Side-Channel Based Disassembler for AVR Micro-Controllers Using Convolutional Neural Networks

Reverse engineering using Side Channel Attacks (SCA) have been known as a serious menace against embedded devices. The attacker could employ side channel data to retrieve some sensitive information from the device, security analysis, existence of a library in the device or execution of a special stream of codes. Side channel data could be gathered from the power consumption or electromagnetic radiations by the device. In this paper, we propose a disassembler to extract the instructions of the device under attack. A deep convolutional neural network is employed to make templates of the target to use it for real-time scenarios. Short Time Fourier Transform (STFT), and Mel-Frequency Cepstrum Coefficients (MFCC) are utilized as feature extractors. The proposed method consists of two different parts: 1) Hierarchical scenario and 2) Sole model. Atmel 8-bit AVR micro-controller is employed as the target device under attack. Our results indicate that, even with an experimental and low cost setup a vast number of instructions are detectable. The proposed method reaches 98.21% accuracy on the real code, outperforms state-of-the-art methods on the proposed dataset.