Physical Unclonable Functions (PUFs) are circuits
which extract a device dependent secret from inherently available
manufacturing variations. This work focuses on evaluating the
quality of such circuits regarding intra-die correlations, because
canonical quality metrics for PUFs do not sufficiently cover
them. Correlations reduce the effort for an attacker to guess
the secret with the same severity as biases, yet canonical tests
focus only on the latter. Three tests which are used to consider
topological properties along with the secret are introduced and
adapted to quality evaluation of PUFs. To show the efficiency and
effectiveness but also the limitations of the tests, we apply them
to three real-world measurement datasets from ring-oscillator
PUFs on FPGAs, standard SRAM and Two-Stage PUFs on ASICs.
The results show that the presented statistical tests are ideal
candidates to complement state-of-the-art metrics for PUF quality.
