AutoProfile: Automated profiling in deep learning-based side-channel analysis

Side-channel analysis (SCA) capitalizes on unintentionally leaked information to extract sensitive data from cryptographic systems. Over recent years, the side-channel community has exhibited a notable surge in interest towards deep learning (DL) techniques. However, the challenge of constructing appropriate networks has been highlighted. This paper presents a novel methodology named AutoProfile to enhance the efficacy of DL-based profiling attack on robust cryptographic system. AutoProfile customizes two core components of Bayesian optimization for SCA: the modeling strategy and acquisition function. The performance of AutoProfile is assessed on publicly available datasets that consist of real side-channel measurements. The experimental results showcase substantial improvements over the state-of-the-art method, achieving an average performance enhancement of 78.4 %. Notably, for more robust targets that employ a combination of masking, random delay and key variation countermeasures, AutoProfile delivers the most significant performance boost. It reduces the number of traces required from the thousands, as typically needed by state-of-the-art methods, to merely dozens to break the targets. Furthermore, the experiments show that AutoProfile is notably faster than baseline methods in identifying effective networks over all tested SCA datasets.