1T Spiking Neuron Using Ferroelectric Junctionless FET with Ultra-Low Energy Consumption of 24 aJ/Spike

• Published in: Neural processing letters Vol. 55; no. 8; pp. 11527 - 11539
• Main Authors: Khanday, Mudasir A., Rashid, Shazia, Khanday, Farooq A.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01-12-2023; Springer Nature B.V
• Subjects: Artificial Intelligence; Circuits; Complex Systems; Computational Intelligence; Computer Science; Current carriers; Design; Electric fields; Energy consumption; Ferroelectric materials; Ferroelectricity; Field effect transistors; Image classification; Neural networks; Neuromorphic computing; Neurons; Physiology; Semiconductor devices; Transistors; Ferroelectric junctionless FET; SNN; 1T LIF neuron; Neuromorphic computing; Image classification
• ISSN: 1370-4621; 1573-773X
• DOI: 10.1007/s11063-023-11387-x

In view of the soaring demand of highly scalable and energy-efficient neuron devices for future neuromorphic computing, this work demonstrates a novel double-gate ferroelectric junctionless field effect transistor (DG-FE-JLFET) neuron in 20 nm technology node. The proposed device has a homogenously doped structure with GaAs as channel material and HfZrO
2
as gate oxide. Using the calibrated simulations in Atlas TCAD, it is confirmed that the proposed neuron accurately mimics the spiking behavior of the biological neuron with an energy consumption of 24 aJ/spike, which is ~ 10
5
folds lesser than the previously proposed single transistor neurons. The proposed design does not need additional circuitry to operate. This greatly simplifies design complexity and can also achieve higher neuron density which is important for designing large scale integration neuromorphic chips. In contrast to the previously reported JLFET neurons that work on impact ionization phenomenon, DG-FE-JLFET works on tunnelling mechanism that eradicates the need to create virtual potential well to accommodate charge carriers in the body. Finally, to validate the practical applicability, the proposed neuron has been explored for image classification with an accuracy of 93.28%.