A Configurable Artificial Neuron Based on a Threshold-Tunable TiN/NbOₓ/Pt Memristor
Yongzhou Wang, Hui Xu, Wei Wang, Xumeng Zhang, Zuheng Wu, Ran Gu, Qingjiang Li, Qi Liu
Abstract
Due to their simple structures and high-density integration, Memristors are employed to construct hardware spiking neurons. In this letter, we present a TiN/NbOx/Pt memristor with a tunable threshold for constructing configurable neurons. By applying positive tuning voltages with different compliance currents, the device exhibits multilevel negative threshold voltages. Additionally, by changing the amplitudes/widths of positive tuning pulses, we obtain a quasi-linear modulation of negative threshold voltages. Based on such a device, we construct a leaky integrate-and-fire neuron with configurable curves between the response speed and the excitatory input. These results indicate that our device is suitable to construct a configurable neuron, which is expected to maintain homeostasis and improve the stability of computing systems.