Quantifying the Speed-Up Offered by Genetic Algorithms during Fault Injection Cartographies

In the context of Fault Injection Analyses, the determination of the correct set of physical perturbation parameters is critical. When searching for vulnerabilities against fault injections, it is then a necessity to carry out a cartography in order to establish which tuples of parameters allow to disturb the target successfully, in a reliable way. In practice, this task is often time consuming because of the large number of dimensions to consider, hence an exhaustive cartography is most of the time impossible.This paper analyses three different cartography strategies: Linear-Scan, Monte-Carlo, and Genetic Algorithm-based methods. We compare them in real Electro-Magnetic Fault Injection Analyses on an hardware device, distinguishing two different contexts, namely with few, and, at the opposite, with more Points of Interest. We show that Genetic Algorithms are always better for identifying Areas of Interest, and so correct injection parameters, which is crucial for characterizing vulnerabilities in security evaluation contexts.