A PUF Based on the Non-Linearity of Memristors

As autonomous devices are increasingly used in security and safety-critical applications the security of the systems they comprise is of increasing concern. In such situations it is important that devices can be securely identified and trusted. When an IC or device is in the supply chain, or in the field, the lack of control over actors who can obtain physical access can compromise the trust and overall security of a system. Counterfeit chips may be incorporated into the device, compromising reliability or security. Additionally, for implemented devices, keys stored on-device may be copied by a bad actor. To help improve the security of such devices this paper proposes a new physical unclonable function (PUF) architecture, based on a TiOx memristor-based resistive memory (RRAM), that exploits the inherent analogue non-linearity in resistance of some memristor technologies. By directly exploiting non-linearity of memristor cells, rather than relying on the devices' absolute resistance at a single test voltage, a multi-bit-per-comparison PUF is created. As the architecture directly exploits cells' non-linearity, an additional source of hard-to-clone entropy is incorporated.

I. Introduction

Trust is an important part of modern security and safety-critical systems. Counterfeit, tampered or otherwise illegiti-mate hardware can compromise the robustness or trustwor-thiness of a system resulting in anything from early failure to compromise by a bad actor. Counterfeit ICs may be unwittingly incorporated into deployed systems. Such a chip may not perform to the expected standard, having subpar performance or even incorporating security vulnerabilities. Vulnerabilities could be introduced either intentionally or unin-tentionally, depending on the motivations of the counterfeiter. To ensure proper security of a deployed device it is imperative that devices in the field can be securely and uniquely identified in a way that is difficult, or impossible, for an attacker to effectively clone.