Paper 2020/899

On the Attack Evaluation and the Generalization Ability in Profiling Side-channel Analysis

Lichao Wu, Léo Weissbart, Marina Krček, Huimin Li, Guilherme Perin, Lejla Batina, and Stjepan Picek

Abstract

Guessing entropy is a common metric in side-channel analysis, and it represents the average key rank position of the correct key among all possible key guesses. By evaluating it, we estimate the effort needed to break the implementation. As such, the guessing entropy behavior should be stable to avoid misleading conclusions about the attack performance. In this work, we investigate this problem of misleading conclusions from the guessing entropy behavior, and we define two new notions: simple and generalized guessing entropy. We demonstrate that the first one needs only a limited number of attack traces but can lead to wrong interpretations about the attack performance. The second notion requires a large (sometimes unavailable) number of attack traces, but it represents the optimal way of calculating guessing entropy. We propose a new metric (denoted the profiling model fitting metric) to estimate how reliable the guessing entropy estimation is. With it, we also obtain additional information about the generalization ability of the profiling model. We confirm our observations with extensive experimental analysis.