Measuring the effectiveness of DPA attacks-from the perspective of distinguishers' statistical characteristics

Jingang Huang, Yongbin Zhou, Jiye Liu

Abstract: Distinguisher serves as an essential component in DPA attacks and should to some extent influence behaviors of these attacks. Motivated by this, we proposed a sound approach to evaluating the effectiveness of DPA attacks from the perspective of distinguishers' statistical characteristics. For this propose, we formally defined the notion of Gaussian Distinguisher in one typical DPA attack setting and then proved that two most widely used DPA distinguishers (namely difference-of-means test and Pearson correlation coefficient) were Gaussian. After that, Distinctive Level, a useful quantitative metric, was introduced to evaluate the effectiveness of DPA attacks. This metric virtually equips the designer with the capability of judging to what extent DPA attacks will succeed. We performed experiments using both simulated and real power traces afterwards, the results of which evidently demonstrated the validity and the effectiveness of the methods we had proposed.