Predicting PUFs faster: using high entropy input codes to improve machine learning attacks on a category of Physical Unclonable Functions

PUFs are a class of security primitives that rely on statistical variations of integrated circuits production processes to provide authentication without the need explicitly storing cryptographic keys. Nevertheless several PUF architectures have shown to be predictable using machine learning techniques. This work extends a code published on related work that maximizes the output entropy for a single PUF architecture to a class of delay based PUF architectures and provides a method to make machine learning attacks more efficient against these devices.