Artificial optoelectronic synapse based on spatiotemporal irradiation to source‐sharing circuitry of synaptic phototransistors

• Published in: InfoMat Vol. 6; no. 2
• Main Authors: Song, Seungho, Choi, Changsoon, Ahn, Jongtae, Lee, Je‐Jun, Jang, Jisu, Yu, Byoung‐Soo, Hong, Jung Pyo, Ryu, Yong‐Sang, Kim, Yong‐Hoon, Hwang, Do Kyung
• Format: Journal Article
• Language: English
• Published: Melbourne John Wiley & Sons, Inc 01.02.2024; Wiley
• Subjects: amorphous oxide semiconductor; analog processing; Artificial neural networks; artificial synapse; Circuits; Data processing; Electronic components; Light; Machine learning; Mathematical analysis; Matrices (mathematics); Neural networks; neuromorphic; Neuromorphic computing; Optical pulses; Optoelectronics; Phototransistors; Synapses
• ISSN: 2567-3165; 2567-3165
• DOI: 10.1002/inf2.12479

To overcome the intrinsic inefficiency of the von Neumann architecture, neuromorphic devices that perform analog vector–matrix multiplication have been highlighted for achieving power‐ and time‐efficient data processing. In particular, artificial synapses, of which conductance should be programmed to represent the synaptic weights of the artificial neural network, have been intensively researched to realize neuromorphic devices. Here, inspired by excitatory and inhibitory synapses, we develop an artificial optoelectronic synapse that shows both potentiation and depression characteristics triggered only by optical inputs. The design of the artificial optoelectronic synapse, in which excitatory and inhibitory synaptic phototransistors are serially connected, enables these characteristics by spatiotemporally irradiating the phototransistor channels with optical pulses. Furthermore, a negative synaptic weight can be realized without the need for electronic components such as comparators. With such attributes, the artificial optoelectronic synapse is demonstrated to classify three digits with a high recognition rate (98.3%) and perform image preprocessing via analog vector–matrix multiplication. We demonstrate an artificial optoelectronic synapse array of 64 pixels based on two indium gallium zinc oxide (IGZO) synaptic phototransistors that respectively correspond to excitatory and inhibitory synapses. This synapse can be modulated by the spatiotemporal irradiation of light pulses to each phototransistor, enabling fully optically‐triggered potentiation and depression. Finally, the synapse array is used to classify three digits with a high recognition rate (98.3%) and perform image preprocessing.