Anomaly detection of CAN bus messages on neuromorphic hardware

Anomaly detection is becoming an essential component of the modern automotive industry. With the increasing prevalence of smart vehicles, the number of Electronic Control Units (ECUs) integrated within automotive systems is also growing. These ECUs communicate through the Controller Area Network (CAN) to perform critical tasks such as braking, seatbelt control, and navigation. However, the lack of security in CAN networks makes them vulnerable to malicious message injections, potentially endangering passengers and drivers. Traditional anomaly detection systems relying on Central Processing Units (CPUs) and Graphics Processing Units (GPUs) are computationally expensive, energy-intensive, and introduce significant latency due to continuous data monitoring. In this work, we propose a low-power anomaly detection system implemented on Intel’s Loihi and BrainChip’s Akida neuromorphic hardware. Our system achieves the same detection accuracy as conventional CPU-based methods while significantly reducing energy consumption. To the best of our knowledge, this is the first anomaly detection implementation on Akida. Additionally, we provide a comparative analysis of accuracy and power efficiency between these neuromorphic processors, marking the first such evaluation for anomaly detection applications. Our system shows it can classify between anomalous and normal data successfully in low power. Our results demonstrate the potential of neuromorphic computing for secure and efficient anomaly detection in power-constrained environments.