AnalogFed: Privacy-Preserving Discovery of Analog Circuits at Scale with Federated Generative AI

AnalogFed is a novel privacy-preserving framework designed for large-scale analog circuit topology discovery, utilizing federated learning and generative AI. It addresses security challenges such as membership inference and model inversion attacks through techniques like dummy token injection and efficient homomorphic encryption. This framework enables collaborative hardware design while maintaining data confidentiality, paving the way for scalable electronic design automation in the semiconductor industry.