Graduation tutor:

Physical Unclonable Functions (PUFs) are an emerging security primitive useful for secure key storage mechanisms and anti-counterfeiting measures. To do this securely, it is imperative that PUFs are unique, i.e., possess enough entropy. The aim of this thesis is to develop and implement new methodologies to accurately estimate the entropy of PUFs. To this end, a novel method is presented which estimates the extractable entropy by calculating the mutual information between enrollment and reconstruction measurements. Furthermore, a method to determine uniqueness from the field of biometrics is modified in such a way that the entropy of PUFs can be estimated. Our newly developed entropy estimation methods are compared to other methods found in the literature using a comprehensive experimental evaluation on synthetically generated PUF data with known entropy. We find that the results from our methods are in line with results found in the literature using methods such as the context-tree weighting compression and inter-device distance tests. The contributions of our methods are that the influence of the reconstruction condition (noise) is included in the entropy estimation as well as providing insight on how entropy develops if more measurements are used during enrollment and reconstruction. We find that the extractable entropy increases as more measurements are used during enrollment and reconstruction, but only if PUF biases are reliable across conditions. Furthermore, an entropy analysis is performed on PUF data from the European Union sponsored UNIQUE project, which consists of 192 application-specific integrated circuits housing multiple instances of the most popular intrinsic PUF types. In particular, we performed a comprehensive uniqueness evaluation of four memory based PUF types: the latch, D flip-flop (DFF), buskeeper and SRAM PUF. Our analysis shows that from the four analysed PUF types, the SRAM PUF has the most extractable entropy. The buskeeper also shows strong PUF behaviour, while the DFF and latch PUF have the least extractable entropy