Portability profiled power analysis based on deep domain adaptation

Abstract

Abstract The security of password information is essential to protect personal privacy and sensitive data. In recent years, deep learning-based side-channel profiled analysis has been more in-depth research. However, the differences between devices caused by circuit design, process parameters and other hardware characteristics have not been well solved, and the generalization ability of pre-trained models on new devices is poor, and the prediction performance is low. Therefore, this paper proposes a new method, Portability Profiled Power Analysis (PPPA) strategy, which combines domain adaptive and deep learning technology to model and calibrate the domain difference between the profiling device and the target device. Improve model adaptability in different device environments. Experiments show that the proposed method can recover the key information of different devices and effectively reduce the influence of domain differences caused by different devices.