Hardware Trojan Detection using Unsupervised Machine Learning Algorithms in the Gate-level Netlist

Now-a-days, the density of gates in VLSI circuits is rapidly increasing, it is becoming more difficult to detect malicious circuits. The malicious nets present in the circuit, named "hardware trojans," can be added in the pre-silicon and post-silicon design stages. In order to detect and classify circuits with and without trojans, efficient methods are required. Here we proposed an unsupervised machine learning model to detect and classify genuine and suspicious signals in the gate-level netlist using the controllability and observability Trojan detection (COTD) technique. This COTD method clusters the controllability and observability of each net of different ISCAS-85 and ISCAS-89 benchmark circuits in a single round and identifies suspicious or hardware trojan signals using K-means clustering and Densitybased clustering algorithms with characteristics that resembled those of HT-free signals. As a result, these signals were missclustered, and high accuracy and low FPR have been recorded. Experimental results on ISCAS-85 and ISCAS-89 benchmark circuits show that 98and 0.8871 silhouette score for density-based clustering algorithm and 96silhouette score for K-means clustering algorithm.