A Metric-Learning Methodology for Side-Channel Analysis

Side-channel analysis has demonstrated its potency in extracting confidential keys by leveraging the flaws within the implementation of encryption algorithms. Over the past few years, sophisticated deep learning methods have been amalgamated into the profiling process of side-channel analysis, leading to exceptional outcomes. Yet, the complexity of the majority of contemporary studies is escalated due to the increased number of hyperparameters that require meticulous tuning. Although the traditional template attack method boasts a limited number of hyperparameters, its foundation in mathematical approaches renders it less adept at managing high-dimensional data. In response to these challenge, we introduce a side-channel analysis framework designated as MSCA, which leverages metric learning to guide the model in capturing the characteristics of leakage traces, and incorporates the principles of template attacks for side-channel analysis. Moreover, We combine the Lifted Structure Loss with label information of Side-channel leakage traces to enhance the model's performance. We implement the model to evaluate the MSCA framework with the publicly available ASCAD dataset. The experimental results on the ASCAD desynchronous dataset show that our method successfully recovers the key within 1146 traces and demonstrate the generality of our proposed framework.