A Lightweight IDS Framework Using FPGA-Based Hardware Fingerprinting on Zynq SoC

AAs cyber-physical systems (CPS) and embedded devices become increasingly interconnected, secure and efficient device authentication is essential. While lightweight cryptographic methods like elliptic curve cryptography (ECC) offer one solution, hardware-based fingerprinting complements them by exploiting intrinsic device-level variations. This paper presents a design approach with practical implementation for a lightweight intrusion detection system (IDS) using an FPGA-based hardware fingerprinting system on a Zynq System-on-Chip (SoC). The framework analyzes inherent electrical imperfections in ADCs, demonstrating that these variations do not affect the uniqueness or stability of intrinsic unclonable hardware fingerprints of electronic devices. Our approach captures CAN-like step signals from hardware devices via an AD9467-FMC ADC, extracts 38 handcrafted time-domain and frequency-domain features, and classifies devices using a Random Forest model optimized for constrained environments. By varying ADCs and probes during evaluation, we confirmed that hardware variability does not impact device identification, with the model maintaining 100% test accuracy. Dimensionality reduction to 19 features further improved computational efficiency without sacrificing accuracy. Recent studies in hardware fingerprinting, particularly in vehicular networks, underscore the importance of modeling ADCinduced variations and employing machine learning for real-time security. Our results affirm the effectiveness of hardware-agnostic fingerprinting combined with a lightweight machine learning model, offering a robust design solution for intrusion detection in embedded systems.