Abstract
Analog/mixed-signal circuits are key for interfacing electronics with thephysical world. Their design, however, remains a largely handcrafted process,resulting in long and error-prone design cycles. While the recent rise ofAI-based reinforcement learning and generative AI has created new techniques toautomate this task, the need for many time-consuming simulations is a criticalbottleneck hindering the overall efficiency. Furthermore, the lack ofexplainability of the resulting design solutions hampers widespread adoption ofthe tools. To address these issues, a novel agentic AI framework forsample-efficient and explainable analog circuit sizing is presented. It employsa multi-agent workflow where specialized Large Language Model (LLM)-basedagents collaborate to interpret the circuit topology, to understand the designgoals, and to iteratively refine the circuit's design parameters towards thetarget goals with human-interpretable reasoning. The adaptive simulationstrategy creates an intelligent control that yields a high sample efficiency.The AnaFlow framework is demonstrated for two circuits of varying complexityand is able to complete the sizing task fully automatically, differently frompure Bayesian optimization and reinforcement learning approaches. The systemlearns from its optimization history to avoid past mistakes and to accelerateconvergence. The inherent explainability makes this a powerful tool for analogdesign space exploration and a new paradigm in analog EDA, where AI agentsserve as transparent design assistants.