A Stochastic Leaky-Integrate-and-Fire Neuron Model With Floating Gate-Based Technology for Fast and Accurate Population Coding

An analytical model has been developed for stochastic leaky-integrate-and-fire (LIF) neurons with floating gate (FG) technology. The stochastic behaviors have been modeled extensively for both individual neurons and populations of neurons. In the FG LIF neurons, the electron injection is governed by the tunneling process through the gate oxide, leading to the exponential distributions of the injection time and inter spike interval (ISI) stochasticity. The concept of the population coding is demonstrated by simulating the stochastic behaviors of the populations of the FG LIF neurons. The ISI stochasticity enables encoding of the input signals to the population outputs. Spike-to-spike stochasticity improves the signal-to-noise ratio of the population outputs. Moreover, the shape of the ISI distribution can be controlled by adjusting the number of electrons to spike (NES). Exponential-like ISI distributions are realized by reducing the NES. With the exponential-like ISI distributions, the population of fast spiking neurons increases significantly (more than 10% of neurons spiking twice faster than the mean ISI), potentially contributing to the fast computation. Finally, step-by-step procedures have been proposed to design the FG LIF neurons exhibiting the desired neuron characteristics including operation voltage (0.5 V to 3 V), leaky time constant (<1 mu s to >10 ms), ISI mean (in the range of 6 orders of magnitude) and stochasticity (similar to 0 % to similar to 60 %) as well as the type of the distribution (exponential-like to Gaussian-like).

Automated summary

An analytical model for stochastic leaky-integrate-and-fire neurons with floating gate technology has been developed. The study explores the stochastic behaviors and population coding of these neurons. It is found that the shape of the inter spike interval distribution can be controlled, improving the population output signal-to-noise ratio and enabling fast computation.