SeeMLess: Security Evaluation of Logic Locking using Machine Learning oriented Estimation
Bulbul Ahmed, M. Sazadur Rahman, Kimia Zamiri Azar, Farimah Farahmandi, Fahim Rahman, Mark Mohammad Tehranipoor · 2024
Although logic locking has been widely known as a promising countermeasure against intellectual property (IP) piracy and overproduction risks, it has been challenged by different attack breeds over the years. Attacks on logic locking, either algorithmic or structural, have been always known as a time-consuming resource-intensive effort. For instance, the Boolean satisfiability (SAT) attack might take weeks to be completed. In this paper, we introduce SeeMLess, a first-of-its-kind ML framework for the security evaluation of logic locking, design and locking agnostic. SeeMLess leverages feature sets computed from different aspects, graph-based, functional, propositional, etc. to accurately estimate the attack time with no attacks running. Our experimental results, on a case study over the SAT attack, show the trained model on a dataset of 5K+ designs locked by various techniques, where SeeMLess achieves Math 1 accuracy in predicting the time of the attack, offering valuable insights into the locking mechanism effectiveness pre-implementation.