A Novel Approach to Detect Hardware Malware Using Hamming Weight Model and One Class Support Vector Machine

Abstract

Internet of Things (IoT) is an upcoming research area in cyber security and since common security measures are not implemented in these devices, they are most vulnerable to cyber attacks. Though IoT devices are vulnerable to both hardware and software malware attacks, the impact of hardware vulnerabilities will be significant since the devices once fabricated cannot be modified or updated. Among the major hardware vulnerabilities, hardware malware also known as hardware trojan (HT) is critical as it can control, modify, disable or monitor the key information in the device. While many techniques have been explored in the literature to detect HT at the gate level netlist, their computational complexity is very high and detection accuracy is relatively low. In order to reduce the computational complexity and improve the detection accuracy, a novel approach is proposed in this work to detect HT in the gate level netlist using Hamming weight model and unsupervised anomaly based detection method. The leakage power supply current signatures for both HT-free and HT-infected circuits are derived from Hamming weights of the random input sequence. The current signatures are then normalized and applied to one class support vector machine which acts as an anomaly detector to identify the HT-infected circuits. The process parameter and environmental noise variations are considered while characterizing the gates for different Hamming weights. The proposed method is evaluated on ISCAS85 C17 benchmark circuit using 16 nm process technology node in HSpice. A detection accuracy of 100% is achieved even when there is a single malicious gate in the HT-infected circuit. Experiments with other benchmark circuits show that the proposed methodology performs well.