Anomalous IoT Behavior Detection by Generated Power Waveforms with Hyper-parameter Tuning

With the recent spread of Internet of Things (IoT) devices, the security issues for hardware devices have increased. When an IoT device runs an application program, the power consumption of the running application is combined with the power consumption of the device hardware itself, resulting in a complex power waveform. To detect anomalous application behaviors using the power waveforms, it is necessary to subtract the steady-state power waveform due to the device hardware from the measured power waveforms and extract only the application power waveform. In this paper, we propose a method for detecting anomalous IoT behaviors using generated power waveforms by introducing hyper-parameter tuning. The proposed method detects anomalous behaviors by generating a highly accurate steady-state power waveform and an application power waveform by adjusting the waveform period through hyper-parameter tuning, even if the measured power waveform includes large noises. Experimental evaluation demonstrates that we successfully detect anomalous behaviors from an AES encryption circuit containing a hardware Trojan on an FPGA device, while the existing state-of-the-art method cannot.