Cross-Device Deep Learning Side-Channel Attacks using Filter and Autoencoder

Side-channel Analysis (SCA) attacks are known as effective methods for extracting the encryption keys. These methods are improved by using deep learning (DL) techniques such that much stronger attacks can be carried out with lower level of effort. However, DL-boosted side channel attacks in cross-device cases is very applicative and challenging. In this kind of attack, profiling done on an existing device while the final attack will done on another device which may be similar to the profiled device but not the same. We proposed a new approach for improving the DL-based SCA attack in cross-device application. Due to significant device-to-device process variations, the accuracy of the existing methods is not adequate for real attacks. The proposed method improves the attack using the pre-processing methods based on a combination of DL-based autoencoder and Gaussian low-pass filter (GLPF). According to our experimental results, the accuracy of attack using the deep learning-based autoencoder can be improved by about 70%. It can also be improved up to 82% by adding the GLPF technique simultaneously.