A 690fJ/Bit ML-Attack-Resilient Strong PUF Based on Subthreshold Voltage Attenuator Ring with Closed-Loop Feedback
Haotao Lin, Haibiao Zuo, Qiaozhou Peng, Xiaojin Zhao · 2023
In this paper, we present a strong physically unclonable function (PUF) based on subthreshold 3T voltage attenuators (VAs), featuring a high energy efficiency of 690fJ/bit in 65-nm CMOS. Provided with a VA array having a number of stages and an arbitrary input challenge, a VA ring with closed-loop feedback can be formed by selecting one VA from each stage. With the VA's gain value designed to be less than 1, an initial voltage can propagate along the VA ring for multiple loops, until a final stabilization is achieved. Meanwhile, even with identical design parameters, the selected VA's gain at each stage is variant due to the inevitable device mismatch, and can be further changed dynamically by its input voltage, both of which lead to extremely high complexity for the above stabilized entropy voltage. By digitizing the entropy voltage with another 3-stage current-starved amplifier chain, the generated challenge-response-pair (CRP) space can exhibit strong resilience to various machine-learning attacks, even with a huge CRP training set size up to 50 million. Moreover, the fabricated prototype chips show excellent randomness by passing the widely-adopted evaluation tools.