LIPSTICK: Corruptibility-Aware and Explainable Graph Neural Network-based Oracle-Less Attack on Logic Locking

TL;DR

The paper addresses vulnerabilities in logic-locked circuits under a zero-trust fabless model by moving beyond HD-based ML attacks and proposing LIPSTICK, a corruptibility-aware and explainable GNN-based oracle-less attack that accounts for both circuit structure and functionality. It formalizes metrics such as the key error rate $ER(K)$ and challenges the sufficiency of $HD(K^a,K^*)$ as a predictor, integrating ER into training and evaluation. LIPSTICK combines a GNN on circuit graphs with resynthesized variants and multiple locking methods, using PGExplainer to provide explanations of the inferred keys. Empirical results on ISCAS'85 benchmarks show improved prediction accuracy and key precision over prior OL attacks, accompanied by interpretable explanations that can guide secure design choices.

Abstract

In a zero-trust fabless paradigm, designers are increasingly concerned about hardware-based attacks on the semiconductor supply chain. Logic locking is a design-for-trust method that adds extra key-controlled gates in the circuits to prevent hardware intellectual property theft and overproduction. While attackers have traditionally relied on an oracle to attack logic-locked circuits, machine learning attacks have shown the ability to retrieve the secret key even without access to an oracle. In this paper, we first examine the limitations of state-of-the-art machine learning attacks and argue that the use of key hamming distance as the sole model-guiding structural metric is not always useful. Then, we develop, train, and test a corruptibility-aware graph neural network-based oracle-less attack on logic locking that takes into consideration both the structure and the behavior of the circuits. Our model is explainable in the sense that we analyze what the machine learning model has interpreted in the training process and how it can perform a successful attack. Chip designers may find this information beneficial in securing their designs while avoiding incremental fixes.

Figures (4)

• Figure 1: Counterexample for hamming distance as a key precision metric
• Figure 2: Counterexample for the model prediction accuracy of state-of-the-art GNN-based attacks
• Figure 3: LIPSTICK attack framework
• Figure 4: Explanation accuracy of LIPSTICK on locking schemes. Bold edges demonstrate a sample pattern that PGExplainer Zhang:PGExplainer was able to find. (a) xor-based locking (b) mux-based locking (c) xor-based and mux-based methods (d) xor-based, mux-based, lut-based, and SAR-Lock methods