Racing APUF: A Novel APUF against Machine Learning Attack with High Reliability
Zheng Li, Lijia Zhu, Minghui Huang, Ziheng Chen, Shuai Chen, Bing Li · 2019
Physically unclonable functions are lightweight, secure primitives based on inherent variance of integrated circuits. Its security of resistance against machine learning attract more attention. Many improved APUF structures reduce stability while improving resistance to machine learning attack. In this paper, we propose a novel racing arbiter physically unclonable function (R-APUF) which has resistance against machine learning attack while keep an acceptable reliability. The proposed R-APUF's path consists of several cascaded sub-chains based on MAX (MIN) operation, which return a MAX (MIN) delay of multi-channel. Software simulation and FPGA implementation are applied to evaluate the performance of the proposed PUF. Evolution strategy used to predict 4-channel 2-stage R-APUF maintain 75% prediction accuracy, while this structure keeps a 94.737% reliability closed to normal APUF.