These attacks can be seen as a classification problem, where the adversary needs to decide to what class (and consequently, the secret key) the traces collected from the victim’s device belong. The research community investigated profiling attacks in-depth, primarily by using an empirical approach. As such, it emerges that a theoretical framework to analyze profiling side-channel attacks comprehensively is still missing.
In this paper, we propose a theory-grounded framework capable of modeling and evaluating profiling side-channel analysis. The framework is based on the expectation estimation problem that has strong theoretical foundations. We quantify the effects of perturbations injected at different points in our framework through the robustness analysis, where the perturbations represent sources of uncertainty associated with measurements, non-optimal classifiers, and countermeasures. Finally, we use our framework to evaluate the performance of different classifiers using publicly available traces.