基于反向神经网络的硬件木马识别

Hardware Trojan Recognition Based on Reverse Neural Network

针对侧信道、逻辑功能、逆向工程等硬件木马检测技术存在高成本、高设备要求、易受工艺噪声影响和不适用于大规模电路等问题,提出了一种基于反向神经网络的门级硬件木马识别方法。通过提取电路的门级网表特征,使用电路特征集构建全新反向神经网络,训练成门级硬件木马分类器。通过不断调整神经网络的隐藏层数和节点数,实现门级硬件木马识别,最终达到99.82%的正常电路识别率、87.83%的木马识别率和99.27%的线网准确度,在正常电路几乎完全识别的前提下,获得了较高的硬件木马识别效果。

Aiming at the problems of side channel, logic function, reverse engineering and other hardware Trojan horse detection technologies, such as high cost, high equipment requirements, vulnerability to process noise and unsuitability for large-scale circuits, a gate-level hardware Trojan horse recognition method based on reverse neural network is proposed. By extracting the gate-level network table features of the circuit, a new back-propagation neural network is constructed using the circuit feature set and trained as a gate-level hardware Trojan horse classifier. By adjusting the number of hidden layers and nodes of the neural network, the gate- level hardware Trojan horse recognition is realized. The recognition rate of 99.82% normal circuit, 87.83% Trojan horse and 99.27% accuracy of the network are achieved. On the premise of almost complete recognition of normal circuit, a higher recognition effect of hardware Trojan horse is obtained.