Spiking frequency adaptability and multi-weight synergy in artificial neuronal modules via bifunctional NbOx memristors.

Abstract

To address the limitations of current artificial neurons in neuromorphic hardware implementation, NbO_x-based bifunctional memristors are fabricated to construct oscillatory units and advanced neuronal modules. NbO_x-based memristors operate as either threshold-switching memristors (TSMs) or dynamic memristors (DyMs), depending on whether electroforming is applied. TSMs are employed to build oscillatory units and further reconfigured into a weighted multi-terminal neuronal module, enabling real-time spatiotemporal summation of input spikes based on the leaky integrate-and-fire model. This module demonstrated the capability to perform spike summation and multi-weight synergy. Leveraging the gradual resistance change characteristic of DyMs, a sequential encoder is implemented, allowing the system to recognize and respond to the temporal order of spiking signals. Additionally, a DyM is integrated into the oscillatory unit to construct intensification and attenuation neurons, enabling short-term spiking frequency adaptation. The versatile spiking performance of our NbO_x bifunctional memristor provides a strategic foundation for developing artificial neurons for next-generation bio-inspired spiking neural networks.