A profiled side‐channel attack detection using deep learning model with capsule auto‐encoder network

Abstract

AbstractSide‐channel analysis (SCA) is a type of cryptanalytic attack that uses unintended ‘side‐channel’ leakage through the real‐world execution of the cryptographic algorithm to crack a secret key of an embedded system. These side‐channel errors can be discovered through tracking the energy usage of the device performing the technique, electromagnetic radiations while the encryption process, execution time, cache hits/misses, and others. Nowadays, deep learning‐based detection techniques are considered as emerging techniques that have been proposed for attack detection. Deep learning architectures have the ability to learn autonomously and concentrate on difficult features, in contrast to machine learning models. In light of these factors, the work's motive is thought to be the proposal of a deep learning‐based attack detection method. Many methods are used to decrease these assaults, however, the majority of them are inefficient and time‐demanding. In order to address these challenges, this study employs a novel deep learning‐based methodology. Pre‐processing, feature extraction, and SCA classification are the three stages of the approach proposed in this work. First, pre‐processing is used to remove unnecessary information and improve the quality of the input using data cleaning and min‐max normalization. The previously processed data are then fed as input into the proposed hybrid deep learning architecture. A Deep Residual Capsule Auto‐Encoder (DR_CAE) model is introduced in the proposed study. The deep residual neural network‐50 (DRNN‐50) is utilized to extract relevant features in this case, while the side channel analysis is done by using capsule auto‐encoder (CAE). The parameters of the proposed model are adjusted using the modified white shark optimization (MWSO) technique to improve its performance. In the results section, the proposed model is compared to various existing models in terms of accuracy, precision, recall, F‐measures, time, and so on. The proposed framework has an accuracy of 98.802%, F‐measures of 98.801%, kappa coefficient of 97.6%, the precision value of 98.81%, and recall value of 98.80%.