四虎影院

In 2024, 60% of cybersecurity incidents in the automotive and smart mobility sectors affected thousands to millions of mobility assets. These assets are increasingly becoming under attack due to legacy in-vehicle networks, over-the-air software updates and vehicle-to-everything communication. While these capabilities are essential for autonomy, they also expose safety-critical control systems to remote cyber-attacks.

The core challenge is how to provide reliable, real-time intrusion detection on in-vehicle networks under the tight computational and energy constraints of Electronic Control Units (ECUs), while remaining effective against both known and novel attacks. Current state-of-the-art Intrusion Detection Systems (IDSs) for in-vehicle networks rely predominantly on conventional deep learning models, which cannot be effectively deployed on embedded ECUs, due to their high computational, memory, and energy demands.

There is an urgent need for highly parallel, low-latency processing with significantly reduced energy consumption. To this end, novel neural network paradigms, such as Neuromorphic computing based on Spiking Neural Networks (SNNs), offer fast event-driven detection at much lower energy cost. Neuromorphic methods for vehicle cyber-security remain largely unexplored.

To address the above challenges, this PhD will utilise neuromorphic, SNN-based IDSs tailored to next-generation autonomous vehicles. The project is novel in seeking to develop and experimentally validate an end-to-end neuromorphic IDS pipeline. It will explicitly address event-driven traffic representations, SNN architectures for anomaly detection and attack classification, and quantitative hardware-level evaluation.

The objectives are:

1. Define requirements for neuromorphic IDS, including representative in-vehicle network architectures, threat models, key attack vectors, and real-time performance and energy constraints, given the distributed nature of modern vehicle networks and the corresponding IDS they require.
2. Develop event-driven encodings and spiking neural network architectures for in-vehicle network traffic, mapping message identifiers, payloads and timing information into spike-based representations suitable for anomaly detection and attack classification.
3. Design and evaluate SNN-driven algorithms that produce hardware-efficient IDS, exploring both direct SNN training and systematic conversion of deep learning IDS models into SNNs, with explicit consideration of robustness to unseen attacks and dynamic traffic patterns.
4. Prototype and benchmark neuromorphic IDS implementations on representative embedded and neuromorphic platforms, quantifying detection performance (e.g. detection rate, false positive rate, F1 score), latency, energy consumption and resource utilisation, and evaluating integration strategies within wider vehicle security architectures.

This PhD research will generate the fundamental knowledge required to implement low energy, near real-time distributed IDSs for low resource platforms. An Innovate UK proposal will follow up to transfer these outcomes to a higher TRL level, via a dedicated hardware design, so that the technology can be scaled up in collaboration with industrial partners.