Artificial Neuron with Spike Frequency Adaptation Based on Mott Memristor

An artificial neuron with spike frequency adaptation (SFA) characteristic is demonstrated by integrating a Mott memristor with adaptation unit. By combining experimental measurements with circuit simulations, an adaptive neuron model is established and applied to a single-layer spiking neural network (SNN) trained with unsupervised spike-timing-dependent plasticity (STDP). The simulated classification accuracy on the MNIST dataset is improved from 80.21 % to 83.3% by incorporating SFA. Furthermore, the resulted uniform neuronal activities also alleviate the device requirements on endurance and uniformity
$(> 4\times)$
in the hardware implementation of SNN.