The hardware resources in embedded systems are limited,and there may be conflicts between custom instructions and original instructions,resulting in reduced accuracy of instruction code recognition and higher running power consumption.To address this problem,this paper put forward a low-power iterative recognition method for cus-tom instructions in embedded processors.Firstly,the instruction codes of the embedded processor were visualized,and then the instruction images were input into a convolutional neural network for detecting malicious codes in the instruc-tion.After that,an open-source compiler was used to transform the code into a control data flow graph.Meanwhile,subgraphs were enumerated and selected.Finally,the recognition for custom instruction of the embedded processor was completed through code transformation.Simulation results show that the proposed method has advantages such as high detection accuracy and high recognition accuracy for malicious codes,and the accuracy rate is consistently above 70% .The average energy consumption is only 89J.
关键词
嵌入式处理器/恶意代码检测/自定义指令/控制数据流图/指令识别
Key words
Embedded processor/Malicious code detection/Custom instructions/Control data flow graph/In-struction recognition
维普
万方数据