Spatial Dependency Analysis to Extract Information from Side-Channel Mixtures

Abstract

Practical side-channel attacks on recent devices may be challenging due to the poor quality of acquired signals. It can originate from different factors, such as the growing architecture complexity, especially in System-on-Chips, creating unpredictable and concurrent operation of multiple signal sources in the device. This work makes use of mixture distributions to formalize this complexity, allowing us to explain the benefit of using a technique like Scatter, where different samples of the traces are aggregated into the same distribution. Some observations of the conditional mixture distributions are made in order to model the leakage in such context. From this, we infer local coherency of information held in the distribution as a general expression of the leakage in mixture distributions. This leads us to introduce how spatial analysis tools, such as Moran's Index, can be used to significantly improve nonprofiled attacks compared to other techniques from the state-of-theart. Exploitation of this technique is experimentally shown very promising, as demonstrated by its application to ASCAD dataset.