Electromagnetic failure analysis of control system processors in the internet of things

Abstract

Control system processors in networked control systems are susceptible to malware attacks and failures. Monitoring stuck-at faults and malware at a distance is more resilient and can be integrated into frameworks for the Internet of Things. We offer a technique for failure/performance analysis based on the electromagnetic spectral domain that is completely decoupled from the controller processor. Without being a part of the monitored system, it can investigate failure causes on a complicated six-stage pipelined microarchitecture. Due to the fact that malware cannot traverse EM side-channels, our monitor is immune to controller-level malware attacks and can also work without requiring controller alteration. The traces were analysed using Support Vector Machines, AdaBoost, Quadratic Discriminant Analysis, Gaussian Process Classifiers, and Naive Bayes in the frequency domain. On a six-stage pipelined ARM Cortex-M7, our results demonstrated an accuracy of over 80% in predicting control system stuck-at errors and various types of malware.