Title:  Leakage Extraction Methodologies from On-Device and Remote Side-Channels

Committee: 

Dr. Mukhopahyay, Advisor     

Dr. Yu, Chair

Dr. Chatterjee

Abstract: The objective of the proposed research is to develop signal processing and machine learning (ML) techniques to extract information leakage from external power, electromagnetic emissions, and on-device power side-channels. We demonstrate the following side-channel attack and defense methodologies. An improved Correlation Frequency Analysis (CFA) attack with narrow band pass filtering and windowed FFT successfully recovers encryption key on unrolled implementations of lightweight block cipher, SIMON. An EM side-channel based defense using spectral features and ML based detection to identify malware applications from benign and detect unknown applications. A ML malware analysis framework trained on features derived from wavelet coefficients of EM side-channel traces to group detected malware into families. Apart from traditional side-channels, proposed research shows dynamic power managers in software that governs voltage, frequency states of CPU, generates unique signature for different application, and can red-flag malware applications from benign. To demonstrate unique correlation, ML models are trained on application specific features derived from DVFS states time-series. The application inference is further extended to demonstrate feasibility of a malware detector for IoT devices. The extension to proposed work would demonstrate a malware detection and analysis framework utilizing multi-channel frequency scaling data from CPU and other devices to improve security of devices through dynamic power management and study energy security trade-offs.