Breaking a Masked AES Implementation Using a Deep Learning-based Attack

Abstract

The block cipher AES (Advanced Encryption Standard) is a cryptographic algorithm used to guarantee the confidentiality of a message. A masked implementation of AES is often used to increase resistance against SCA (Side Channel Attacks). This paper presents some deep learning-based attacks for extracting AES secret keys embedded in cryptographic devices. The proposed attack methods represent new approaches to computing the secret key by applying the mask profiling techniques. The MLP (Multi-Layer Perceptron) and CNN (Convolutional Neural Network) deep learning models are developed to break the masked AES implementation. Our experimental results show the overwhelming advantages of the novel attack methods when targeting both unmasked and masked implementation of AES.