Highly Stable Artificial Synapses Based on Ferroelectric Tunnel Junctions for Neuromorphic Computing Applications. Issue 7 (13th January 2022)

• Record Type: Journal Article
• Authors: Song, Sungmun; Ham, Woori; Park, Gyuil; Kho, Wonwoo; Kim, Jisoo; Hwang, Hyunjoo; Kim, Hyo‐Bae; Song, Hyunsun; Ahn, Ji‐Hoon; Ahn, Seung‐Eon
• Is Part Of: Advanced materials technologies. Volume 7:Issue 7(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-01-13
• Subjects: synaptic devices -- HfO 2 -- ferroelectricity -- FTJ -- neuromorphic computing; Materials science -- Periodicals; Technological innovations -- Periodicals; Materials science; Technological innovations; Periodicals; 620.1105
• DOI: 10.1002/admt.202101323
• Languages: English
• ISSNs: 2365-709X

Abstract

Abstract: Owing to the limited processing speed and power efficiency of the current computing method based on the von Neumann architecture, research on artificial synaptic devices for implementing neuromorphic computing capable of parallel computation is accelerating. The potential application of artificial synapses composed of ferroelectric tunnel junctions based on metal–hafnium zirconium oxide–metal structure for neuromorphic computing is investigated. Multiple resistance levels are implemented through partial polarization switching control, and synaptic plasticity is successfully imitated based on a high level of device stability and reproducibility. In addition, this device exhibits linear symmetric long‐term potentiation and long‐term depression using a highly variable pulse driving scheme. Finally, the artificial neural network applied with this synaptic device shows high classification accuracy (95.95%) for the Mixed National Institute of Standards and Technology handwritten digits. Abstract : Synaptic functionalities are demonstrated using ferroelectric tunnel junction devices based on HfZrO2 ferroelectrics. Multilevel conductance states are programmed by controlling the ferroelectric partial polarization switching via various input voltage pulses. A simulated artificial neural network based on a synaptic device with excellent stability and retention characteristics exhibits a classification accuracy of 95.95% for the Mixed National Institute of Standards and Technology handwritten digits.