文献类型：期刊文献

英文题名：Scarefusion:Side Channel Analysis Data Restoration with Diffusion Model

作者：Lu, Zeng[1,2]; Longde, Yan[1,2]; Fei, Wan[1,2]; Ning, Yang[1,2]; Xiang, Li[1,2]; Aidong, Chen[1,2,3]; Jiancheng, Zhang[1,2]; YanLong, Zhang[4]; Shuo, Wang[4]; Jing, Zhou[4]

第一作者：Lu, Zeng

机构：[1] Beijing Key Laboratory of Information Service Engineering, Beijing Union University, Beijing, 100101, China; [2] College of Robotics, Beijing Union University, Beijing, 100101, China; [3] Research Centre for Multi-Intelligent Systems, Beijing, 100101, China; [4] Beijing Union University, China

第一机构：北京联合大学北京市信息服务工程重点实验室

年份：2024

外文期刊名：SSRN

收录：EI(收录号：20240262520)

语种：英文

外文关键词：Deep learning - Diffusion - Quality control - Side channel attack - Signal sampling

摘要：Side-channel analysis based on deep learning is sensitive to data quality. We propose SCARefusion, a latent diffusion model using U-Net, to address data imbalance in power consumption traces. SCARefusion retains high-resolution information through Conditional Nonlinear Activation Free Blocks (CNAFBlocks), downsampling, and upsampling modules. This model generates balanced class data, mitigating dataset imbalance and avoiding GAN instability. Using Correlation Power Analysis (CPA), we compare correlation coefficients between hypothetical and measured traces. The generated traces' effectiveness is validated using CNN attacks. Experimental results on the ASCADf dataset show that SCARefusion generates traces consistent with original leakage points and successfully extracts the correct key. We also discuss strategies to reduce learning rate and optimize network architecture to address loss oscillation, enhancing the performance and robustness of deep learning in side-channel analysis. ? 2024, The Authors. All rights reserved.