ResNet‐based Trojan detection methodology for protected ICs

Abstract

Hardware Trojan detection plays a significant role in building a trusted integrated circuit (IC) industry. Unfortunately, the conventional Trojan detection techniques may fail when the feature of the embedded Trojan is masked by the feature of the random side-channel leakages. To overcome the constraint, in this Letter, a novel machine learning technique based on residual neural networks (ResNet) is proposed to classify the different features to achieve the Trojan detection. In the proposed Trojan detection methodology, two different Trojan-free IC chips are used for training the ResNet to study the feature of the random side-channel leakages. Subsequently, a suspected IC chip is tested by the well-trained ResNet to infer whether the feature associated with the suspected IC chip is caused by the random side-channel leakages or the embedded Trojan. As demonstrated in the result, after enabling about 250,000 data to train the ResNet, the hardware Trojan can be detected by using the proposed methodology.