基于电磁辐射与深度学习的芯片指令分析

Chip instruction analysis based on electromagnetic radiation and deep learning

针对传统的芯片指令分析的局限性以及指令恢复准确率低下等问题,文中结合旁路分析与深度学习算法实现芯片指令的逆向分析。由于芯片在工作时能够通过不同的方式泄露旁路信号,因此,首先对芯片特性进行分析,找出不同执行指令、操作数据、指令地址之间电磁旁路信号的差异形式;然后,利用深度神经网络构建指令逆向分析模型,在分析连续执行指令信号时,对多周期指令进行分割并提出分阶段识别的方法恢复指令;最后,在STC89C52RC芯片上进行实验并分析,验证该方案的可行性与有效性。

In allusion to the limitations of traditional chip instruction analysis and the low accuracy of instruction recovery,the reverse analysis of chip instruction is realized by combing with the side-channel analysis and deep learning algorithm.The chip characteristics are analyzed to find out the side-channel signals′different forms caused by different execution instructions,operating data and instruction addresses because the chip can leak side-channel signals in different ways when it is working.The deep neural network is used to construct the instruction reverse analysis model is constructed.When analyzing the continuous execution instructions signal,the multi-cycle instruction is segmented and the method of phased recognition is proposed to restore the instruction.The experiments and analyses were performed with the STC89C52RC chip to verify the feasibility and effectiveness of the scheme.