Date: Wednesday, Aug. 10, 2022
Time: 11:30 am (EDT)
Location: Hybrid at this zoom.gatech.us link and TSRB Room 423
Erik Jorgensen
Machine Learning PhD Candidate
Electrical and Computer Engineering
Georgia Institute of Technology
Committee:
- Dr. Alenka Zajić, School of Electrical and Computer Engineering, Georgia Tech (Advisor)
- Dr. Matthieu Bloch, School of Electrical and Computer Engineering, Georgia Tech (Co-Advisor)
- Dr. David Anderson, School of Electrical and Computer Engineering, Georgia Tech
- Dr. Mark Davenport, School of Electrical and Computer Engineering, Georgia Tech
- Dr. Milos Prvulovic, School of Computer Science, Georgia Tech
- Dr. Mikko Lipasti, Department of Electrical and Computer Engineering, University of Wisconsin-Madison
ABSTRACT
The authenticity of integrated circuits is of increasing concern as more steps in the device manufacturing supply chain are outsourced, especially considering current global semiconductor shortages. Common methods for integrated circuit validation rely on either destructive techniques with high resolution imaging of the circuit interconnects or functional testing of a variety of test inputs with automated test equipment. These methods are time-consuming or even intractable to detect counterfeit components or stealthy modifications of underlying circuitry. The objective of the proposed research is to combine the non-destructive monitoring advantages of standard and backscattering electromagnetic side channels with modern machine learning techniques to efficiently validate the authenticity of individual integrated circuits installed on a motherboard.
First, we apply deep learning methods to classify and detect counterfeits of major ICs on a variety of motherboards. Second, we leverage hyperspectral scanning with the backscattered EM side-channel and design a novel active learning method to detect dormant hardware trojans several times smaller than previously possible. Finally, we develop a compressed sensing approach to heavily reduce sampling for hardware trojan detection while defining a hyperspectral characterization of expected and anomalous circuits.