Random Forest Classifier For Hardware Trojan Detection

Hardware Trojans, which exist in the integrated circuits in the form of additional logic units, will affect the supply voltage of nearby units. In this paper, the ring oscillator is introduced to diagnose the state of the chip for its high sensitivity to the fluctuation of supply voltage. Combining the ring oscillator network, and the machine learning method, a random forest classifier is presented to detect hardware Trojans. The corresponding chips were designed and fabricated. The sample data, obtained from chips, were used to train the random forest classifier. The data can be processed by principal component analysis to optimize the random forest classifier. Then the accuracy of the random forest classifier is improved. The experimental results show that the accuracy of this method is 100% for the UART-T100 chip with a Trojans ratio of 3.22% and the UART-T800 chip with a Trojan ratio of 1.67%.