Research on electromagnetic attack of advanced encryption standard based on long short-term memory and sparse autoencoder
Deep learning techniques have been widely used in the field of Side Channel Attack (SCA), which poses a serious threat to the security of cryptographic algorithms. However, deep learning-based side channel attack also has problems such as inefficient models, poor robustness, and longtime consumption. To address these problems, this paper focuses on the performance of Long Short-term Memory(LSTM) combining with the dimensional compression technique of Sparse Auto Encoder (SAE), and validates it on fully synchronized and unsynchronized EM traces captured under first-order bool mask protection. The experimental results show that compared with multilayer perceptron (MLP) and convolutional neural network (CNN), LSTM achieves more than 90% training accuracy and test accuracy, with higher robustness, lower parameters and faster convergence speed, even when the jitter in the dataset increases from 0 to 50 and 100.
Paper
Full text
Research on electromagnetic attack of advanced encryption standard based on long short-term memory and sparse autoencoder
Semantic Scholar · Computer Science · 2022
Abstract
Deep learning techniques have been widely used in the field of Side Channel Attack (SCA), which poses a serious threat to the security of cryptographic algorithms. However, deep learning-based side channel attack also has problems such as inefficient models, poor robustness, and longtime consumption. To address these problems, this paper focuses on the performance of Long Short-term Memory(LSTM) combining with the dimensional compression technique of Sparse Auto Encoder (SAE), and validates it on fully synchronized and unsynchronized EM traces captured under first-order bool mask protection. The experimental results show that compared with multilayer perceptron (MLP) and convolutional neural network (CNN), LSTM achieves more than 90% training accuracy and test accuracy, with higher robustness, lower parameters and faster convergence speed, even when the jitter in the dataset increases from 0 to 50 and 100.