Session

DescriptionAs AI accelerators become increasingly integral to modern computing, ensuring their security and privacy is paramount. This session explores the key challenges in this domain, focusing on side-channel vulnerabilities, privacy-preserving computation techniques, and the design of secure deep learning hardware architectures. Topics include power-based attacks on XGBoost accelerators, efficient zero-knowledge proofs for verifiable computing, and fully homomorphic encryption (FHE) acceleration for client-side privacy. Additionally, the session covers secure DNN accelerators leveraging hardware/software co-design and lightweight reconfigurable computing to protect AI intellectual property.