On-Chip Photonic Synapses with All-Optical Memory and Neural Network Computation

Area of Science:

• Optoelectronics
• Materials Science
• Computational Neuroscience

Background:

• Traditional computing faces bottlenecks, driving research into brain-inspired neural network computing.
• Integrated design of artificial neural networks remains a significant challenge.

Purpose of the Study:

• To demonstrate a readily integrated membrane-system photonic synapse.
• To explore the tunable excitatory and inhibitory functions of the photonic synapse.
• To showcase the application of photonic synapses in neural network functions and digital recognition.

Main Methods:

• Development of a membrane-system photonic synapse.
• Pre-pulse training at 1064 nm for device regulation.
• Operation and testing of the photonic synapse at tunable wavelengths (1200-2000 nm).
• Integration of photonic synapses into a network to demonstrate memory and weight functions.
• Construction and testing of a single-layer perceptron neural network using photonic synapses.

Main Results:

• A readily integrated membrane-system photonic synapse was successfully demonstrated.
• The photonic synapse exhibited tunable excitatory and inhibitory functions.
• The device showed memory functions and multiple weight capabilities within an integrated network.
• Digital recognition was successfully achieved using a single-layer perceptron neural network built with these photonic synapses.

Conclusions:

• The demonstrated photonic synaptic network effectively replicates most biological synaptic functions.
• The photonic synapse offers advantages such as a compact structure, scalability, and adjustable wavelength.
• This work presents a novel approach for the study and development of neural synaptic networks.