Dual Channel EM/Power Attack Using Mutual Information and its Real-time Implementation

Cryptosystem implementations often leak information about a secret key due to correlation with side channels such as power, timing, EM, etc. Based on this principle, statistical and machine-learning-based side-channel attacks have been investigated, most often using a single channel or modality such as power; however, EM is growing in popularity. Since power and EM channels can leak distinct information, the combination of EM and power channels could increase side-channel attack efficiency. In this paper, we combine EM and power channels in a linear fashion by using mutual information to determine the optimal coefficients for each feature. Mutual information is also systematically applied for lightweight dimensionality reduction. Further, the proposed methodology is implemented onto a platform to simultaneously measure power and EM traces and process them in real time to extract AES subkeys. With the proposed dual channel approach, the success rate increases by at least 30% compared to single power/EM channels in the offline mode and over 50% in the real-time mode.