A robust Euclidean metric based ID extraction method using RO-PUFs in FPGA

Abstract

Main problems in FPGA-based ring oscillator (RO) PUFs are that the RO frequencies are highly sensitive to operating conditions and other types of global variations. In addition, the RO frequencies are highly correlated by local variations. Therefore, in practice, conventional RO-PUF application schemes using Hamming distance normally require complex identification (ID) extraction algorithms and/or a large number of ROs to ensure a high level of uniqueness and reliability of the extracted IDs. In this work, we proposed a novel scheme of ID extraction based on Euclidean distances. Our proposed scheme stably and reliably generates the ID using a non-selective small number of ROs. Specifically, the generated IDs are mostly non-sensitive to global variables and operating conditions such as ambient temperature. In oppose to Hamming-based extraction, the close-frequencies ROs are normally removed to avoid flipping bits, our proposed scheme takes all those into account and simplifies the extraction process. Experiments on our available hardware have shown a very good level of reliability and uniqueness with ID collision rate is estimated less than 2×10−9.