Exploitable Fault Space Characterization: A Complementary Approach

Abstract

Consider the problem of estimating the success rate of an attacker for a given fault location. Intuitively, the success rate indicates the likeliness of an attack for returning the key bits within a practical computational bound. It is a quantity worth estimating as it helps in choosing cipher constructs and sub-operations showing a certain amount of robustness against fault attacks. In this chapter, we present a framework, which can estimate quantities like success rate and can provide new insights into the cipher structures in the context of fault attacks. The main idea is to utilize the exact encoding of algebraic fault attacks (AFA) for exploring the fault space without getting affected by its prohibitive time complexity. We propose a machine learning (ML) based speedup strategy to make the AFA suitable for characterizing huge fault spaces statistically. This statistical approach is found to be extremely informative for cipher designers and evaluators.