Affiliation: Université catholique de Louvain, B1348 Louvain-la-Neuve
Multi-Tuple Leakage Detection and the Dependent Signal Issue
Leakage detection is a common tool to quickly assess the security of a cryptographic implementation against side-channel attacks. The Test Vector Leakage Assessment (TVLA) methodology using Welch’s t-test, proposed by Cryptography Research, is currently the most popular example of such tools, thanks to its simplicity and good detection speed compared to attack-based evaluations. However, as any statistical test, it is based on certain assumptions about the processed samples and its detection performances strongly depend on parameters like the measurement’s Signal-to-Noise Ratio (SNR), their degree of dependency, and their density, i.e., the ratio between the amount of informative and non-informative points in the traces. In this paper, we argue that the correct interpretation of leakage detection results requires knowledge of these parameters which are a priori unknown to the evaluator, and, therefore, poses a non-trivial challenge to evaluators (especially if restricted to only one test). For this purpose, we first explore the concept of multi-tuple detection, which is able to exploit differences between multiple informative points of a trace more effectively than tests relying on the minimum p-value of concurrent univariate tests. To this end, we map the common Hotelling’s T2-test to the leakage detection setting and, further, propose a specialized instantiation of it which trades computational overheads for a dependency assumption. Our experiments show that there is not one test that is the optimal choice for every leakage scenario. Second, we highlight the importance of the assumption that the samples at each point in time are independent, which is frequently considered in leakage detection, e.g., with Welch’s t-test. Using simulated and practical experiments, we show that (i) this assumption is often violated in practice, and (ii) deviations from it can affect the detection performances, making the correct interpretation of the results more difficult. Finally, we consolidate our findings by providing guidelines on how to use a combination of established and newly-proposed leakage detection tools to infer the measurements parameters. This enables a better interpretation of the tests’ results than the current state-of-the-art (yet still relying on heuristics for the most challenging evaluation scenarios).
Leakage Certification Revisited: Bounding Model Errors in Side-Channel Security Evaluations 📺
Leakage certification aims at guaranteeing that the statistical models used in side-channel security evaluations are close to the true statistical distribution of the leakages, hence can be used to approximate a worst-case security level. Previous works in this direction were only qualitative: for a given amount of measurements available to an evaluation laboratory, they rated a model as “good enough” if the model assumption errors (i.e., the errors due to an incorrect choice of model family) were small with respect to the model estimation errors. We revisit this problem by providing the first quantitative tools for leakage certification. For this purpose, we provide bounds for the (unknown) Mutual Information metric that corresponds to the true statistical distribution of the leakages based on two easy-to-compute information theoretic quantities: the Perceived Information, which is the amount of information that can be extracted from a leaking device thanks to an estimated statistical model, possibly biased due to estimation and assumption errors, and the Hypothetical Information, which is the amount of information that would be extracted from an hypothetical device exactly following the model distribution. This positive outcome derives from the observation that while the estimation of the Mutual Information is in general a hard problem (i.e., estimators are biased and their convergence is distribution-dependent), it is significantly simplified in the case of statistical inference attacks where a target random variable (e.g., a key in a cryptographic setting) has a constant (e.g., uniform) probability. Our results therefore provide a general and principled path to bound the worst-case security level of an implementation. They also significantly speed up the evaluation of any profiled side-channel attack, since they imply that the estimation of the Perceived Information, which embeds an expensive cross-validation step, can be bounded by the computation of a cheaper Hypothetical Information, for any estimated statistical model.