In this extended abstract a large-scale analysis of 4-
way Arbiter PUFs is performed with measurement results from
1000 RFID tags. Arbiter PUFs are one of the most important
building blocks in PUF-based protocols and have been the
subject of many papers. However, in the past often only software
simulations or a limited number of test chips were available for
analysis. Therefore, the goal of this work is to verify earlier
findings in regard to the uniqueness and reliability of Arbiter
PUFs by using a much larger measurement set. Furthermore, we
used machine learning algorithms to approximate and compare
the internal delay differences of the employed PUF. One of the
main research questions in this paper is to examine if any
“outliers” occurred, i.e., if some tags performed considerably
different. This might for example happen due to some unusual
manufacturing variations or faults. However, our findings are that
for all of the analyzed tags the parameters fell within the range
of a Gaussian distribution without significant outliers. Hence, our
results are indeed in line with the results of prior work.
