Investigation of the Influence of Multi-Valued Symbol Extraction on PUF Randomness

Physical unclonable functions (PUFs) are used to generate fingerprints of circuits or to derive unique keys for hardware identification and cryptographic applications. For security applications, these keys must meet strict requirements, such as randomness, uniqueness, and steadiness. The latter two are mostly covered via inter- and intra-hamming-distance, while the randomness is often neglected in many PUF publications. Even if the uniqueness and steadiness requirements are met, this does not guarantee that the keys do not follow a predictable pattern. This is particularly important for multi-valued (MV) PUF response symbols which extract multiple bits per PUF-cell, causing potential correlation of the bits extracted from the same source of entropy. However, the impact of MV-readouts on randomness has not been investigated in detail. In this paper, we evaluate the randomness of a publicly available PUF dataset, focusing on different multi-valued extraction approaches, and find that the number of extracted symbols together with the chosen encoding have a large impact on the randomness.