Dual-Wavelength Synaptic Simulator ReS₂/Ta₂NiSe₅ for Multi-Timescale Learning in Neuromorphic Computing

Area of Science:

• Materials Science
• Neuromorphic Engineering
• Optoelectronics

Background:

• Silicon-based devices face limitations in neuromorphic computing.
• Conventional single-wavelength devices lack multi-timescale adaptability.
• Novel materials and device architectures are needed for advanced AI.

Purpose of the Study:

• To propose a dual-wavelength photomodulated synaptic device for enhanced neuromorphic computing.
• To achieve wavelength-selective synaptic plasticity using a ReS₂/Ta₂NiSe₅ heterojunction.
• To enable multi-timescale learning for improved AI performance.

Main Methods:

• Fabrication of a ReS₂/Ta₂NiSe₅ heterojunction device.
• Utilizing photothermal effect (1550 nm) and photoelectric effect (520 nm) for synaptic plasticity.
• Implementing dual-wavelength cooperative learning strategy.
• Training a convolutional neural network (CNN) on the Extended MNIST (EMNIST) dataset.

Main Results:

• Achieved distinct relaxation timescales for synaptic weight updates (≈17 s at 1550 nm, ≈6 s at 520 nm).
• Demonstrated wavelength-selective synaptic plasticity and multi-timescale learning.
• Improved training efficiency and generalization in CNNs.
• Attained a final accuracy of 95.5% on the EMNIST dataset.

Conclusions:

• The dual-wavelength heterojunction photonic synaptic device offers a scalable approach for adaptive neuromorphic computing.
• This technology enables precise control over synaptic weight adaptation for multi-timescale learning.
• The findings pave the way for future photonic artificial intelligence (AI) architectures.