CPU-Doctor: when a device’s heart-beat can be an acoustic side-channel disassembler

In recent years, the ever increasing need of computing has lead to design of modern embedded computing devices that are dedicated to provide enhanced system performance. But, due to inadequate security monitoring and the challenges of ongoing operating systems’ patching, modern embedded computing systems are not supposed to be growing as much as seen in recent years. Specifically, embedded systems are applied in industry and household devices with some hesitation from people, because they are susceptible to malware, software piracy and data exfiltration. Therefore, it is vital to protect embedded devices from malicious activities and safeguard the integrity of executable software. In this paper, we propose (to the best of our knowledge) the first acoustic side-channel-based disassembler to investigate the real-time functioning of embedded systems at the instruction level. More specifically, we highlight the fact that the Central Processing Unit (CPU) (micro-controller in case of edge/embedded devices) can have a heart-beat (sound). This heart-beat extraction and analysis methodology are discussed in detail in this work. To design our proposed disassembler, we initially collect templates from a source device and then apply machine learning algorithms to uniquely identify instructions executed on the device. For this purpose, we use a hierarchical classification framework, to implement an acoustic side-channel disassembler “CPU-Doctor” for ATMEGA328P and ARM Cortex A53. “CPU-Doctor” exactly identifies group of the instructions with 100% ssaccuracy and uniquely determines the instruction with 96.67% accuracy in verification phase. Although we have presented the experimental analysis on ATMEGA328P and ARM Cortex A53, our approach is generic in nature and can be applied to any processor.