Abstract:We propose a new nonprofiled side-channel analysis (SCA) method called binary classification-based SCA (BCSCA), which accomplishes the tasks of the traditional nonprofiled SCA by solving binary classification problems with the help of neural networks. In addition, to improve the analytical efficiency, we design a new type of neural network and propose a fast implementation of BCSCA. We evaluate BCSCA on both simulated and public side-channel traces and compare BCSCA with correlation power analysis (CPA), mutual information analysis (MIA), and the state-of-the-art nonprofiled SCA method based on neural networks (NNSCA). The experimental results show that BCSCA outperforms NNSCA and MIA in all cases, and it is better than CPA on the nonlinear traces and comparable to CPA on the linear traces.

Binary Classification-Based Side-Channel Analysis.