This paper focuses on the innovative developments in AI/ML as a data-driven approach for analog circuit design automation. In particular, there is growing interest in research on automatically discovering novel analog circuit topologies using generative AI. However, due to the unique characteristics of analog circuit design (including confidential circuit structures and underlying commercial semiconductor processes), it is difficult to access large and diverse datasets. To address these issues, this paper proposes AnalogFed, which enables collaborative topology discovery among distributed clients such as individual researchers or institutions. AnalogFed enables collaboration without sharing raw private data, and introduces techniques required to apply FedL to analog design, such as generative model development, data heterogeneity handling, and privacy-preserving strategies. Through extensive experiments on various client numbers and dataset sizes, we demonstrate that AnalogFed achieves comparable performance to centralized baselines while maintaining strict data privacy. In particular, the generative AI model within AnalogFed achieves state-of-the-art efficiency and scalability in analog circuit topology design.
