An FPGA-Based Intrusion Detection System Using Binarised Neural Network for CAN Bus Systems

As vehicles are made more intelligent with numerous installed sensors, the in-vehicle networks become more prone to critical security issues. The absence of any authentication process for Controller Area Network (CAN), which is the most widely used communication protocol among the Electronic Control Units (ECUs), makes the network vulnerable to attacks. By connecting some malicious nodes to the vehicle CAN bus, the intruders can capture and manipulate the vehicle data. Intrusion Detection Systems (IDSs) utilising machine learning models have shown outstanding performance in detecting incoming attacks, and even better, they do not reduce the communication speed when compared to adding the authentication mechanism into the CAN protocol. This paper presents an FPGA-based IDS using Binarised Neural Network (BNN) for CAN bus systems. The developed 1-bit BNN model makes the implementation fit in a low-cost FPGA device. Moreover, to maximise efficiency, the model is structured in a novel two-stage cascaded architecture called Coarse-to-Fine (C2F) model. Using this architecture, a Coarse model is first processed to detect whether any attack has taken place. Then, only when any attack occurrence is identified, a Fine model is executed for attack-type detection. Results show that our approach achieves more than 99% in every evaluation metric for attack detection and classification. Furthermore, compared to the state-of-the-art FPGA-based IDS, the proposed IDS reduces power consumption by 40% with a comparable speed.