Anomaly detection in embedded systems using simultaneous power and temperature monitoring

Embedded systems were traditionally considered impenetrable against security violations for a long duration. The recently discovered security attacks like W32.Stuxnet and the Duqu worm changed this almost established belief. Further, usual methods of anomaly detection cannot simply be employed in embedded systems since they are usually resource-constrained and the detection methods might interfere with performance of the main system tasks. However, behavior of applications running in embedded systems is relatively known, and profiling the correct behavior and comparing it with the runtime behavior is accounted as an effective approach for anomaly detection with no such interferences. In this paper, we consider the inter-effects of processor temperature and power consumption of embedded systems, and present a novel technique to detect malicious activities with minor delays with respect to the occurrences. This is done through analyzing the simultaneously observed power and temperature of the processor. The efficacy of the proposed method is evaluated by using the power and temperature traces of a real processor obtained through detailed simulations.