Title:  Leveraging Electromagnetic Side-Channel for System Profiling

Committee:

Dr. Milos Prvulovic, CoC, Chair, Advisor

Dr. Alenka Zajic, ECE, Co-Advisor

Dr. Hyesoon Kim, ECE

Dr. Moinuddin Qureshi, CoC

Dr. Tushar Krishna, ECE

Dr. Alessandro Orso, CoC

Abstract: With growing demand for efficient Internet of Things and embedded devices, system profiling in such highly resource-constrained systems is a huge challenge. Traditional profilers have dependent upon heavily modifying the system to monitor system activity, and such approaches add either a lot of program instrumentation or rely on hardware-support from the device itself. These methodologies have been known for interfering with the native application events beyond recognition. It has been well-known that side-channels (unintentional leakages) from a device contain system activity information, and this information can in turn be used to achieve system profiling and monitoring. However, there has been insufficient research efforts to systematically correlate the application execution to the underlying architectural and micro-architectural activity of the system. This thesis attempts to address these challenges by creating and developing frameworks to extract and profile various performance-affecting activities by leveraging the physical side-channel, specifically the electromagnetic side-channel, of the devices in a completely contact-less manner. Specifically, the first objective of this thesis is to conceptualize a new profiler to extract memory access features and profile the memory subsystem completely externally. The second objective of this thesis is to utilize the electromagnetic emanations to model architectural events that are nearly impossible to profile using on-device software infrastructure. To demonstrate this, we profile asynchronously occurring events such as interrupts and exceptions, that IoT and embedded devices depend heavily upon for their correct functionality. Having developed a good understanding of the system activity using the aforementioned profilers, we present the third objective of this thesis to achieve application fingerprinting and monitoring using the identified signatures of system activity. As a proof-of-concept, we design a novel approach for identification, profiling and analysis of IoT devices' network operations in a completely remote manner, that exhibits very high accuracy in determining the underlying network protocol and providing more information about individual transactions within a protocol.