Non-Profiled Side-Channel Attack Based on Deep Learning Using Picture Trace

Abstract

Over the years, deep learning algorithms have advanced a lot and any innovation in the algorithms are demonstrated and benchmarked for image classification. Several other field including sidechannel analysis (SCA) have recently adopted deep learning with great success. In SCA, the deep learning algorithms are typically working with 1-dimensional (1-D) data. In this work, we propose a unique method to improve deep learning based side-channel analysis by converting the measurements from raw-trace of 1-dimension data based on float or byte data into picture-formatted trace that has information based on the data position. We demonstrate why "Picturization" is more suitable for deep learning and compare how input and hidden layers interact for each raw (1-D) and picture form. As one potential application, we use a Binarized Neural Network (BNN) learning method that relies on a BNN's natural properties to improve variables. In addition, we propose a novel criterion for attack success or failure based on statistical confidence level rather than determination of a correct key using a ranking system. INDEX TERMSBinarized neural network, Deep learning, Multi-layer perceptron, Non-profiled sidechannel attack

Deep-Learning-Based Side-Channel Analysis of Block Cipher PIPO With Bitslice Implementation

“…In addition, according to [13], if the ID of the S-Box output is used as the classification label, then the learning levels of all VOLUME 10, 2022 guessed keys' neural networks are Thus, ID labeling is not used in DL-based non-profiled SCA. Later studies also performed DL-based non-profiled SCA using MSB or LSB labeling on the LUT implementation of AES [14]- [17]. In [18], Xiangliang et al performed DL-based non-profiled SCA on SM4 and DES using the MSB, HW labeling, and they got the result that the MSB labeling has better performance than HW labeling.…”

“…DL-based non-profiled SCA was proposed by Timon in 2019, and analysis was performed on the AES ASCAD dataset using the most significant bit (MSB), least significant bit (LSB), and HW labeling of the S-Box output [13]. In addition, in other studies, DL-based non-profiled SCA was also performed using the MSB, LSB, or HW label of the target block cipher's S-Box output [14]- [18]. Unlike traditional non-profiled SCA, in DL-based non-profiled SCA, bit models such as MSB and LSB exhibited better performance than HW models.…”

Side-channel Attack Using Word Embedding and Long Short Term Memories

“…The dataset has been divided into two datasets: DPAV2 Template Dataset.h5 and DPAV2 Public Dataset.h5. The first dataset consists of: plaintext(1000000, 16), first round key (16,1), last round key (16,1), ciphers (1000000,16), and traces(1000000,3253).…”

“…plaintext (20000,16), first round key (16,1), last round key (16,1), ciphers (20000,16), and traces (20000,3253), where the (. .…”

“…Moreover, they firstly utilized classical data augmentation tool to improve attack performance. In 2021 year, lots of algorithms demonstrated or benchmarked for image classification which are typically working with 1-dimensional (1-D) data, Yoo-Seung Won proposed a novel criterion for attack success or failure based on statistical confidence level rather than determination of a correct key using a ranking system and Binarized Neural Network (BNN) learning method that relies on a BNN's natural properties to improve variables [16]. J Wei [17], in 2020 year, firstly combined the LSTM from Hochreiter S [18] with side channel attack, then Paguada S in 2021 improve the efficiency of LSTM on side channel attack [19].…”