On the Trade-Offs Between the Security and PUFs Properties of Delay-Based PUFs

Abstract

Physical Unclonable Functions (PUFs) are hardware-based security primitives that were promoted as a lightweight solution to secure authentications among Internet of things (IoTs) devices. The security of PUFs was recognized from the beginning and remains central in evaluating the suitability of each proposed instance. In this dissertation, we thoroughly examine the security of two recently proposed designs, which are the Double Arbiter PUFs and Multiplexers PUFs variant called (rMPUFs). We provide a comprehensive analysis on the security of both designs using different implementations. We show that there exist inevitable trade-offs between enhancements in the tolerance against machine learning and other PUFs properties. We compare our findings on both rMPUFs and DAPUFs on XOR PUFs to show the area of enhancement brought by these instances and their drawback compared to previously proposed designs. Reported results and derived conclusions in our work are based on extracted silicon data from Field Programmable Gate Arrays (FPGAs). Our results reveal the vulnerability of all tested designs of DAPUFs and rMPUFs using different numbers of components and challenge sizes. We show that increasing both the number of components and the number of challenge sizes has a positive impact on the security, while the former has a stronger influence in enhancing the tolerance against modeled attacks. Our findings from implementing larger sizes of DAPUFs, including 5-1 DAPUF, show that enhancement on the resilience of the design came at the cost of a deterioration in the reliability and the stability of the generated responses. Nevertheless, the generated challenge-response pairs (CRPs) of DAPUFs have high uniqueness rates, which exhibit a sufficient utilization of the fingerprints property of PUFs. Our experimental results also show that rMPUFs with up to seven multiplexers (MUX) stages are vulnerable to machine learning attacks, and further increase in the number of multiplexers is not desirable since the hardware complexity is already high with seven multiplexers. Nevertheless, our experimental studies confirm the original claim that rMPUFs is similar to MPUF in that rMPUFs also have reasonably good reliability. Therefore, rMPUFs could be a good candidate for weak PUFs with restricted access protocols due to the stability of the generated responses that do not deteriorate with high complexity in their structural design.