A switchable multifunctional artificial neuron with rate coding and time‐to‐first‐spike coding for accurate image recognition

Abstract Inspired by the structure and principles of the human brain, the development of artificial neurons for spiking neural network (SNN) has been stimulated. Threshold switching memristors offer a viable pathway for the emulation of biological neurons. However, existing artificial neurons primarily rely on a singular data coding scheme, which diminishes the capacity of artificial neurons as computational units within SNN. In this study, we introduce a switchable multifunctional artificial neuron (SMAN) capable of encoding information via varying spiking frequencies (rate coding) and time‐to‐first‐spike coding. SMAN leverages the conductive filament conduction mechanism and stress characteristics of materials to achieve varying ionic dynamics by simply altering the position of the electrode contacts. Finally, an SNN based on SMAN has been designed, demonstrating effective performance in the Modified National Institute of Standards and Technology image classification task under both initial and noise‐added conditions. This device‐level selective coding scheme significantly enhances the capability of SNN to recognize various types of images. We believe that our successful implementation will promote the universality of artificial neurons across various tasks, bringing innovation and development to the field of neuromorphic hardware.