Leakage Assessment through Neural Estimation of the Mutual Information

Valence Cristiani (1) , Maxime Lecomte (1) , Philippe Maurine (2)

1 CEA-LETI - Commissariat à l'énergie atomique et aux énergies alternatives - Laboratoire d'Electronique et de Technologie de l'Information
2 SmartIES - Smart Integrated Electronic Systems

A large variety of side-channel attacks have been developed to extract secrets from electronic devices through their physical leakages. Whatever the utilized strategy, the amount of information one could gain from a side-channel trace is always bounded by the Mutual Information (MI) between the secret and the trace. This makes it, all punning aside, a key quantity for leakage evaluation. Unfortunately, traces are usually of too high dimension for existing statistical estimators to stay sound when computing the MI over full traces. However, recent works from the machine learning community have shown that it is possible to evaluate the MI in high dimensional space thanks to newest deep learning techniques. This paper explores how this new estimator could impact the side channel domain. It presents an analysis which aim is to derive the best way of using this estimator in practice. Then, it shows how such a tool can be used to assess the leakage of any device.

Mots clés: Side channel analysis; Mutual information; Deep learning; Hardware security

Domaines: Informatique [cs]; Mathématiques [math]

Valence Cristiani, Maxime Lecomte, Philippe Maurine. Leakage Assessment through Neural Estimation of the Mutual Information. ACNS 2020 - International Conference on Applied Cryptography and Network Security, Oct 2020, Rome, Italy. pp.144-162, ⟨10.1007/978-3-030-61638-0_9⟩. ⟨hal-02980501⟩