HT-EMIS: A Deep Learning Tool for Hardware Trojan Detection and Identification through Runtime EM Side-Channels

Hardware Trojans (HTs) are malicious circuits planted in Integrated Circuits (ICs). Multiple techniques using Side-Channel signals to detect HTs have been developed over the past decade. However, most of this research focuses on HT detection. Few of them explore the possibility of either identifying different Hardware Trojans implemented inside ICs or detecting inactive HTs. We propose a runtime EM side-channel analysis workflow (HT-EMIS) that uses a convolutional neural network to address the shortcomings above. By analyzing EM side-channel leakage from an FPGA, our tool can identify known types of HTs implemented inside a design and reports whether they are inactive or active with 100% accuracy. Additionally, we are able to successfully detect new unseen HTs with this model in 98.7% of test cases, due to the fact that HTs inserted at the Register Transfer Level with similar triggers and payloads often have similar effects on a floorplan, and thus the EM radiation of a device.