Statistical Tools Flavor Side-Channel Collision Attacks

By examining the similarity of side-channel leakages, collision attacks evade the indispensable hypothetical leakage models of multi-query based side-channel distinguishers like correlation power analysis and mutual information analysis attacks. Most of the side-channel collision attacks compare two selective observations, what makes them similar to simple power analysis attacks. A multi-query collision attack detecting several collisions at the same time by means of comparing the leakage averages was presented at CHES 2010. To be successful this attack requires the means of the side-channel leakages to be related to the processed intermediate values. It therefore fails in case the mean values and processed data are independent, even though the leakages and the processed values follow a clear relationship. The contribution of this article is to extend the scope of this attack by employing additional statistics to detect the colliding situations. Instead of restricting the analyses to evaluation of means, we propose to employ higher-order statistical moments and probability density functions as the figure of merit to detect collisions. Thus, our new techniques remove the shortcomings of the existing correlation collision attacks using first-order moments. In addition to the theoretical discussion of our approach, practical evidence of its suitability for side-channel evaluation is provided. We provide four case studies, including three FPGA-based masked hardware implementations and a software implementation using boolean masking on a microcontroller, to support our theoretical groundwork.