关键词: 侧信道攻击, 均匀流型逼近与投影, 相关功耗分析, 数据降维, 加权邻接图

Abstract: The efficiency of side channel attack and the accuracy of key recovery have been greatly reduced by the high noise and dimension of energy trace collected in side channel attacks. To solve these problems, a correlation power analysis method of Advanced Encryption Standard (AES) algorithm based on Uniform Manifold Approximation and Projection (UMAP) was proposed. The UMAP method was used to calculate the set of neighboring points of traces based on Euclidean distance. First, to capture the position relationships of the energy trace collected and preserve the local structural features of the traces, a weighted adjacency matrix was proposed by constructing an adjacency graph and calculating the similarity between neighboring nodes. Then, the structure relationship of the adjacency graph was described using the Laplacian matrix, and the Initialized low-dimensional data was represented as the eigenvectors corresponding to smaller eigenvalues extracted from the adjacency graph by feature decomposition. Meanwhile, the objective function represented by the binary cross-entropy method was used to adjust the position of the data in the low-dimensional space, to preserve the global structural features of the data. Furthermore, to further improve the computational efficiency, the force-oriented graph layout algorithm is introduced in the gradient descent process. Finally, power consumption attacks were performed on the reduced dimensional data to recover the key. The experimental results show that the UMAP method can effectively preserve the local and global structural features of the original energy trace data. The correlation power analysis based this UMAP method can improve the correlation between energy trace data and assumed power leakage models, and reduce the number of energy traces required for key recovery. The number of energy traces required to recover a single key byte is 180, and the number of energy traces required to recover all 16 key bytes is 700. Compared to the Isometric feature mapping (Isomap) dimensionality reduction method, the number of traces required to recover all key bytes with the UMAP method is reduced by 36.4%.

Key words: Keywords: side channel attack, uniform manifold approximation and projection, correlation power analysis, data dimensionality reduction, weighted adjacency graph