Side channels are created unintentionally as a byproduct of computer system implementations, and they can potentially leak confidential and sensitive information. By using the consumed power, temperature changes, acoustic emanations, and electromagnetic (EM) emanations, several side-channel attacks have been reported in the literature. Side channels are also used in non-adversarial applications, including malware detection, program profiling, reverse engineering, etc. However, these approaches have the following drawbacks: 1) they are coarse-grained and cannot detect small changes; 2) they are limited to simple computers with low operating clock frequencies; and 3) they do not consider the microarchitecture-level features of the computers, leading to inaccurate modeling and tracking.

To address these challenges, this thesis describes a systematic framework for analyzing and modeling software activities of modern computing devices using the electromagnetic side channel. This framework integrates various methodologies and techniques to address the limitations of current approaches at finer granularity, particularly on modern computers with pipelined processors and high clock frequencies. Specifically, the first contribution of this thesis is a technique that increases the effective sampling rate and signal-to-noise ratio (SNR) of the signals collected with readily available measurement devices. The second contribution is PITEM (Permutations-Based Instruction Tracking via Electromagnetic Side-Channel), a mechanism to 1) identify groups of instructions that have similar EM signatures, and 2) track all possible orderings, i.e., permutations, of these instruction groups. The final contribution is machine-learning-based modeling of the electromagnetic side-channel signals by indirectly addressing several microarchitecture-level properties.

This thesis provides other researchers with tools and insights that can be used to leverage the electromagnetic side channel in various applications.