A Highly Secure Reconfigurable Memory-Based Strong PUF for Device Authentication in Internet of Things

Physically Unclonable Functions (PUFs) are hardware-based security primitives that utilize the mechanical and electrical variations that occur during manufacturing to realize binary keys or binary functions. The low overhead from area and power requirements makes it a good application for key generation and authentication in resource constrained devices such as Internet of Things (IoT) devices. However, there are vulnerabilities that PUFs face due to machine learning based attacks on strong PUFs. The nature of a PUF architecture utilized in device authentication is the challenges and responses are transmitted without any encryption. This allows the attacker to capture these challenge-response pairs CRPs to build a Machine Learning model that can predict the responses of new or unused challenges. This paper introduces a novel and highly secure 64-bits reconfigurable memory-based controlled PUF (ReMc-PUF). The proposed architecture generates a fingerprint using the bits generated by powering on a Synchronous Random Access Memory (SRAM) PUF with a unique control logic. Due to the reconfigurability nature of this architecture, a single challenge may result in multiple responses. This feature allows the server to reuse a challenge multiple times in the authentication process. The design synthesis yielded an overhead of 1.879 kGE and 0.2787 mW for area and power respectively. The architecture is lightweight and displayed a strong resiliency against CMA-ES attacks as the highest prediction accuracy obtained was 49.37%.