A Novel Relative Frequency Based Ring Oscillator Physical Unclonable Function

Abstract

Silicon physical unclonable function (PUF) exploits the random variation in the manufacturing process to extract unique information for each chip. Ring oscillator PUF (RO-PUF) is easy to implement for both ASIC and FPGA, and it becomes one of the most popular techniques to build a PUF. However, the unwanted systematic process variation degrades the randomness and uniqueness of RO-PUF. To reduce the effect of the systematic process variation and improve the randomness and uniqueness of RO-PUF, we propose a novel relative frequency based RO-PUF which exploits the differences between relative frequencies instead of absolute frequencies to generate unique secrets. Our proposed RO-PUF is lightweight, and it does not need extra hardware resources compared with the conventional RO-PUF. Experimental results on 193 FPGAs showed that the proposed RO-PUF could improve the average fractional hamming weight of a n-bit response from 25.34% to 50.40% and the average inter-die fractional hamming distance from 31.40% to 46.83%. Moreover, we propose a post-processing scheme to improve the reliability of the enrolled challenge response pairs, and it could reduce the error rate of the relative frequency based RO-PUF from 17.05% to 5.73% in the worst case.