Bioinspired Quantum Dots Light-Emitting Synapse with Efficient Parallel Processing Capability for High-Precision 3D...

3D reconstruction is crucial in computer vision, especially in medical imaging and human-computer interaction. However, traditional reconstruction methods face challenges like energy inefficiency and memory limitations due to the storage-computation-separated architecture. Neuromorphic devices, inspired by the brain's architecture, offer a solution for efficient data processing. Electrical output synapse devices for 3D reconstruction face delays in coloring point clouds after depth processing, leading to errors. In this work, a co-planar quantum dot (QD) light-emitting synapse is proposed for high-precision 3D reconstruction. By using the light-emitting synapse, handwritten digit recognition achieved 92.35% accuracy in just 20 epochs. Depth and grayscale information are independently processed through electrical and optical outputs, allowing for parallel processing that enhances reconstruction quality. This method decreases losses by 46.3% and reduces the reconstruction pixel error rate by over 21% in comparison to the single output approach. This study demonstrates significant potential of light-emitting synapses in computer vision applications.