AI-Enabled Digital Twin Framework for TSCA-like Anomaly Detection in FPGA-SoC-Based Industrial Cyber-Physical Systems

Field-Programmable Gate Array System-on-Chip (FPGA-SoC) platforms are increasingly adopted in modern industrial Cyber-Physical Systems (CPSs), enabling real-time control, monitoring, and automation of critical industrial processes. The increasing integration density of modern FPGA-SoC architectures introduces new thermal security challenges, where heat evolves from a reliability concern into a potential source of information leakage. Thermal Side-Channel Attacks (TSCAs) exploit runtime thermal variations to infer sensitive operational, architectural, or cryptographic information from the underlying hardware. While this study is centered on FPGA-SoC platforms, comparable thermal security challenges are increasingly reported across other densely integrated computing architectures, including Multiprocessor System-on-Chip (MPSoC), System-in-Package (SiP), and emerging Three-Dimensional Integrated Circuit (3D-IC) technologies. Consequently, the detection of thermal side-channel intrusions has become a critical hardware security challenge for next generation industrial CPS infrastructures. To address this challenge, an AI-enabled Digital Twin (DT) framework is introduced for TSCA detection in densely integrated FPGA-SoC microarchitectures. By combining thermal behavioral modeling, feature engineering, and machine learning-based anomaly detection, the proposed framework extends conventional Thermal Digital Twin (TDT) approaches beyond monitoring and mitigation toward autonomous thermal threat detection. The proposed framework is experimentally validated using an NI myRIO-1900 platform integrating a Xilinx Zynq-7010 FPGA-SoC representative of modern industrial embedded control architectures. Experimental results demonstrate the feasibility of the proposed framework, achieving an accuracy of approximately 75% with an Area Under the ROC Curve (AUC) of 0.76 using a lightweight Isolation Forest model. These results validate the capability of the proposed AI-enabled Digital Twin framework to learn normal thermal behavioral patterns and autonomously detect anomalous thermal activities potentially related to TSCAs.