Committee:

Dr. Alenka Zajic, ECE, Chair , Advisor

Dr. Milos Prvulovic, CS, Co-Advisor

Dr. Morris Cohen, ECE

Dr. David Anderson, ECE

Dr. Azadeh Ansari, ECE

Dr. Alessandro Orso, CS

Abstract: Side-channels cause unintentional information leakage as a side-effect of hardware activity due to legitimate program execution. While attackers have traditionally exploited side-channel analysis for extracting sensitive information from target systems, recent research has utilized side-channels for non-adversarial monitoring of program execution. This approach can be especially useful for securing resource-constrained security-critical embedded systems. This thesis develops methods that leverage electromagnetic (EM) side-channel signals for non-adversarial and non-intrusive monitoring of embedded and cyber-physical systems. Our research provides techniques for identifying anomalous/malicious program behavior by detecting deviations in EM emanations and presents a framework for end-to-end basic-block program execution tracking by monitoring the device's EM side-channel signal. In this thesis, we 1) designed an intrusion detection system that learns a dictionary of reference EM signatures and exploits the dictionary for identifying anomalous/malicious program behavior, 2) designed neural network to model the monitored device's EM side-channel signal and detect stealthy malware activities through deviations in EM emanations, 3) designed a novel framework that performs basic-block program execution tracing by monitoring device's EM side-channel signal, and 4) demonstrated that even a single instruction deviation in program execution can be detected with high accuracy via EM side-channel signals captured by a readily available measurement device.