director of the Institute of Computing Technology at the Chinese Academy of Sciences, said at the AI parallel forum. On one hand is AI for IC, using artificial intelligence to automatically design better chips; on the other is IC for AI, using chips to make AI inference and training more efficient. He further envisioned that combining these two directions could even form a new RSI (recursive self-improvement) loop in the future—AI designs better AI chips, and AI chips in turn feed back into better AI technology.

Chen Yunji gave an example: last year his team had AI automatically design the Qimeng No. 2 chip, which has capabilities such as automatic pipeline slicing and automatic cache addition, with performance comparable to ARM A53, roughly the level of mainstream human-designed CPUs from more than a decade ago. "We want to keep going down this path; sooner or later, processor chips automatically designed by AI will be comparable to, or even better than, human-designed chips."

On the manufacturing and equipment side, AI's driving force is also direct. Lu Guangquan, chairman of Piotech, said in a keynote speech at the Future Semiconductor parallel forum that as design costs at the 2nm node soar, only 5% of IC/ASIC projects will achieve first-pass tape-out success in 2026 (that figure was 32% in 2012). Faced with high trial-and-error costs, semiconductor technology is moving toward three major trends: smaller transistors, denser packaging, and smarter equipment.

Lu Guangquan introduced that semiconductor equipment is extremely complex, and traditional physical debugging is costly and time-consuming. Equipment giants are bringing AI into the interior of equipment to achieve "virtual manufacturing." Giants such as Lam use physical models to build virtual fabs (digital twins), increasing the speed of specific etching modeling steps by up to 30 times; Applied Materials (AMAT) uses an AI platform for real-time process analysis, processing more than one million three-dimensional wafer measurements per hour.

At Piotech, through massive two-way data mapping between physical chambers and virtual chambers, real-time process prediction, equipment health monitoring (predictive maintenance), and wafer transfer scheduling optimization can be achieved.

Lu Guangquan also revealed a striking efficiency revolution: "In scenarios such as failure analysis, AI agents can automatically execute test tasks, compressing work that originally required hundreds of labor-hours to within 5.5 hours."

Regarding talent cultivation in the chip industry, Wu Hanming, director of the Faculty of Information Technology at Zhejiang University, believes that in the context of AI-driven automation, talent must transform from the "pure layout drafters" of the past, who focused only on geometric layout and manual adjustment, to "system architects" who emphasize system specifications, algorithm design, and architectural exploration. The logic of education must shift from "implementation-centered" to "innovation-centered."

To bridge the gap between production lines and the classroom, Wu Hanming's team is promoting the "Virtual-First, Physical-Validation" FabGPT solution and implementing "one student, one chip," allowing university students to actually handle silicon wafers on 12-inch production lines and experience a real production-line environment.

English text is being prepared.

This view was repeatedly mentioned in the remarks of the top scientists attending the forum. Michael Levitt, winner of the 2013 Nobel Prize in Chemistry and professor of structural biology at Stanford University, said at the forum opening ceremony: "With artificial intelligence, young researchers can verify in an afternoon ideas that previously took a whole year to test. So the scarce resource is no longer hard work, but high-quality scientific questions."

"AI cannot ask a good question," Yuan Junying, director of the Interdisciplinary Research Center for Biology and Chemistry at the Shanghai Institute of Organic Chemistry, Chinese Academy of Sciences, said in an interview with media including Yicai. "Now our team is 'forcing' students to ask questions."

She shared her own research experience. Years ago, while still a PhD student at Harvard University, she raised a simple question that was considered obscure at the time and was even discouraged by her advisor: "How do cells die?" It was precisely this "good question," for which few peers could even be found in the literature at the time, that ultimately opened up an entirely new field of programmed cell death research.

"AI is a very good accelerator, but it currently cannot solve all physical mechanism problems. Don't think that because we have AI, we no longer need to think; asking good questions is still the most critical thing," said Chen Peining, a professor at Fudan University and a winner of this year's WLF award, in response to a question from a Yicai reporter.

Faced with the efficiency revolution and research anxiety brought by AI, Shao Feng, a winner of this year's WLF award and deputy director of the National Institute of Biological Sciences, Beijing, called in an interview for recognizing that the essence of scientific research is exploring the unknown, and that the entire ecosystem should allow young scientists to focus on scientific questions themselves, dare to try, and dare to take risks.

"Experimental scientific research is destined to be accompanied by failure; be mentally prepared that most attempts will go wrong, and then keep moving forward," Shao Feng said.