Generalized Bio-Plausible Neuron Model with Refractoriness, Frequency Adaptation and Dynamic Threshold

Abstract

Real biological neurons are dynamical systems that can process a nonlinear, arbitrary input signal with a dynamic threshold voltage, exhibit refractory time, and have a smooth [Formula: see text]–[Formula: see text] curve. This paper presents a generalized bio-plausible neuron model with refractoriness, frequency adaptation, and dynamic threshold. The spiking characteristics of the generalized neuron are enhanced with less circuitry and fewer parameters compared to older circuits. The proposed silicon neuron can generate all known spiking behaviors, just like a biological neuron. The circuit is designed using a 180[Formula: see text]nm standard CMOS process. The entire design uses 22 transistors with a total power consumption of only 2[Formula: see text]nW for a 1[Formula: see text]nA step current at a spike frequency of 122[Formula: see text]Hz. The power consumption and spike frequency rate depend on the input current. The proposed neuron is also capable of displaying both Type I and Type II frequency response characteristics. In addition, our neuron model underwent Monte Carlo simulation with 1000 sample runs to assess its robustness and stability against process variations and uncertainties. The results indicate that the average energy consumption of the proposed neuron model falls within the range of 16[Formula: see text]pJ to 68[Formula: see text]pJ.