CHEST color logo
CHEST University logos image of a line

CHEST 2020 Research Project Abstracts


P13_20: Cognitive Obfuscation Securing Circuits by Graph Convolutional Networks
Topic Areas: Reverse Engineering, Logic Locking
Principal investigator: Dr. Houman Homayoun, University of California Davis
PI Email


Abstract:
The current heuristic based logic obfuscation methods are facing several challenges including: 1) Handcrafted obfuscation rules could be incomplete, biased, and even controversial. 2) Obfuscation strategy tailored for one attacking strategy usually does not work for another, while attackers can easily use various strategies, and 3) Hand-crafted heuristics and case-by-case obfuscation strategy are inefficient and expensive to cope with the fast-developing attacking strategies. To address these challenges, for the first time, we propose a framework for end-to-end obfuscation optimization which automatically estimates and maximizes the attackers’ efforts, and also provides explanations and new insights on the optimized obfuscation strategies. This project aims to develop the cognitive obfuscation that predicts the runtime of attacks based on the recently developed powerful method of graph convolutional network classifier. To design a defense against emerging SAT attack threats, it is non-trivial to first understand and characterize the properties of the design that makes it SAT-hard and post SAT attack-hard. We perform a design space exploration (DSE) in this project, to study the security impacts of different existing defenses against existing and developed security threats. This information will be utilized to build a graph convolutional network (GCN)-based deobfuscation time predictor to verify the security of an obfuscation instantaneously. Lastly, we deploy non-convex optimization approaches (including black-box stochastic optimizations and white-box gradient descent optimizations for deep learning) to determine the obfuscation parameters to secure the IC against existing SAT attacks as well as new powerful attacks.

Back