PUF modeling attacks: An introduction and overview

Abstract

Machine learning (ML) based modeling attacks are the currently most relevant and effective attack form for so-called Strong Physical Unclonable Functions (Strong PUFs). We provide an overview of this method in this paper: We discuss (i) the basic conditions under which it is applicable; (ii) the ML algorithms that have been used in this context; (iii) the latest and most advanced results; (iv) the right interpretation of existing results; and (v) possible future research directions.