Wire Load Variation-Based Hardware Trojan Detection Using Machine Learning Techniques

Abstract

Detection of malicious form of hardware is commonly referred to as hardware Trojan and had become a major challenge. Especially when we go down to deep submicron technology, it becomes really difficult to detect the presence of Trojan using conventional testing approaches. Logic testing proves to be effective only when the number of inputs to trigger the Trojan is of small number. Further using side-channel approaches, the complexity of triggering the entire Trojan circuit is reduced because partial activation of Trojan will cause a considerable change in the measured parameter that is used to differentiate between the original circuit and circuit infected with Trojan. This work provides a non-invasive hardware Trojan detection methodology which uses side-channel power of the circuit to detect the presence of the Trojan. Moreover, the proposed method deviates from other existing techniques by accounting the change in side-channel power, due to interconnects that is the distributed resistance and capacitance of the wire connecting different standard cells in the circuit, by taking wire load variations into consideration. The power profile from Trojan-infected and Trojan-free circuits is used for training the machine. The machine predicts with much accuracy whether the circuit consists of Trojan or not based on data we have trained, therefore effectively categorizing the circuits and thus eliminating the errors caused due to manual intervention. The proposed work is validated by using ISCAS 85 and ISCAS 89 benchmark circuits.