Comparison between side-channel analysis distinguishers

Abstract

Side-channel analyses allow to extract keys from devices whatever their length. They rely on tools called "distinguishers". In this paper, we intend to compare two generic distinguishers per se: we provide a characterization environment where all the implementation details are equal, hence a fair comparison. In the field of distinguishers that use a model, the notion of equivalence between distinguishers has already been studied in some seminal works [6,13]. However, no such work has been carried out for generic distinguishers, that work on observable values distributions rather than on their values themselves. In this paper, we set up simulations that aim at showing experimentally that two generic distinguishers are different. Then, we develop a theory to actually prove that one distinguisher is better than the other. © 2012 Springer-Verlag.