AnalogSAGE: Self-evolving Analog Design Multi-Agents with Stratified Memory and Grounded Experience

2025-12-26

AI Summary

Analog circuit design heavily relies on expert knowledge, and existing large language model (LLM)-based approaches—constrained by template-based or prompt-driven paradigms—struggle to meet complex, multi-objective specifications. Method: This paper proposes a simulation-feedback-driven self-evolving multi-agent framework. It introduces a novel four-level hierarchical memory mechanism and a simulation-anchored, three-stage collaborative exploration paradigm, enabling template-free, prompt-free end-to-end topology generation and parameter optimization. The framework integrates closed-loop ngspice simulation, SKY130 PDK process modeling, and specification-guided search. Contribution/Results: Evaluated on ten operational amplifier benchmark tasks, the framework achieves a 10× improvement in overall pass rate, a 48× increase in Pass@1, and a 4× reduction in parameter search space. These results demonstrate substantial gains in reliability, generalizability, and automation capability for analog circuit synthesis.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Cognitive Modeling & Cognitive Systems: Analogy
• Multiagent Systems: Mechanism Design

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Web data generation and simulation

Abstract

Analog circuit design remains a knowledge- and experience-intensive process that relies heavily on human intuition for topology generation and device parameter tuning. Existing LLM-based approaches typically depend on prompt-driven netlist generation or predefined topology templates, limiting their ability to satisfy complex specification requirements. We propose AnalogSAGE, an open-source self-evolving multi-agent framework that coordinates three-stage agent explorations through four stratified memory layers, enabling iterative refinement with simulation-grounded feedback. To support reproducibility and generality, we release the source code. Our benchmark spans ten specification-driven operational amplifier design problems of varying difficulty, enabling quantitative and cross-task comparison under identical conditions. Evaluated under the open-source SKY130 PDK with ngspice, AnalogSAGE achieves a 10$ imes$ overall pass rate, a 48$ imes$ Pass@1, and a 4$ imes$ reduction in parameter search space compared with existing frameworks, demonstrating that stratified memory and grounded reasoning substantially enhance the reliability and autonomy of analog design automation in practice.

Problem

• Automates analog circuit design using multi-agent framework with memory layers.
• Improves reliability and autonomy over existing LLM-based topology generation methods.
• Reduces parameter search space and increases pass rates for operational amplifiers.

Innovation

• Multi-agent framework coordinates three-stage explorations
• Four stratified memory layers enable iterative refinement
• Simulation-grounded feedback enhances reliability and autonomy