Optical synaptic devices with ultra-low power consumption for neuromorphic computing

Area of Science:

• Materials Science
• Optoelectronics
• Computer Engineering

Background:

• Neuromorphic computing offers energy-efficient parallel processing for massive data.
• Photonic synapses are crucial for artificial neural systems but face high power consumption challenges.

Purpose of the Study:

• To develop an artificial photonic synapse with ultra-low power consumption.
• To demonstrate its potential in energy-efficient neuromorphic vision systems.

Main Methods:

• Fabrication of a BP/CdS heterostructure-based artificial photonic synapse.
• Characterization of its optoelectronic properties, including negative light response and responsivity.
• Simulation of a fully-connected optoelectronic neural network (FONN) for image recognition.

Main Results:

• The device exhibited a remarkable negative light response with high responsivity (up to 4.1 × 10^8 A/W).
• Achieved ultra-low average power consumption of 4.78 fJ per training process, the lowest reported.
• Simulated FONN achieved a maximum image recognition accuracy of 94.1%.

Conclusions:

• The BP/CdS heterostructure provides a novel concept for designing energy-efficient artificial photonic synapses.
• This technology holds significant potential for advancing high-performance neuromorphic vision systems.