Hai (Helen) Li is the Marie Foote Reel E’46 Distinguished Professor and Department Chair of the Electrical and Computer Engineering Department at Duke University. She received her B.S. and M.S. degrees from Tsinghua University, and her Ph.D. degree from Purdue University. Her research interests include neuromorphic circuits and systems for brain-inspired computing, machine learning acceleration and trustworthy AI, conventional and emerging memory design and architecture, and software and hardware co-design. Dr. Li has received many awards, including the IEEE A. Richard Newton Technical Impact Award, the IEEE Charles A. Desoer Technical Achievement Award, the IEEE Edward J. McCluskey Technical Achievement Award, Ten Year Retrospective Influential Paper Award from ICCAD, TUM-IAS Hans Fischer Fellowship from Germany, ELATE Fellowship, and ten best paper awards from IEEE/ACM. Dr. Li is a fellow of AAAS, ACM, IEEE, and NAI.

The semiconductor industry faces a growing gap between rapidly increasing design complexity and the productivity of traditional electronic design automation (EDA) methodologies. While machine learning has delivered significant advances in prediction tasks such as timing estimation, routability analysis, and design optimization, recent breakthroughs in foundation models have opened new opportunities for generative and autonomous design workflows. This talk reviews the evolution of AI for chip design, from early machine-learning-assisted EDA to the emerging era of large language models (LLMs), diffusion models, and agentic AI systems. Through a series of case studies, we demonstrate how domain-adapted foundation models can generate analog circuit topologies, produce manufacturable layout patterns, optimize EDA tool parameters, and automate RTL-to-GDSII design flows. We highlight key enabling techniques including retrieval-augmented generation (RAG), specialized circuit representations and tokenization, supervised post-training, diffusion-based customization, and Model Context Protocol (MCP)-based agent frameworks. Beyond automation, we discuss a broader vision for AI-native chip design in which autonomous agents coordinate complex workflows, explore design spaces at unprecedented scale, and increasingly move beyond human-crafted heuristics. The talk concludes with opportunities and challenges for building the next generation of intelligent EDA systems that combine reasoning, generation, and orchestration to accelerate semiconductor innovation.