AnalogFed: Privacy-Preserving Discovery of Analog Circuits at Scale with Federated Generative AI
Researchers have developed AnalogFed, a novel framework that combines federated learning and generative AI to enable privacy-preserving discovery of analog circuit topologies. This approach addresses the challenge of using proprietary and siloed hardware datasets for large-scale electronic design automation by allowing collaborative training without centralizing sensitive data. AnalogFed incorporates defenses against membership inference and model inversion attacks, demonstrating its effectiveness in protecting privacy while maintaining model utility for next-generation hardware design. AI
IMPACT Enables collaborative, privacy-preserving AI development for hardware design, potentially accelerating innovation in the EDA field.