On-receptor computing utilizing vertical-structured cost-effective memristor

Abstract

A breakthrough in this field is the introduction of vertical resistive random-access memory (VRRAM), which features stacked electrodes capable of incorporating multiple devices in the same area, similar to the 3D NAND structure. Despite different fabrication processes, VRRAM retains the simple electrode–switching layer–electrode structure of traditional memristors, making it suitable for mimicking the synapse and neural connections in our brain, thus enhancing its adaptability to neuromorphic computing. Recent studies have aimed to advance functional computing systems, particularly in the realm of “on-receptor computing.” For this system adaptation, memristors need to exhibit sensor-perception behavior alongside traditional computing capabilities. One approach involves emulating the biological sensing process of the human body by establishing artificial nociceptors, which are dorsal root ganglion under the skin capable of sensing external stimuli. In this study, to achieve cost-effectiveness with a high storage density, a four-floor VRRAM is fabricated for implementing on-receptor computing. By applying controlled pulse streams to the VRRAM, the key nociceptive functions, including threshold, relaxation, no adaptation, and sensitization, are successfully demonstrated, highlighting the filament formation-based sensing properties of the memristor. Furthermore, by testing the reactance of the neuron to the applied action, namely, action-dependent synaptic plasticity, behaviors similar to those of biological neurons are demonstrated. Finally, by utilizing the multilevel properties of the VRRAM, a neural network-based pattern recognition system is constructed, showcasing the on-receptor computing capabilities of the VRRAM. © 2024 Elsevier B.V.