Energy-Efficient ReS₂-Based Optoelectronic Synapse for 3D Object Reconstruction and Recognition

Area of Science:

• Materials Science
• Computer Science
• Electrical Engineering

Background:

• Traditional machine vision systems struggle with 2D image processing limitations, leading to spatial cognition errors and low-precision interpretation.
• Existing neuromorphic vision systems (NVS) lack stereo vision, hindering 3D object recognition capabilities.

Purpose of the Study:

• To develop an NVS with stereo vision for accurate 3D object recognition, inspired by the human visual system.
• To utilize ReS2 optoelectronic synapses for low-power, high-performance 3D vision applications.

Main Methods:

• Development of a ReS2 optoelectronic synapse exhibiting persistent photoconductivity and wavelength-dependent plasticity.
• Integration of color planar and depth information capture, leveraging synapse's distance-dependent channel conductance.
• Reconstruction of 3D object images through fusion of planar and depth data.

Main Results:

• The ReS2 optoelectronic synapse demonstrated ultralow power consumption (12.12 fJ) and excellent synaptic plasticity.
• Successful discrimination and in-situ storage of color planar information based on wavelength-dependent plasticity.
• Achieved 97.0% recognition accuracy for 3D objects, significantly outperforming 2D object recognition (32.6%).

Conclusions:

• The developed 3D-NVS based on ReS2 synapses effectively enables 3D object recognition with high accuracy.
• The system demonstrates robustness against 2D photo-spoofing, suitable for security applications like facial recognition.
• This advancement paves the way for more sophisticated machine vision systems capable of complex spatial cognition.