Concepts de base
DECOR is a randomized algorithm-based method that significantly decreases the correlation between locked circuit netlist and correct key values, enhancing resilience against ML-based attacks.
Résumé
Logic locking (LL) has been a crucial measure for IC protection, but recent ML-based attacks have exposed vulnerabilities. DECOR introduces a generic LL enhancement method based on randomized alterations to the locked circuit function. By strategically modifying UDC cofactors, DECOR confuses ML models without affecting legal users' access. Experimental results show that DECOR-XBI and DECOR-SARLock reduce key prediction accuracy to around 50%, providing significant security improvements. The area overhead varies across benchmark circuits but remains acceptable for large systems. PCA analysis demonstrates the effectiveness of DECOR in decorrelating features from labels in training data sets.
Stats
Numerical results show that the proposed method can efficiently degrade the accuracy of state-of-the-art ML-based attacks down to around 50%, resulting in negligible advantage versus random guessing.
Previous efforts to robustify LL against ML-based attacks have mostly focused on making a set of localized, predetermined transformations of the circuit structure with the goal of reducing the correlation between the circuit structure and the correct key.
ML-based attacks consist of two phases, namely, training and inference. The training phase includes three steps: reference netlist generation, training data extraction, and model training.
Citations
"DECOR introduces a generic LL enhancement method based on randomized alterations to the locked circuit function."
"Experimental results show that DECOR-XBI and DECOR-SARLock reduce key prediction accuracy to around 50%."