Data-driven detection of peripheral module activity in embedded systems using side-channel power modeling

Peripheral modules in embedded systems frequently constitute the most energy-consuming elements, thereby rendering their power characterization indispensable for both system optimization and anomaly detection. Energy profiling in embedded systems is a critical task for optimizing performance and ensuring reliability. In circumstances where source code is not accessible, the development of power models provides a viable alternative to application-level analysis. This study introduces a generalized methodology for the monitoring and characterization of total power consumption, for creating precise power models for peripheral modules through the application of pattern recognition techniques. While the proposed framework is broadly applicable, it is particularly pertinent to domains such as cybersecurity, intrusion detection, and power optimization, wherein the detection of anomalous activity is paramount. The methodology encompasses the experimental measurement of an embedded system’s power consumption with a specific emphasis on peripheral modules, the extraction of salient features from the acquired power traces, and the subsequent classification of these features to construct accurate power models. Empirical results demonstrate that discrete peripheral module activities can be reliably inferred from aggregate power measurements, thereby providing comprehensive insights into their energy consumption behaviors.