Single Germanium MOSFET-Based Low Energy and Controllable Leaky Integrate-and-Fire Neuron for Spiking Neural Networks

In this work, a single transistor based on germanium (Ge) is used to construct a leaky integrate-and-fire (LIF) neuron with significant improvement in energy efficiency, area efficiency, and reduction in cost. Using 2-D calibrated simulation, we validated that Ge-MOSFET LIF neuron is able to imitate...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on electron devices 2022-08, Vol.69 (8), p.4265-4270
Main Authors: Khanday, Mudasir A., Bashir, Faisal, Khanday, Farooq A.
Format: Article
Language:English
Subjects:
Behavioral sciences
Breakdown
Electric potential
Germanium
Germanium (Ge)
Integrated circuit modeling
Ionization coefficients
leaky integrate-and-fire (LIF) neuron
Logic gates
Mathematical models
MOSFET
MOSFETs
Neural networks
Neurons
Semiconductor process modeling
Silicon
Spiking
spiking neural network (SNN)
Transistors
Voltage
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In this work, a single transistor based on germanium (Ge) is used to construct a leaky integrate-and-fire (LIF) neuron with significant improvement in energy efficiency, area efficiency, and reduction in cost. Using 2-D calibrated simulation, we validated that Ge-MOSFET LIF neuron is able to imitate the neuron behavior accurately. The Ge-MOSFET shows low breakdown voltage, high impact ionization coefficient, and sharp breakdown. All these factors are responsible for achieving low energy per spike and higher spiking current. The proposed Ge-MOSFET-based spiking LIF neuron needs only 8 pJ/spike of energy as compared to recently reported silicon-based silicon-on-insulator (SOI) MOSFET, which needs 45 pJ/spike of energy. The use of gate voltage makes Ge-MOSFET LIF neuron firing controllable, which can improve the energy efficiency of the spiking neural network (SNN) by inducing sparse action.
ISSN:0018-9383
1557-9646
DOI:10.1109/TED.2022.3186274