Improved Hybrid Approach for Side-Channel Analysis Using Efficient Convolutional Neural Network and Dimensionality Reduction

摘要: Deep learning-based side channel attacks are burgeoning due to their better efficiency and performance, suppressing the traditional side-channel analysis. To launch successful attack on a particular public key cryptographic (PKC) algorithm, large number of samples per trace might need be acquired capture all minor useful details from leakage information, which increases features instance. The decreased instance-feature ratio computational complexity deep attacks, limiting efficiency. Moreover, data class imbalance can hindrance in accurate model training, leading an accuracy paradox. We propose efficient Convolutional Neural Network (CNN) based approach dimensionality dataset is reduced, then processed using proposed CNN model. In model, optimal convolutional blocks used build powerful extractors within cost limit. have also analyzed presented impact Synthetic Minority Over-sampling Technique (SMOTE) performance. that data-balancing step should mandatory for analysis scenario. provided performance-based comparative between existing learning models unprotected protected Elliptic curve (ECC) Montgomery Power ladder implementations. reduced network complexity, together with improved efficiency, promote effectively attacks.