Photonic neuromorphic computing using symmetry-protected zero modes in coupled nanolaser arrays

Area of Science:

• Optoelectronics
• Artificial Intelligence
• Materials Science

Background:

• Photonic neuromorphic computing offers energy-efficient artificial neural networks (ANNs).
• Nanolasers are attractive for ANNs due to low power and nonlinear properties.
• Symmetry-protected zero modes in coupled nanolaser arrays are proposed for computation.

Purpose of the Study:

• To propose and demonstrate a photonic neuromorphic computing architecture.
• To leverage symmetry-protected robust zero modes in nanolaser arrays for computation.
• To showcase the classification capabilities of coupled nanolasers.

Main Methods:

• Proposed a photonic neuromorphic computing architecture using coupled semiconductor nanolaser arrays.
• Utilized symmetry-protected robust zero modes at the center of the optical spectrum.
• Experimentally demonstrated classification tasks with a 2x2 nanolaser array.

Main Results:

• A small set of coupled nanolasers demonstrated inherent non-convex classification capabilities.
• A 2x2 nanolaser array solved the XNOR logical gate, acting as a nonlinear recurrent layer.
• Robust classification performance was achieved even with highly compressed handwritten digits.

Conclusions:

• Symmetry-protected modes in nanolaser arrays enable robust optical connections for complex problem-solving.
• This approach offers an energy-efficient alternative to traditional artificial neural networks.
• The findings suggest potential for tackling complex computations without scaling neuron numbers.