A statistical noise model for a class of Physically Unclonable Functions

TL;DR

This paper introduces a Bayesian statistical noise model tailored for Physically Unclonable Functions, accounting for varying error probabilities across bits, and evaluates its effectiveness with real device data.

Contribution

It presents a novel Bayesian noise model for PUFs that captures non-uniform error behaviors and compares different parameter estimation methods.

Findings

The model accurately reflects the varying error probabilities in PUF outputs.

Parameter estimation methods differ in accuracy and computational efficiency.

Applying the model improves error correction strategies for PUF-based security.

Abstract

The interest in "Physically Unclonable Function"-devices has increased rapidly over the last few years, as they have several interesting properties for system security related applications like, for example, the management of cryptographic keys. Unfortunately, the output provided by these devices is noisy and needs to be corrected for these applications. Related error correcting mechanisms are typically constructed on the basis of an equal error probability for each output bit. This assumption does not hold for Physically Unclonable Functions, where varying error probabilities can be observed. This results in a generalized binomial distribution for the number of errors in the output. The intention of this paper is to discuss a novel Bayesian statistical model for the noise of an especially wide-spread class of Physically Unclonable Functions, which properly handles the varying output stability and also reflects the different noise behaviors observed in a collection of such devices. Furthermore, we compare several different methods for estimating the model parameters and apply the proposed model to concrete measurements obtained within the CODES research project in order to evaluate typical correction and stabilization approaches.