Multivariate TVLA - Efficient Side-Channel Evaluation using Confidence Intervals

Securing cryptographic hardware designs and software implementations against side-channel attacks that leverage the power consumption or electromagnetic emanations of a device is an active topic of research. Different countermeasures against these attacks have been published, many of which rely on masking where sensitive information is split into multiple shares. Here, the information is hidden in higher statistical moments of the leakage if processed at the same time (univariate) or in combinations of side-channel information from different points in time (multivariate) if processed sequentially. Test Vector Leakage Assessment (TVLA) is a common evaluation technique to address the growing number of specific attacks. However, the assessment of multivariate leakage requires the evaluation of all possible combinations of sample points, massively slowing down the evaluation and in turn the development of countermeasures due to computational complexity.

In this work, we develop and compare techniques to determine clock cycle combinations that leak information in a multivariate setting. We develop an efficient multivariate assessment framework and show how this approach can be used to generate evaluation results that satisfy a desired confidence level. Eventually, we demonstrate the practical relevance of our approach by applying it to two masked implementations of block ciphers.