Flexible Artificial Neurons Based on NbO x for Robust Tactile Perception in a Wide Temperature Range

Abstract

Neuromorphic tactile sensing that emulates biological touch enables artificial systems to work with high efficiency and low energy consumption. As tactile-sensing scenarios may involve curved surfaces and different temperatures, artificial spiking afferent neurons require both flexibility and temperature robustness at the front end of artificial systems. Here, we report flexible artificial spiking neurons based on niobium oxide (NbOx) threshold-switching (TS) memristors prepared on polyimide (PI) substrates. Temperature-dependent transport analysis indicates that their temperature robustness is associated with a bulk switching pathway in the NbOx film. By integrating an oscillation circuit, the NbOx spiking afferent neurons convert analog inputs into frequency-modulated voltage spikes and retain oscillatory capability from –80 to 80 °C at a bending radius of 5 mm. Moreover, pressure sensors are coupled with the NbOx neurons to form mechanoreceptors that encode pressure as oscillation frequency. In the numerical tactile-recognition workflow, a convolutional neural network (CNN) classifies spatial frequency maps generated from the fitted pressure-to-frequency relations, achieving accuracies of 97.5% under the baseline condition and 93.2% at 80 °C at a bending radius of 5 mm. This work demonstrates that flexible NbOx neurons can serve as compact event-driven front ends for wide-temperature robotic tactile sensing.