Molecularly triggered multilevel conductance of oxidized MXene chemimemristor, mimicking biological olfactory system at single-device level

Abstract

In sensor-computing systems inspired by biological sensory systems and neural networks exhibit considerable potential for addressing the energy inefficiencies of traditional computing architectures. In this study, we present neuromorphic systems integrated with tunability of the molecular dynamics, embodied by the concept of a “Chemimemristor.” The outstanding performances are achieved, including 4 to 6 resistive switching ranges, ~104 to ~106 ON/OFF ratio, 2.7 V to 3.8 V switching voltage, stable retention, and endurance over 1000 cycles. Moreover, the device demonstrates basic synaptic behaviors, emulations of excitatory and inhibitory postsynaptic currents (EPSC and IPSC), and short-term plasticity with tunable dynamic ranges under CO2 trigger. In addition, the conductance of the active layers depends on the molecular concentration, consistent with molecularly triggered multilevel conductance with stable retention, making them ideal for in-sensor computing applications. The image classification tasks, using the Fashion-MNIST dataset, show device accuracy under CO2 concentrations from 5 to 25 ppm of about 72 % to 80 %. The synaptic behavior remains stable after 100 times of bending. The fabricated oxidized MXene-based device holds promise for learning molecular information, providing a platform for real-time, molecule-responsive neuromorphic systems, as a planar and flexible electronic device.