SimLL: Similarity-Based Logic Locking against Machine Learning Attacks
DescriptionLogic locking is a promising technique for protecting outsourced integrated circuit designs. Recently, graph neural network (GNN)-based link prediction attacks have been developed that successfully break all multiplexer-based learning-resilient locking techniques. We present SimLL, a novel similarity-based locking technique that locks a design using multiplexers and is robust against all existing structure-exploiting oracle-less learning-based attacks. SimLL introduces key-controlled multiplexers between topologically and functionally similar logic gates/wires in a design. Choosing similar nodes/wires successfully confuses the ML models. Numerical results show that SimLL can degrade the accuracy of the existing ML-based attacks to ~50%, resulting in a negligible advantage over random guessing.
TimeThursday, July 13th4:55pm - 5:10pm PDT
Location3003, 3rd Floor
Hardware Security: Attack and Defense