Date of Graduation

5-2024

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Engineering (PhD)

Degree Level

Graduate

Department

Computer Science & Computer Engineering

Advisor/Mentor

Di, Jia

Committee Member

Panda, Brajendra N.

Second Committee Member

Nelson, Alexander H.

Third Committee Member

Dix, Jeff

Keywords

Computer Security; Graph Neural Networks; Hardware Security; Hardware Trojan; Machine Learning; Trojan Detection

Abstract

The integrated circuit (IC) industry has experienced exponential growth, particularly in the complexity and scale of hardware designs. To sustain this growth, faster development cycles and cost-effective solutions have been the focus for many companies. One strategy to maintain this growth is through the incorporation of third-party intellectual property (IP) into the IC design process. Outsourcing the production of sub-components reduces development time and enables faster time-to-market, however, this approach also introduces the threat of Hardware Trojans. Hardware Trojans, defined as any malicious modification or addition to an IC, pose significant security risks due to their small size, low activation frequency, and complex obfuscation techniques. Subsequently, efficient and cost-effective detection of Hardware Trojans has become a crucial area of research. This dissertation research proposes an improvement to the Trojan detection mechanisms incorporated in the Structural Checking Tool, a Trojan detection tool that focuses on the identification of logical Trojans embedded within soft IPs. This tool facilitates direct conversion of HDL source code into a structural representation using directed graphs. Leveraging these graph structures and signal-level features, this research develops a new dataset and three graph neural networks architectures. Each of the three neural networks correspond to classical graph neural network layers and execute graph-level probabilistic binary classification of Trojan inclusion. Through rigorous testing with two potential sets of node-level feature vectors, this dissertation offers a faster, more accurate, and more adaptable approach than those existing within the current tool.

Download

Share

COinS