Toward Fast Neural Computing using All-Photonic Phase Change Spiking Neurons.

Area of Science:

• Photonics
• Neuromorphic Engineering
• Materials Science

Background:

• The demand for brain-inspired computing is growing, but current CMOS-based neuromorphic systems face limitations in speed, density, and energy efficiency.
• Electronic implementations of neurons and synapses are constrained by switching speed, packing density, and interconnect losses.
• Photonic approaches to neuromorphic engineering are emerging as a promising alternative to overcome these limitations.

Purpose of the Study:

• To propose and demonstrate a purely photonic Integrate-and-Fire Spiking neuron.
• To leverage the phase-change dynamics of Ge2Sb2Te5 (GST) for efficient neuromorphic hardware.
• To explore the integration of these photonic neurons with on-chip synapses for all-photonic neural networks.

Main Methods:

• Utilized Ge2Sb2Te5 (GST) phase-change material integrated with a microring resonator.
• Designed a photonic neuron based on the phase change dynamics of GST.
• Investigated the potential for integration into an all-photonic spiking neural network framework.

Main Results:

• Demonstrated a purely photonic operation of an Integrate-and-Fire Spiking neuron.
• Showcased how GST phase change dynamics can alleviate energy constraints compared to electrical PCMs.
• Proposed a pathway for integrating photonic neurons and synapses for ultrafast neural network inference.

Conclusions:

• The proposed photonic neuron offers a significant advancement in energy efficiency for neuromorphic computing.
• All-photonic spiking neural networks based on this technology promise ultrafast inference and large operating bandwidth.
• This work paves the way for next-generation, high-performance photonic neuromorphic systems.