GNNUnlock+: A Systematic Methodology for Designing Graph Neural Networks-Based Oracle-Less Unlocking Schemes for Provably Secure Logic Locking

Leading-edge design houses outsource the fabrication process to pure-play foundries eliminating the expenses of owning and maintaining a fab. The intellectual property (IP) of an outsourced design is now subject to IP piracy, which drives the need for a protection mechanism. Logic locking is a technique that aims to thwart IP piracy throughout the supply chain. However, state-of-the-art, provably secure logic locking (PSLL) techniques are vulnerable to functional and structural analysis-based attacks. Few removal attack protection mechanisms have been developed, such as diversified tree logic and wire entanglement, to protect PSLL against structural attacks. In this work, we significantly enhance GNNUnlock (GNNUnlock+) and demonstrate how the most advanced PSLL techniques armed with removal attack protection have no impact on its effectiveness. Our evaluation demonstrates that GNNUnlock+ is 89.66%-100% successful in breaking benchmarks locked using 9 different PSLL techniques-Stripped functionality logic locking, tenacious and traceless logic locking, Anti-SAT, SAT attack resistant logic locking (SARLock), Anti-SAT with diversified tree logic (Anti-SAT-DTL), Anti-SAT with wire entanglement, SARLock-DTL, corrupt and correct (CAC) and CAC-DTL. GNNUnlock+ can break the considered techniques under different parameters, synthesis settings, and technology nodes. Moreover, GNNUnlock+ successfully breaks corner cases where even the most advanced state-of-the-art attacks fail.