Experimental evaluation of Physically Unclonable Functions in 65 nm CMOS

Abstract

We present a silicon characterization vehicle implementing six different constructions of intrinsic Physically Unclonable Functions (PUFs). The design contains four different memory-based PUFs, one of which is a novel buskeeper PUF, and two different delay-based PUFs. Test chips are fabricated in 65 nm Low Power (LP) technology, using a standard cell ASIC design flow for the memory-based PUFs and a full custom flow for the delay-based ones. This test vehicle enables a comprehensive experimental evaluation of individual PUF implementations as well as a comparative analysis across different PUF types for the same silicon technology. PUF responses are obtained from 192 device samples and the uniqueness and reliability of the implemented PUFs are evaluated. In addition, the effects of varying temperature and silicon device ageing on the PUF characteristics are extensively studied.

Physically unclonable function using CMOS breakdown position

“…In CMOS technologies, the widely discussed SRAM PUF reads out the power-up patterns, which are ideally random and unique for different cells and dies [4], [7]. The entropy source of the SRAM PUF is process variations in a pair of matched invertors, resulting in the different pull-up and pull-down strengths.…”

“…A PUF implemented in commercial silicon technology enables chip level cryptographic functions such as secret key generation, entity authentication and IP protection [1]- [4]. Ideally, a silicon PUF utilizes intrinsic sources of variation on devices or interconnections, to extract a unique data pattern which is unpredictable and reproducible in various noisy environments.…”

Bias-based modeling and entropy analysis of PUFs

“…This is no surprise, since SRAM PUFs were found to be the most reliable and unique PUFs in [9,13]. Furthermore, the number of PUF cells per mm 2 is also highest for the SRAM PUF.…”

“…Unfortunately two Latch PUFs per ASIC are unusable due to faults in the addressing logic. Furthermore, during preliminary testing, one DFF instance was found to be very unreliable when compared to the other instances (also noted in [9,13]). This instance we also excluded from the test data.…”

“…Note also that even at k = 1 (a single enrollment measurement) it is advantageous to take > 1. Finally, based on the mutual entropy results and known size of the PUF instances on the UNIQUE ASIC (based on [13]) the minimum number of extractable bits per mm 2 of each PUF type can be calculated. The results of the calculation can be found in Table III.…”