SiameseHTD: Golden-Chip-Free Hardware Trojan Detection Using Power Traces
Abstract
Hardware Trojans (HTs) are considered one of the most significant threats in terms of security of integrated circuits (IC) by facilitating actions such as leakage of information, functional alteration, and cryptanalytic attacks. This research offers a novel golden-chip-free Hardware Trojan Detection (HTD) solution through the application of Siamese Neural Network (SNNs) based on power side-channel traces collected using the SAKURA-G FPGA platform. In particular, a four-stage preprocessing scheme is introduced to extract discriminative features from power traces. Unlike prior side-channel HTD methods that require a golden reference chip or rely on single-channel preprocessing, this work introduces a golden-chip-free, dual-channel preprocessing strategy that jointly preserves amplitude and switching-rate information. Moreover, three different Siamese models, namely, CNN, GRU, and LSTM, are implemented for evaluating 10,000 traces from the AES-T1300 benchmark set. It was found that the Siamese CNN model has the highest accuracy and ROC-AUC metrics, reaching 89.58% and 0.9519, respectively, compared with other models, namely, Siamese GRU and LSTM with accuracies of 89.17% and 72.92%, respectively.