GoCLIP: Graph one-class CLassification for Intellectual Property Circuit Identification

Circuit Intellectual Property (IP) Identification is crucial to prevent malicious misuse of IP cores/blocks in Integrated Circuits (ICs). The main objective in IP identification is to identify a known set of IP cores/blocks in the circuit netlists extracted from manufactured IC products. Conventional methods are usually based on similarity matching to a library of IPs. Such methods are tedious and their performance can be compromised by small changes to the recovered circuit netlists caused by imperfections in the recovery process. In this paper, we propose GoCLIP, a machine learning based graph one-class classification method for IP identification. We reformulate circuit IP identification into a graph one-class classification problem and employ Graph Neural Network (GNN) for graph structure learning. Our results show that our machine learning models produce highly differentiable graph features for different circuits with promising results in IP identification of mean Average Precision (mAP) >95%.