AgenticECO: AI Agent Framework for 3D-IC Engineering Change Orders

A recent paper published on arXiv (2608.03738) presents AgenticECO, a novel framework aimed at managing engineering change orders (ECO) in three-dimensional integrated circuits (3D-ICs). With Moore's law decelerating, the industry is transitioning to 3D integration; however, these merged 3D-IC processes reveal bond-level defects absent in 2D designs, and the post-route ECO process remains reliant on manual expertise. The conventional method of editing followed by complete rerouting complicates repairs, preventing the assignment of signoff numbers to individual edits. AgenticECO utilizes an evidence-gated, tool-assisted agent workflow based on the open-source TaiWei flow, along with EcoRoute, a minimal-disturbance ECO-routing layer that ensures each repair corresponds to its edit. In trials involving nine matched natural defect scenarios with identical budgets, AgenticECO successfully resolved seven cases, while both full reroute and stock repair managed only two, achieving an average disturbance of 0.66% and leaving clock nets untouched. A rerun across the same sealed contract also proved successful. The paper's abstract emphasizes the challenges of 3D-IC ECO and the effectiveness of the proposed solution.