Neural morphology perception system based on antiferroelectric AgNbO3 neurons

Neural morphology perception system based on antiferroelectric AgNbO3 neurons

Abstract Biologically inspired neuromorphic perceptual systems have great potential for efficient processing of multisensory signals from the physical world. Recently, artificial neurons constructed by memristor have been developed with good biological plausibility and density, but the filament‐type...

Full description

Saved in:
Bibliographic Details
Main Authors: Jianhui Zhao, Jiacheng Wang, Jiameng Sun, Yiduo Shao, Yibo Fan, Yifei Pei, Zhenyu Zhou, Linxia Wang, Zhongrong Wang, Yong Sun, Shukai Zheng, Jianxin Guo, Lei Zhao, Xiaobing Yan
Format: Article
Language:English
Published: Wiley 2025-03-01
Series:InfoMat
Subjects:
antiferroelectric memristor
antiferroelectric neuron
neuromorphic perception system
spiking neural networks
Online Access:https://doi.org/10.1002/inf2.12637
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Abstract Biologically inspired neuromorphic perceptual systems have great potential for efficient processing of multisensory signals from the physical world. Recently, artificial neurons constructed by memristor have been developed with good biological plausibility and density, but the filament‐type memristor is limited by undesirable temporal and spatial variations, high electroforming voltage and limited reproducibility and the Mott insulator type memristor suffer from large driving current. Here, we propose a novel antiferroelectric artificial neuron (AFEAN) based on the intrinsic polarization and depolarization of AgNbO3 (ANO) antiferroelectric (AFE) films to address these challenges. The antiferroelectric memristor exhibits low power consumption (8.99 nW), excellent durability (~105) and high stability. Using such an AFEAN, a spike‐based antiferroelectric neuromorphic perception system (AFENPS) has been designed, which can encode light level and temperature signals into spikes, and further construct a spiking neural network (SNN) (784 × 196 × 10) for optical image classification and thermal imaging classification, achieving 95.34% and 95.76% recognition accuracy on the MNIST dataset, respectively. This work paves the way for the simulation of spiking neurons using antiferroelectric materials and promising a promising method for the development of highly efficient hardware for neuromorphic perception systems.
ISSN:2567-3165

Similar Items