Hardware Trojan Detection in Chips by Removing Dependencies Between Features in Machine Learning

Globally, there has been an increase in demand for System on Chip (SoC) applications, active medical implants, and Internet of Things (IoT) devices. However, due to challenges in the global supply chain, the design, fabrication, and testing of Integrated Circuits are often outsourced to untrusted third-party entities around the world rather than a single trusted entity. This situation presents an opportunity for adversaries to compromise the device\u27s integrity, performance, and functionality by inserting malicious modifications known as Hardware Trojans (HTs) into the original design. HTs can also create a backdoor in the system for malicious alterations. In this research, a solution to the issue of hardware trojan is presented through the utilization of machine learning models that rely on supervised and unsupervised learning. The proposed method involves providing the netlist features of the digital hardware design post-synthesis to the machine learning model and removing any interdependence between features to prevent overfitting of the training dataset. The supervised model showed a 99.2\% true positive and true negative rate, as well as an F-measure of 99.3\%, while the unsupervised model achieved a 99.5\% true positive rate with the use of random projection, thereby offering a more resilient machine learning-based method for detecting hardware trojans