A deep learning based solution for data disproportionproblem in side channel attacks using intelligent sensors

Abstract

Recently, Deep learning (DL) based Side Channel Attacks (SCAs) has been emerged as new paradigm in which the cryptographic devices are attacked through the side channel information. SCAs use external characteristics like power consumption, electromagnetic radiation, sound etc. of cryptographic devices to attack and estimate the secret key. However, the accomplishment of Deep learning for SCAs has not been fully analyzed especially at the data used to train and test. The major problem observed for DL based SCAs are Data Disproportionation Problem (DDP) using Intelligent Sensors which results in low success rate. Methods like data augmentation are used to make the data proportionate, but they resulted in poor accuracy because the original data will get disturbed. Hence, this paper proposed an ew solution to solve DDP without affecting the original data distribution. Unlike the traditional methods which predict the secret key based on Hamming Weight based likelihood function, the proposed solution uses Key value based likelihood function. We explore the validity of proposed solution through extensive simulations over the standard and public ASCAD dataset. The obtained results prove the superiority of proposed solution from the state-of-the-art methods.