Horizontal Attacks using K-Means: Comparison with Traditional Analysis Methods

AI means are widely used to detect correlations in large data sets. This makes them an ideal candidate to improve side channel analysis attacks as the core feature if these attacks is to reveal the correlation between measurement values and the key bits processed. In this paper we present an assessment of AI means, i.e. k-means. We investigated the success rate of attacks against three designs with different levels of vulnerability. The result was that even though counter intuitive approaches such as the Pearson correlation coefficient outperform k-means. The highest success rate of the latter was 68.7 per cent of an uncompressed trace and 88.3 per cent for a compressed trace whereas the Pearson correlation coefficient achieved 91.7 per cent for the uncompressed trace and 89.3 per cent for the compressed trace.