Detection method of Golden Chip-Free Hardware Trojan based on the combination of ResNeXt structure and attention mechanism

Since 2007, the use of side-channel data to detect hardware Trojans (HT) has been widely studied. Machine learning methods are widely used in the detection of hardware Trojans, but with the development of integrated circuits (IC), machine learning methods are not able to obtain a higher accuracy rate compared to deep learning. In this paper, we propose to use an architecture inspired by the ResNeXt network architecture and combine it with an attention mechanism, referred to as the Attention-Res-Attention (ARA) network. Firstly, the side channel data are extracted by convolutional layer with features and focus on important points under the attention module; then, the feature map enters the ResNeXt architecture that achieves classification accuracy improvement by adding the attention module; finally, the data are classified by the fully connected layer. Our proposed solution is observable to natural variations that may occur in side-channel measurements, and can accurately detect abnormal behavior of the chip when HT is triggered. And using a self-referential method for HT detection eliminates the need for a golden chip. The effectiveness of the method proposed in this paper is evaluated based on the AES series hardware Trojans publicly provided by TrustHub. Experimental results show that the method proposed in this paper has a high accuracy rate when a single Trojan exists, and can effectively detect the existence of hardware Trojans. And when a variety of hardware Trojans exist at the same time, the method used in this paper can effectively distinguish the types of hardware Trojans, and the highest average accuracy rate reached 97% during the experiment. Compared with the existing deep learning methods, the network model in this paper has higher classification accuracy.