DCC-PUF: A Delay-Composition-and-Coupling Physical Unclonable Function Design for FPGA

Abstract

This paper proposes a Delay-Composition-and-Coupling Physical Unclonable Function (DCC-PUF). Experimental results on a 28 nm Xilinx Artix-7 FPGA show that eight isomorphic instances deployed across different clock regions achieve a region-level uniqueness of 47.18%. Under ten repeated measurements with majority voting, the proposed structure attains an average reliability of 99.87%. In addition, the measured uniformity is 48.57%, and autocorrelation analysis indicates weak low-order linear correlation in the generated responses. Modeling attack experiments based on 50,000 challenge-response pairs (CRPs) show that the DCC-PUF can effectively suppress the learning capability of conventional models, including logistic regression (LR), least squares (LS), and support vector machines (SVM), while limiting the prediction accuracy of covariance matrix adaptation evolution strategy (CMA-ES) to around 60%. However, high-capacity neural networks can still achieve high prediction accuracy.