Challenging the Security of Logic Locking Schemes in the Era of Deep Learning: A Neuroevolutionary Approach

Abstract

Logic locking is a prominent technique to protect the integrity of hardware designs throughout the integrated circuit design and fabrication flow. However, in recent years, the security of locking schemes has been thoroughly challenged by the introduction of various deobfuscation attacks. As in most research branches, deep learning is being introduced in the domain of logic locking as well. Therefore, in this article we present SnapShot, a novel attack on logic locking that is the first of its kind to utilize artificial neural networks to directly predict a key bit value from a locked synthesized gate-level netlist without using a golden reference. Hereby, the attack uses a simpler yet more flexible learning model compared to existing work. Two different approaches are evaluated. The first approach is based on a simple feedforward fully connected neural network. The second approach utilizes genetic algorithms to evolve more complex convolutional neural network architectures specialized for the given task. The attack flow offers a generic and customizable framework for attacking locking schemes using machine learning techniques. We perform an extensive evaluation of SnapShot for two realistic attack scenarios, comprising both reference combinational and sequential benchmark circuits as well as silicon-proven RISC-V core modules. The evaluation results show that SnapShot achieves an average key prediction accuracy of 82.60% for the selected attack scenario, with a significant performance increase of 10.49 percentage points compared to the state of the art. Moreover, SnapShot outperforms the existing technique on all evaluated benchmarks. The results indicate that the security foundation of common logic locking schemes is built on questionable assumptions. Based on the lessons learned, we discuss the vulnerabilities and potentials of logic locking uncovered by SnapShot. The conclusions offer insights into the challenges of designing future logic locking schemes that are resilient to machine learning attacks.

Designing ML-Resilient Locking at Register-Transfer Level

“…The extractor uses the Pyverilog library [21]. Instead of one neural network type as in [6], we use auto-sklearn [13], a library for automatic ML (auto-ml) model exploration. Auto-ml searches for an ML model among the implementations and optimizes the hyperparameters.…”

“…Since ML-driven attacks do not need a working chip (oracle) to succeed [6], our threat model includes the following assumptions.…”

“…An ML-based attack has to exploit structural key-related patterns to produce a (correct) key prediction. Thus, we selected the OL SnapShot attack [6] as a basis for the evaluation. SnapShot was initially designed to attack locked gate-level netlists, following four major steps (Fig.…”

“…The training set comprises locking samples that are added in additional relocking rounds of the target with known keys. This process is also known as self-referencing [6,18]. The attacker uses the training set to collect observations about the relation between the locking pairs and the key.…”

“…ML-driven attacks exploit the predictable relation between the key value and the functional or structural aspects of locking. This has led to potent attacks that can either predict the correct key value or remove the locking circuitry from the netlist [6,11,12,18]. While ML-driven attacks often lack output certainty, their applicability adds another requirement for logic locking success-prevention of key-related residue within the locking mechanics.…”