Application of LSTM Auto Encoder in Hardware Trojan Detection
Alex Sumarsono, Zachary Masters · 2023
Along with the demands for rapid development of integrated circuits (IC), the threats posed by malicious modifications to the original intent of the design, also known as hardware trojans (HT), have also increased significantly. HT can be inserted at certain weak points in the IC development flow by untrustworthy sources. Based on the activation characteristics, HT can be categorized as always-on or condition-based. The latter, especially, can be quite challenging to detect. In this paper, a novel HT detection technique using a Long Short Term Memory neural network coupled with an autoencoder (LSTM-AE) is proposed. The viability of LSTM-AE as an HT detection algorithm is demonstrated with always-on HT and condition-based HT that have been inserted into an up-down digital counter and a router verilog model, respectively. The condition-based HT is triggered after 18,000 packets are received. The experimental results show that LSTM-AE is capable of reliably detecting both types of HT.