Learning to attack from electromagnetic emanation

Abstract

Sensitive information processed by the circuitry in electronic security devices can be leaked via physical characteristics of the device, such as power consumption, electromagnetic (EM) emanation, timing, etc. These techniques are known as Side-Channel Attacks (SCA). To date, a significant amount of research has been carried out into side channel attacks, which uses statistical processing techniques to analyse the information leaked from the device. This work formalized the problem of studying the relation between EM emanation and encryption key as a supervised learning task. The considered technique is Support Vector Machine (SVM). The chosen side channel is the EM emanation and the target is a software implementation of the Data Encryption Standard (DES). In this study, several feature selection techniques are compared in a real experimental setting. Our promising results regarding the DES encryption scheme confirms the importance of adopting SVM in cryptanalysis and the effectiveness of our approach in feature selection.