Evaluation of Hardware-Trojan Detection by Ensemble Learning Model for Circuits Inserted with a Trojan by an Automated Framework
Sho Yoshimi, Yuka Ikegami, Ryotaro Negishi, Kota Hisafuru, Nozomu Togawa · 2025
During the design and manufacturing stages of IoT devices, there is a risk of Hardware Trojans (HTs) being inserted into circuits due to the intervention of outside companies. One method for effectively detecting HTs from gate-level netlists is to use an ensemble learning model. In this paper, we use four ensemble learning models: Random Forest, XGBoost, LightGBM, and CatBoost, and evaluate the accuracy of HT detection by adding a new Trojan circuit generated using an automatic HT generation framework as a netlist for training and evaluation. We also use SMOTE, ADASYN, and Borderline-SMOTE as oversampling methods used in training, and evaluate the HT detection accuracy when the hyperparameters of each method are optimized.