Support vector regression: exploiting machine learning techniques for leakage modeling

Abstract

Side-channel analysis (SCA) is a serious threat to embedded cryptography. Any SCA has two important components: leakage modeling and distinguisher. Although distinguisher has received much research efforts, leakage modeling still lies on couple of classical techniques like Hamming weight or linear regression. In this paper, we propose a novel support vector machine based technique for efficient leakage modeling. The technique is called support vector regression (SVR) and can be used in both profiled and non-profiled settings. We provide proper theoretical background of SVR with practical application on AES implementation running on an AVR microcontroller.