基于MLP算法的Glitch PUF机器学习攻击

Machine learning attack to Glitch PUF based on MLP algorithm

随着攻击技术的不断进步,基于机器学习(Machine Learning,ML)、深度学习(Deep Learning,DL)等技术的建模攻击严重威胁了PUF的安全。针对Glitch PUF单元电路静态输出的缺陷,首次提出使用多层感知器(Multilayer Perceptron,MLP)算法对Glitch PUF进行机器学习,解决了Glitch PUF输出为非线性可分数据的问题,能够对Glitch PUF攻击并预测其输出。实验表明,对比于逻辑回归(Logistic Regression,LR)算法和随机森林(Random Forest,RF)二分类算法,提出的MLP算法显著降低了预测错误率。

With the development of attack technology,modeling attacks based on ML,DL and other technologies threaten the security of PUF seriously.In view of the flaw in Glitch PUF static output,this paper first proposes the multilayer perceptron algorithm for Glitch PUF machine learning to analysis nonlinear output data.The experimental results show that the MLP algorithm proposed in this paper can significantly reduce the prediction error rate compared with the logistic regression algorithm and the random forest classification algorithm.