Model building attacks on Physically Unclonable Functions using genetic programming

Abstract

Design, implementation and characterization of Physically Unclonable Functions (PUFs) in silicon have drawn considerable research interest in recent years. However, side-by-side, there are concerns that these PUF circuits, although physically unclonable, might be vulnerable to computational model-building attacks that compromise their security. In this work, we have used ideas from evolutionary computation, specifically genetic programming, to build accurate and compact mathematical models to approximate the response of FPGA-based ring oscillator PUFs (RO-PUFs). Hence, we have demonstrated the feasibility of a computationally simple scheme to model FPGA-based PUFs, and we believe this work will pave the way for similar attempts to attack more sophisticated PUF implementations.