MRS Meetings and Events


EL21.12.04 2023 MRS Spring Meeting

Novel and Tunable Wavelength-Selective Artificial Synapses Based on Inorganic Thin Film Photovoltaic Devices

When and Where

Apr 13, 2023
4:15pm - 4:30pm

Moscone West, Level 3, Room 3011



Zacharie Jehl Li-Kao1,Kunal Tiwari1,Axel Gon Medaille1,Alex Jimenez1,Eloi Costals1,Sergio Giraldo1,Marcel Placidi1,Edgardo Saucedo1

Polytechnic University of Catalonia1


Zacharie Jehl Li-Kao1,Kunal Tiwari1,Axel Gon Medaille1,Alex Jimenez1,Eloi Costals1,Sergio Giraldo1,Marcel Placidi1,Edgardo Saucedo1

Polytechnic University of Catalonia1
Artificial synapses are an important building block for neuromorphic computing, holding the promise to reproduce fundamental brain functions and markedly improve the energy efficiency of future computers beyond the Von Neumann architecture while offering new possibilities in terms of unsupervised learning and parallel computing with a high resilience to faults. As the human brain receives nearly 80% of its inputs from visual perception, the development of efficient and tunable optoelectronic artificial synapses appears as the next frontier going forward, and several proof-of-concept devices have been reported using both organic and inorganic materials, albeit with a comparatively low degree of tunability. While carrier trapping from native and extrinsic defects is often seen as a drawback in inorganic thin film photovoltaic devices, it holds a remarkable yet currently untapped potential to simulate synaptic plasticity.<br/>In this work, we demonstrate for the first time the properties of <b>long term potentiation, short term potentiation, and synaptic depression using light as a stimulus</b> and persistent photoconductivity as a figure of merit to characterize the plasticity in a Cu<sub>2</sub>ZnSnSe<sub>4</sub>/CdS/ZnO-based photovoltaic device artificial synapse. The dark persistent photoconductivity was found to last for <b>several days after a stimulation</b> with white light ranging from durations of a few minutes to several hours and at a power of 1000 W.m<sup>-2</sup>. Furthermore, illumination in short bursts (few seconds and less) reveals a typical <b>short term plasticity behavior with a memory effect akin to that of an RC integrator</b>, a key element for digital computing. It is to our knowledge the first time that such behaviors are deliberately achieved and characterized for this class of materials and devices.<br/>Even more remarkably, the fabricated devices are found to operate with some level of <b>wavelength selectivity</b>. By maintaining the incident illumination power constant and varying the illumination wavelength with 28 different LEDs over a range from 350 nm up to 1500 nm, significant changes in the synaptic response are observed which widens the number of possible programming states of the system, and relates to the different transition energies of the trapping states involved in the plasticity of the device. The response to <b>ultraviolet and blue illuminations below 500 nm is particularly efficient at producing both a long term and short term potentiation of the synapse</b>, hinting at the charging of trap states located at the vicinity of the front p-n interface. In contrast, most of the wavelengths from the visible spectrum do not lead to a measurable short term plasticity response, which could be interpreted as either the absence of transition corresponding to those wavelengths or to a screening of the photoconductivity by the photocurrent. The case of infrared illumination beyond 1200 nm is particularly peculiar as it <b>simultaneously can lead to a short term synaptic potentiation following illuminations of a few seconds or less, and a long term synaptic depression following an illumination of fifteen minutes</b> of an already potentiated synapse. Such reaction to a single stimulus has to our knowledge never be observed in either electronic or optoelectronic synapses.<br/>Finally, it is found that the synaptic behavior of the devices can be tuned through the variation of the oxygen content in the ZnO window layer, offering further insights on the defect trapping mechanics involved in the plasticity of those artificial synapses. The complete set of results will be presented, including the current limitations of this work, and a model involving two trap levels located at the vicinity of the p-n interface will be proposed and discussed in the context of the electrical characterizations of the artificial synapses and previous knowledge acquired in CZTSe-based solar cells. This type of synapses represents an entirely novel research opportunity for ultra-low energy computing.


defects | thin film

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

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

Raith America, Inc.

Publishing Alliance

MRS publishes with Springer Nature