InGaZnO Optoelectronic Synaptic Transistor for Reservoir Computing and LSTM‐Based Prediction Model DOI
Suyong Park, Seong‐Min Kim,

Sungjoon Kim

et al.

Advanced Optical Materials, Journal Year: 2025, Volume and Issue: unknown

Published: May 12, 2025

Abstract This study presents a reservoir computing (RC) system utilizing an indium gallium zinc oxide (IGZO)‐based optoelectronic synaptic transistor (OST) for neuromorphic applications. The proposed IGZO‐based OST harnesses the effects of persistent photoconductivity in IGZO channel and charge trapping at IGZO/tantalum interface to emulate short‐term behavior. By optical stimuli, device achieves dynamic states with nonlinear time‐dependent characteristics, enhancing its capability temporal data processing. Moreover, effectively performs pattern recognition tasks, attaining high classification accuracies 95.75% 85.02% on MNIST Fashion datasets, respectively. Additionally, replicates nociceptive behaviors, such as allodynia hyperalgesia, under stimulation, showcasing potential bio‐inspired sensory An LSTM‐based prediction model is developed using Jena climate data, incorporating method that mimics weight variation assess impact performance. approach demonstrates feasibility hardware‐friendly neural networks via biologically inspired adjustments, outperforming conventional forecasting models. Notably, normalized root mean square error (NRMSE) low 0.0145, highlighting accuracy.

Language: Английский

InGaZnO Optoelectronic Synaptic Transistor for Reservoir Computing and LSTM‐Based Prediction Model DOI
Suyong Park, Seong‐Min Kim,

Sungjoon Kim

et al.

Advanced Optical Materials, Journal Year: 2025, Volume and Issue: unknown

Published: May 12, 2025

Abstract This study presents a reservoir computing (RC) system utilizing an indium gallium zinc oxide (IGZO)‐based optoelectronic synaptic transistor (OST) for neuromorphic applications. The proposed IGZO‐based OST harnesses the effects of persistent photoconductivity in IGZO channel and charge trapping at IGZO/tantalum interface to emulate short‐term behavior. By optical stimuli, device achieves dynamic states with nonlinear time‐dependent characteristics, enhancing its capability temporal data processing. Moreover, effectively performs pattern recognition tasks, attaining high classification accuracies 95.75% 85.02% on MNIST Fashion datasets, respectively. Additionally, replicates nociceptive behaviors, such as allodynia hyperalgesia, under stimulation, showcasing potential bio‐inspired sensory An LSTM‐based prediction model is developed using Jena climate data, incorporating method that mimics weight variation assess impact performance. approach demonstrates feasibility hardware‐friendly neural networks via biologically inspired adjustments, outperforming conventional forecasting models. Notably, normalized root mean square error (NRMSE) low 0.0145, highlighting accuracy.

Language: Английский

Citations

0