PFS - A Platform for Structural Analysis of Logic Locking Using Machine Learning

Hardware obfuscation or logic locking (LL), relate to a method of safeguarding hardware intellectual property (IP) blocks against a variety of attacks, such as, IP theft, reverse engineering, and malicious alterations. Modern locking strategies primarily focus on preventing illegal use of a design by limiting correct functionality; they rarely address preventing extraction of design secrets through structural analysis. The locking gates are also vulnerable to removal attempts due to a lack of structural obfuscation. In this work, we analyze the structural alterations caused by LL and introduce a new class of attacks (SAIL, SURF) and associated CAD platform that take advantage of these artifacts. SAIL can expose the structural vulnerabilities of a locked design through a systematic set of steps and open it up for subsequent key-retrieval/reverse-engineering attacks. SAIL is far stronger than prevalent attacks such as SAT-based attacks because SAIL does not require an unlocked-IC or golden I/O responses to carry out the attack. We have also proposed the SIVA-Metric which can be used to quantify the existing structural vulnerabilities of a given design. It can be directly computed without additional training requirements. The SURF attack leverages the structural analysis output of SAIL to mount a subsequent key-retrieval attack. Through extensive quantitative analysis, we have demonstrated the efficacy of the proposed attacks/metrics.