EMSim+: Accelerating Electromagnetic Security Evaluation with Generative Adversarial Network
Ya Gao, Haocheng Ma, Jindi Kong, Jiaji He, Yiqiang Zhao, Yier Jin · 2023
Electromagnetic side-channel analysis (EM SCA) attack is a serious threat to integrated circuits (ICs). In order to detect vulnerabilities in time at the pre-silicon stage and to improve the chip's robustness to EM SCA attacks, several EM simulation methods have emerged for EM side-channel leakage evaluation. Although the simulated results are accurate, the chip security evaluation in practice requires up to hundreds of millions simulation traces, which imposes an unrealistic computational and time overhead on these simulator-based methods. In this paper, we develop a tool named EMSim+. Different from the general EM security evaluation process, EMSim+ introduces machine learning (ML) to accelerate the simulation of layout-level EM emanations. Based on the generative adversarial network (GAN), a well-trained EMSim+ model can accept the cell current and power grid information of the chip and rapidly predict the EM emanation of the chip surface. We apply EMSim+ to a series of representative cryptographic circuits and compare the simulation results with the state-of-the-art EM simulation method and silicon measurements. The experimental results prove that EMSim+ has high simulation accuracy and achieves more than 242 times evaluation time reduction for 1 M sample data.