A physical neural network training approach toward multi-plane light conversion design

TL;DR

This paper introduces a physical neural network approach for designing multi-plane light converters, enabling flexible, comprehensive optimization of complex photonic structures supporting hundreds of modes.

Contribution

It presents a novel PNN-based method that models light propagation and phase modulation, allowing optimization of all design parameters simultaneously, surpassing traditional MPLC design techniques.

Findings

Supports high-order MPLC design with 45 modes using mini batches

Enables tuning of training hyperparameters like learning rate and batch size

Demonstrates insensitivity to input and target mode numbers during training

Abstract

Multi-plane light converter (MPLC) designs supporting hundreds of modes are attractive in high-throughput optical communications. These photonic structures typically comprise >10 phase masks in free space, with millions of independent design parameters. Conventional MPLC design using wavefront matching updates one mask at a time while fixing the rest. Here we construct a physical neural network (PNN) to model the light propagation and phase modulation in MPLC, providing access to the entire parameter set for optimization, including not only profiles of the phase masks and the distances between them. PNN training supports flexible optimization sequences and is a superset of existing MPLC design methods. In addition, our method allows tuning of hyperparameters of PNN training such as learning rate and batch size. Because PNN-based MPLC is found to be insensitive to the number of input and target modes in each training step, we have demonstrated a high-order MPLC design (45 modes) using mini batches that fit into the available computing resources.