Digitization algorithms in ring oscillator physically unclonable functions as a main factor achieving hardware security

Abstract

Since the discovery of the physical random functions and their subsequent refinement into physical unclonable functions (PUF), a great effort has been made in developing and characterizing these objects attending to their physical properties as well as conceiving a myriad of different examples in the search for a better application-specificity and suitability. However, comparatively little time has been devoted to the analysis of entropy extraction algorithms beyond the recognition of some limitations due to the environment influencing the PUF behavior. In this article we focus on well known PUF candidates based on ring oscillator delay, which are ideal for FPGA prototyping due to their tolerance to asymmetries in routing. We have studied the impact that different digitization algorithms of the responses have over their security properties. Specifically, we have analyzed the response probability distributions that arise from some popular techniques of digitization called "compensated measuring" methods, highlighting their lack of uniformity and how this might translate into cryptanalytically exploitable vulnerabilities. Furthermore, we propose a new family of digitization schemes named k-modular that exhibit both uniformity in response distribution and high entropy density on both physical and response space.