AnalogAgent: Self-Improving Analog Circuit Design Automation with LLM Agents

Abstract

Recent advances in large language models (LLMs) suggest strong potential for automating analog circuit design. Yet most LLM-based approaches rely on a single-model loop of generation, diagnosis, and correction, which favors succinct summaries over domain-specific insight and suffers from context attrition that erases critical technical details. To address these limitations, we propose AnalogAgent, a training-free agentic framework that integrates an LLM-based multi-agent system (MAS) with self-evolving memory (SEM) for analog circuit design automation. AnalogAgent coordinates a Code Generator, Design Optimizer, and Knowledge Curator to distill execution feedback into an adaptive playbook in SEM and retrieve targeted guidance for subsequent generation, enabling cross-task transfer without additional expert feedback, databases, or libraries. Across established benchmarks, AnalogAgent achieves 92% Pass@1 with Gemini and 97.4% Pass@1 with GPT-5. Moreover, with compact models (e.g., Qwen-8B), it yields a +48.8% average Pass@1 gain across tasks and reaches 72.1% Pass@1 overall, indicating that AnalogAgent substantially strengthens open-weight models for high-quality analog circuit design automation.

Figures (6)

• Figure 1: Illustration of context attrition on the hard Task 25. As context accumulates across iterations, LLM agents tend to compress it into shorter, less informative summaries, diminishing instruction salience and degrading detailed guidance. AnalogCoder Pro takes $\sim$7 minutes to produce its first correct circuit, whereas AnalogAgent succeeds in 17.13 seconds with learned knowledge.
• Figure 2: AnalogAgent Framework Overview.Left: Existing LLM-based methods typically rely on a single LLM with static prompts, which often produce suboptimal solutions even when augmented with external circuit libraries. Right: AnalogAgent coordinates multiple agents and curates reusable design knowledge. It retrieves prior knowledge from SEM to refine instructions and perform self-improvement, improving convergence toward functionally complete circuits without additional libraries.
• Figure 3: Self-Improving Refinement Case Study. The flow shows an error-driven refinement loop for Hard Task 25, in which execution failures are diagnosed, distilled into curator rules, and written into Adaptive Design Playbook to guide subsequent generations and accelerate convergence. Hard Task 25 is an op-amp comparator that outputs high when $V_{\text{in}} > V_{\text{ref}}$ and low when $V_{\text{in}} < V_{\text{ref}}$. The final iterations correct the topology and pin semantics to meet this specification.
• Figure 4: Cross-Task Transfer across Amplifiers. Reusable knowledge distilled from the basic Amplifier Task 1 transfers to Tasks 9 and Task 10, improving success rates and demonstrating that long-term memory is effectively reusable across tasks.
• Figure 5: Performance comparison of Qwen base models, AnalogCoder-Pro, and AnalogAgent. AnalogAgent consistently enhances compact Qwen models across all model scales.