Highly scalable In_0.95Ga_0.05As based controllable leaky-integrate-fire neuron for high performance spiking neural network and applications

Abstract In this work, we design and simulate a leaky integrate-and-fire (LIF) neuron based on a single Indium Gallium Arsenide (In 0.95 Ga 0.05 As) MOSFET. The proposed implementation results in significant improvements in energy, area, and cost reduction. The In0.95Ga0.05As -MOSFET LIF neuron imitates neuron behaviour accurately, with low breakdown voltage, high impact ionization coefficient, and sharp breakdown. Due to the storage of induced holes in the potential well of the In 0.95 Ga 0.05 As -MOSFET, impact ionisation is the main mechanism creating the spiking characteristic in this case. It requires only 7.11 pJ/spike of energy, compared to the silicon-based SOI MOSFET, which requires 45 pJ/spike. The LIF neuron exhibits a 0.5 MHz high spiking frequency, which is 5 times higher than real nerves in the human body. Compared to biology, MHz operation provides appealing hardware acceleration. In0.95Ga0.05As -MOSFET LIF neuron firing may be controlled through the application of gate voltage, which can increase the spiking neural networks (SNN) energy efficiency by causing sparse activity. The effects of gate metal work function, Gallium mole fraction, and temperature on spike current variations are also investigated. This neuron circuit has promising potential for large-scale integration in spiking neural network hardware. Reconfigurable threshold logic gates have been investigated for the proposed neuron in order to implement universal digital logic functions, such as NAND and NOR.