Title:A Challenge Obfuscating Interface for Arbiter PUF Variants against Machine Learning Attacks

Abstract:Security is of critical importance for the Internet of Things (IoT). Many IoT devices are resource-constrained, calling for lightweight security protocols. Physical unclonable functions (PUFs) leverage integrated circuits' variations to produce responses unique for individual devices, and hence are not reproducible even by the manufacturers. Implementable with simplistic circuits of thousands of transistors and operable with low energy, Physical unclonable functions are promising candidates as security primitives for resource-constrained IoT devices. Arbiter PUFs (APUFs) are a group of delay-based PUFs which are highly lightweight in resource requirements but suffer from high susceptibility to machine learning attacks. To defend APUF variants against machine learning attacks, we introduce challenge input interface, which incurs low resource overhead. With the interface, experimental attack study shows that all tested PUFs have substantially improved their resistance against machine learning attacks, rendering interfaced APUF variants promising candidates for security critical applications.