Compact and voltage-tunable surface plasmon polariton-based optical neural networks

Area of Science:

• Photonics
• Nanotechnology
• Artificial Intelligence

Background:

• Optical neural networks (ONNs) promise efficient parallel processing and low power consumption.
• Current ONN designs struggle with miniaturization, stability, and integration challenges.
• Advanced materials and designs are needed to overcome these limitations.

Purpose of the Study:

• To propose a novel graphene surface plasmon polariton (GSPP) waveguide switch array for all-optical neural networks.
• To demonstrate a compact and tunable platform for simulating synaptic weights.
• To evaluate the performance of the proposed ONN architecture on a benchmark dataset.

Main Methods:

• Design and simulation of a GSPP waveguide switch array with a lateral area of 0.045 μm².
• Numerical analysis of transmission rate tunability across a frequency range of 30.2 to 49.4 THz.
• Implementation and testing of the array for image recognition on the CIFAR-10 dataset.

Main Results:

• Achieved tunable transmission rates from 0 to 0.875, effectively simulating synaptic weights.
• Demonstrated high recognition accuracy of 93.83% on the CIFAR-10 dataset.
• The compact design facilitates integration and scalability for advanced computing.

Conclusions:

• The GSPP waveguide switch array presents a promising solution for miniaturized, stable, and tunable ONNs.
• This technology can pave the way for next-generation high-speed, low-power, and integrated neural network hardware.
• Further research can explore advanced functionalities and larger-scale implementations.