A Low Power Diode-Clamped Inverter-Based Strong Physical Unclonable Function for Robust and Lightweight Authentication
Yuan Cao, Chao Qun Liu, Chip-Hong Chang · IEEE Transactions on Circuits and Systems I Regular Papers · 2018
Strong physical unclonable function (PUF) transcends the limitations of legacy secure key storage methods as emerging security primitive for cryptography and device identification/authentication. In this paper, a new low power and reliable mono-stable strong PUF is proposed. Its primary entropy is derived from the process variations of the parallel diode-clamped single inverter ring working in the subthreshold region. Due to manufacturing process variations, the monostable state of the output voltage of single inverter ring is Gaussian distributed around the half Vdd point. The spread of the Gaussian is broadened by mixing it with another Gaussian distributed trip point obtained from a diode-clamped parallel inverter stage. As the main entropy of the raw response bits is derived from a mono-stable circuit, it has greater immunity to perturbances introduced by operating environments. In addition, as the mono-stable state for the output voltage is a non-linear combination of individual inverter rings, the resilience against machine learning attacks can be improved. The prototype chip was fabricated using a commercial 40-nm CMOS technology. The measurement results show that the power consumption of the 64-bit mono-stable PUF is merely 3.85 μW. The native bit error rate is <;8% at 0.9 ~ 1.3 V and -40~ 90°C, which can be further reduced to <;1% using the proposed thresholding technique. The proposed PUF reduces the accuracy of support vector machine and reliability-based covariance matrix adaptation evolution strategy attacks by 36× and 75×, respectively, over that of arbiter PUF.