Human in the Loop AI EDA for Chip Design

Currently, chip design automation is hampered by factors such as data scarcity, difficulty in sharing, and high costs associated with design rules and verification. Based on the research trajectory of human-machine collaborative learning, this paper proposes a modular framework that embeds expert feedback into three stages: data processing, model training, and system-level intervention, implementing it within the placement and routing workflow. With less human input, the number of iterations decreases. Economic accounting connects marginal quality gains with human and computing costs, exhibiting diminishing returns and context-dependent optimal feedback budgets.