Active and Passive Physical Attacks on Neural Network Accelerators

This article emphasizes the growing importance of studying defenses for physical attacks on neural network accelerators, and describes approaches based on side-channel attacks and fault-injection attacks.