Geopolymer memristor-based physical reservoir computing for digit recognition

Memristors are solid-state devices that share many information processing capabilities of biological synapses, allowing brain-inspired memory and computing. Among memristor-based neuromorphic computing systems, physical reservoir computing has gained significant attention in recent years due to its low computational cost and suitability in handling multiple tasks by leveraging the short-term memory property of artificial synapses. In our previous studies, we developed geopolymer (GP)-based low-cost memristors and demonstrated their activity-dependent behaviors, such as short-term plasticity behaviors [including Paired-Pulse Facilitation (PPF) and Paired-Pulse Depression (PPD)] and long-term plasticity behaviors [including spike-timing-dependent plasticity and Spike-Rate-Dependent Plasticity (SRDP)]. GPs are a class of inorganic polymers formed by the alkali activation of aluminosilicate precursors. In this study, we present an efficient and low-cost geopolymer-based reservoir computing system capable of recognizing a computer-generated 5 × 5 binary digit set and a handwritten Modified National Institute of Standards and Technology digit dataset with up to an accuracy of 89%, combining both experimental- and simulation-based approaches. The pattern recognition capability of GP memristors suggests their potential application in integrated energy-efficient and real-time structural health monitoring.