SecCAN : Secure CAN Controller with Integrated IDS

This repository contains the hardware implementation of a Secure CAN Controller with an Integrated Intrusion Detection System (IDS). The design features the ability to flag suspicious messages for the receiving ECU, enhancing the security of the Controller Area Network (CAN).

Key Features

• Extension of Open-Source CAN Controller:
  Built upon an existing Verilog-based open-source CAN controller implementation.
• Integrated IDS:
  • The IDS is embedded within the BSP module of the controller.
  • It appends a flag bit to each message for real-time attack detection.
• Advanced Threat Detection:
  • A 4-bit quantized Multi-Layer Perceptron (MLP) serves as the IDS, operating as a binary classifier.
  • Trained to detect DoS/Flooding, Fuzzing, and Malfunction attacks.
  • Utilizes datasets like the Open Car Hacking Dataset and Survival Analysis Dataset for robust training.

Technical Highlights

• Hardware IP for Quantized IDS:
  Developed using the open-source FINN toolchain from AMD Research Labs.
• Training with Quantization-Aware Tools:
  The IDS model is trained using Brevitas, an open-source quantization-aware training library from AMD Research Labs.

This project demonstrates a novel approach to enhancing the security of CAN systems by integrating machine learning-based intrusion detection into the hardware.