Demonstration of integrate-and-fire neuron circuit for spiking neural networks

Abstract

An integrate-and-fire (IF) neuron and a voltage level shifter circuit were fabricated and investigated for hardware-based SNN architectures. To verify an IF function of neurons, the fabricated IF neuron circuit consists of a synapse, an integration/reset part, and a fire/trigger circuit. By observing a membrane potential of the IF neuron circuit, an integration and reset operations are successfully implemented. In the fabricated IF neuron circuit, the number of output spikes is 2, 5, 10, and 20 at tpulses of 0.4 μs, 1 μs, 2 μs and 4 μs, respectively. The firing rate of the neuron circuit linearly increases as tpulse increases. These measurement results demonstrate that the fabricated IF neuron circuit implements IF function and reset operation well with linear activation function. For suppression of a gate induced drain leakage (GIDL) at high drain voltages (>6 V), the voltage level shifter consists of MOSFETs with a lightly doped drain (LDD). The generation of high voltage pulses (>6 V) for a synaptic weight update is observed through the voltage level shifter circuit.