Title: Software Profiling via Electromagnetic Side-Channel Signal

Alireza Nazari

Ph.D. student in Computer Science

School of Computer Science

College of Computing

Georgia Institute of Technology

Date: Thursday, February 14, 2019

Time: 9:00 - 11:00 AM (EST)

Location: Klaus 2100

Committee:

Dr. Milos Prvulovic (Advisor), School of Computer Science, Georgia Institute of Technology

Dr. Alenka Zajic(Co-advisor), School of Computer Science, Georgia Institute of Technology

Dr. Alessandro Orso, School of Computer Science, Georgia Institute of Technology

Dr. Moinuddin Qureshi, School of Electrical and Computer Engineering, Georgia Institute of Technology

Dr. Tushar Krishna, School of Electrical and Computer Engineering, Georgia Institute of Technology

Abstract:

Electronic circuits within computers generate electromagnetic (EM) emanations as a consequence of changes in current flow within a computing device. It is well documented in the literature that EM emanations often contain some information about program activity in the system. Most research work on EM emanations has focused on the potential risk that they create as side-channels, and on countermeasures against such attacks. However, the wide range of information that can be gathered from such signals suggests that EM side-channel signal is potentially useful for far more purposes and its application should not be limited to hostile activities. Unfortunately, since the relation between side-channel signals and this information is conventionally perceived as "complex" few bodies of work have tried to explore these non-hostile applications.

This proposal presents my research on effectiveness of profiling applications via EM side-channel signals in a non-hostile environment. It explores various levels of information that can be extracted from it and their potential application. The main goal of this work is to explore generalized profiling methods via EM signal where method is not bound to specific application or hardware. First, I present Spectral Profiling, a new profiling approach that allows highly accurate profiling of loops and other repetitive activity, without perturbing the profiled system, the program it runs, or the characteristics of the execution, in any way. Second, I show how leveraging EM side-channel signals can be used to detect deviations in program execution. Lastly, I present EMProf, a memory profiler that is completely external to the profiled system, it does not change the behavior of the profiled memory subsystem, and requires no hardware support, no memory, counter or other resources, and no instrumentation on the profiled system.