AnalogCoder: Analog Circuit Design via Training-Free Code Generation

模拟电路设计是现代芯片技术中的重要任务，它着眼于选择组件类型、连接性和参数以确保电路功能的正确性。虽然大型语言模型在数字电路设计方面取得了进展，但模拟电路中的复杂性和数据稀缺性带来了重大挑战。为了缓解这些问题，我们介绍了AnalogCoder，这是第一个通过Python代码生成进行模拟电路设计的无需训练的大型语言模型代理。首先，AnalogCoder采用带有定制域特定提示的反馈增强流程，实现了自动化和自我纠正的模拟电路设计，成功率高。其次，它提出了一个电路工具库，将成功的设计存档为可重复使用的模块化子电路，简化了复合电路的创建。第三，广泛的实验覆盖了各种模拟电路任务，结果显示AnalogCoder优于其他基于大型语言模型的方法。它成功设计了20个电路，比标准的GPT-4o多5个。我们相信AnalogCoder能够显著提高劳动密集型的芯片设计过程，使非专业人士能够高效地设计模拟电路。代码和基准可以在https://github.com/anonyanalog/AnalogCoder 上找到。

Analog circuit design is a challenging task due to the complexity and scarcity of data, and the paper aims to introduce a training-free Large Language Model (LLM) agent called AnalogCoder to automate the design process.

AnalogCoder uses a feedback-enhanced flow with tailored domain-specific prompts and proposes a circuit tool library to archive successful designs as reusable modular sub-circuits, enabling the automated and self-correcting design of analog circuits with a high success rate. AnalogCoder outperforms other LLM-based methods in extensive experiments on a benchmark designed to cover a wide range of analog circuit tasks.

AnalogCoder is the first training-free LLM agent for designing analog circuits through Python code generation. It simplifies composite circuit creation by proposing a circuit tool library to archive successful designs as reusable modular sub-circuits. AnalogCoder successfully designed 20 circuits, 5 more than standard GPT-40, in extensive experiments on a benchmark designed to cover a wide range of analog circuit tasks. The codes and benchmark are provided at https://github.com/anonyanalog/AnalogCoder.

Recent related work includes advances in Large Language Models for digital circuit design, but AnalogCoder is the first to focus on analog circuit design through Python code generation.