Wide collisions in practice

Abstract

Detecting collisions in the power consumption of a modern cryptographic engine is usually difficult due to small leakages and possible countermeasures. Wide collisions offer a much stronger leakage that significantly facilitates their detection. This is the first time wide collisions are exploited in a power analysis attack. In this work we introduce a collision detection method based on the detection of characteristic outliers. Detection results are compared to optimized subspace-based templates. We show that the outlier detection method, while not requiring a template building phase, is almost as effective in detecting collisions as the template-based approach. © 2012 Springer-Verlag.