MRS Meetings and Events

 

EL21.08.06 2023 MRS Spring Meeting

Analog Synaptic Characteristics of CeO2-based Memristor Crossbar Array for Neuromorphic Computing

When and Where

Apr 12, 2023
5:00pm - 7:00pm

Moscone West, Level 1, Exhibit Hall

Presenter

Co-Author(s)

Kitae Park1,Jiyeon Ryu1,Dwipak Sahu1,Tae-Sik Yoon1

Ulsan National Institute of Science and Technology1

Abstract

Kitae Park1,Jiyeon Ryu1,Dwipak Sahu1,Tae-Sik Yoon1

Ulsan National Institute of Science and Technology1
Artificial synapses with analog resistive switching devices have been actively researched since they are essential elements for brain-inspired neuromorphic systems as promising computing systems particularly for data-centric applications such as artificial intelligence.<sup>[1] </sup>In this study, the CeO<sub>2</sub>-based memristor was fabricated as 32x32 crossbar array structure. Two-terminal of access lines with Pt electrodes were deposited by e-beam evaporation and patterned with 10 µm width by photolithography and lift-off process and the switching layer (CeO<sub>2</sub>) was deposited by RF magnetron sputtering. The synapse array exhibited analog, gradual, and voltage-polarity dependent resistive switching, which emulates biological synaptic behaviors of synaptic weight updates. The conductance of the device in array, corresponding to synaptic weight, increased upon consecutive positive voltage pulses and reversibly decreased upon negative pulses. In addition, its non-linear current-voltage characteristics matched well with Schottky conduction, which efficiently reduced sneak current in array architecture working as Schottky diode selector at the same time. Moreover, several electrical properties analogous to those of biological synapse were observed such as paired-pulse facilitation, spike rate-dependent plasticity. Time-dependent conductance decay could also be fitted well with multiple exponential decay functions that could separately assess short-term memory and long-term memory, and the retention at long-term memory turned out to be ~55 % after 1 hour. The resistive changing is thought to be induced by redistribution of oxygen vacancies in CeO<sub>2</sub> switching layer, particularly at the interface with electrodes. X-ray photoelectron spectroscopy analysis unveiled that CeO<sub>2</sub> layer has considerable amounts of oxygen vacancies to induce resistive switching. Furthermore, memristor with NbO<sub>x</sub>/CeO<sub>2</sub> bilayer exhibited enhanced resistive switching dynamic range and non-linearity. This is because NbO<sub>x</sub> layer worked as not only an oxygen vacancies reservoir, but also threshold switching selector.<sup>[2] </sup>This study demonstrated that analog synaptic CeO<sub>2</sub>-based memristor crossbar array could be promising synaptic elements for neuromorphic computing.<br/><br/>[1] K. Park, P. H. Chung, D. P. Sahu, and T.-S. Yoon, Mater. Sci. Semicond. Process. 147, 106718 (2022).<br/>[2] K. Park, J. Ryu, D. P. Sahu, and T.-S. Yoon<i>, </i>RSC Adv. 12, 18547 (2022).

Symposium Organizers

Iuliana Radu, Taiwan Semiconductor Manufacturing Company Limited
Heike Riel, IBM Research GmbH
Subhash Shinde, University of Notre Dame
Hui Jae Yoo, Intel Corporation

Symposium Support

Gold
Center for Sustainable Energy (ND Energy) and Office of Research

Silver
Raith America, Inc.

Publishing Alliance

MRS publishes with Springer Nature