Halide perovskite memristors as flexible and reconfigurable physical unclonable functions

Despite the inspiring demonstrations of security primitives with CMOS electronics, realising efficient roots of trust for resource-constrained hardware remains a significant challenge. Here, we exploit the stochastic memristive switching behaviour in one dimensional halide perovskites to solve this.

Like Comment
Read the paper

Background: With the proliferation of Information and Communication Technologies (ICT), “information” or “data” has become the most valuable commodity and secure information storage is the most sought after technology today. The inability of software-oriented security alone to combat the increasing threats of the cyberspace has called for the development of hardware cryptographic solutions.

 

Physical Unclonable Functions (PUFs) are among the most promising hardware cryptographic primitives capable of generating random, unique secret bit-strings on the fly. Their feasibility stems from the unique signatures induced in a component’s physical property via stochastic variations in the manufacturing process, thus making cloning impossible even by the original manufacturer. Despite impressive demonstrations of PUF designs based on CMOS technology- e.g. arbiter PUF, ring oscillator PUF and SRAM PUF; the emergence of nanoelectronics, and breakthroughs in new semiconducting materials and device configurations have provided computer engineers with entirely new state variables such as electron spin and resistance to represent information. These technologies could potentially trigger new modalities for attack and render existing defences unacceptable. Therefore, there is a critical need to evaluate robust “roots of trust” in the context of new nanoscale devices.

 

Memristor is one such technology that is envisioned to push the future of computing beyond Moore’s law by enabling new neuromorphic concepts such as in-memory computing and edge analytics. In addition to their data storage and computing capabilities, intrinsic stochasticity of the memristive switching physics could enable novel PUF designs with enhanced entropy, but is hitherto undemonstrated.

 

Research Gap: Organic electronic materials with a rich reservoir of exotic switching physics offer an attractive, inexpensive option to design efficient hardware crypto, but have not been investigated till date.

 

Our Hypotheses: An ideal PUF electronic material should possess multi-dimensional entropy- co-existence of intimately coupled, multiple charge transport properties that would act as a black box to not only the attacker, but also to the original manufacturer. Organic materials possess high degrees of freedom such as tunable optoelectronic charge transport, and exotic switching dynamics often incorporating redox reactions, all tailorable via facile chemical synthesis and functionalization. All these phenomena can be viewed as novel entropy sources to design highly-secure PUFs, but have not been investigated till date.

Hybrid organic-inorganic memristors based on ABX3 halide perovskites is one such material platform that possesses very large degrees of freedom or state variables, making them ideal candidates for next-generation digital fingerprints/PUFs. The intimately coupled ionic and electronic conductivity, localized self-doping, switchable majority carrier concentrations and electrochemical reactions with metal electrodes are excellent sources of entropy for cryptographic primitives to generate device-specific secret bits. Here, we hypothesize that embracing these intrinsically coupled sources of entropy in HP memristor PUFs (HP memPUFs) in combination with advanced bit-stabilization techniques will be advantageous to create robust hardware security primitives for secure key generation and device authentication.

 

Our Approach: We experimentally demonstrate reconfigurable weak PUFs or physically obfuscated keys (POKs) with a 1kb array of HP memristors. We utilize a pyridinium-based HP material featuring molecular one-dimensional (1-D) lead-iodide lattices, namely PrPyr[PbI3] for this purpose. This is the very first implementation of an non-oxide memPUF and also the largest ever implemented HP memristor array till date to the best of our knowledge. The secret keys are generated from the measured resistance distribution of HP memPUFs as the source of entropy. In this work, we demonstrate

(i) both weak and strong PUF modes,

(ii) write-back assisted near perfect uniqueness, reliability and randomness,

(iii) reconfigurability to refresh keys when necessary, as well as

(iv) a recurrent scheme of response generation that enhances security of small crossbar arrays against machine learning-based modelling attacks.

Fig 1. Halide perovskite memristor PUFs (HP memPUFs). a HPs possess a rich reservoir of intimately coupled charge transport properties that could serve as sources of entropy to design new kinds of PUFs. Here, stochasticity in the high resistance state of HP memPUF cells in the crossbar provides parametric support for the generation of unique cryptographic keys. Concept schematic of product authentication. The flexible HP memPUFs developed can be attached onto the packaging of goods or products by the original manufacturer. Each individual product would have a unique PUF ID. End users (e.g. consumers) can validate the product and certify the provenance by accessing the enrolled keys in a secure cloud database.

 

Compared to oxide PUFs, our devices are flexible and can be facilely printed on large areas while providing comparable performance. With respect to other recently reported oxide memristive PUFs, our devices pass all the 15 NIST randomness tests and are shown to thwart machine learning attacks on strong PUF mode without increasing the number of crossbars, but by utilizing recurrence.

 

In short, since this is the very first report on a non-oxide memristor PUF, all the analyses and benchmarking metrics presented in this work represent the state-of-the-art and represents a significant advance for organic electronics. Our work extends the frontiers of current silicon and oxide memristor-based PUF implementations by demonstrating unmatched stochasticity in the charge transport properties of organic semiconductors, especially halide perovskites, for applications in hardware security and cryptography. The proposed approach can be generalized to a wide variety of memristor platforms, opening up new possibilities with improved scalability and CMOS compatibility, making this report of interest to wide scientific audiences in materials science, computing, applied physics and electrical engineering.

 

Please check out our recent work published on Nature Communications: “Halide Perovskite Memristors as Flexible and Reconfigurable Physical Unclonable Functions” at the link: https://www.nature.com/articles/s41467-021-24057-0

Arindam Basu

Professor, City University of Hong Kong

Neuromorphic circuits and algorithms, Brain-machine interfaces