hylock: hybrid logic locking based on structural fuzzing for resisting learning-based attacks

The growing popularity of the fabless manufacturing model and the resulting threats have increased the importance of logic locking as a key-based method for intellectual property (ip) protection. recently, machine learning (ml)-based attacks have broken most existing locks by exploiting structural traces or undoing optimizations that obfuscate them. a common limitation of these attacks, however, is their reliance on the correlation between the locked circuit structure and the correct key value. in this paper, we introduce structural fuzzing as a simple, nondeterministic, non-optimizing heuristic algorithm that can obfuscate the lock against learning-based attacks, preventing the attacker from predicting the key. we proceed to apply structural fuzzing to multiplexer-based logic locking and propose hylock, a logic lock with improved resilience against learning-based attacks. in common benchmarks, when compared with a state of the art logic lock, there is on average a 17% decrease in the number of correctly predicted key bits.