Demultiplexing through a multimode fiber using chip-scale diffractive neural networks

TL;DR

This paper demonstrates for the first time a purely optical, chip-scale AI solution for high-mode isolation, speed-of-light demultiplexing of MMF modes using a three-dimensional diffractive neural network (DNN).

Abstract

In today's information age, advanced fiber optic transmission technology is of paramount importance. Multimode fibers (MMFs) using space-division multiplexing (SDM) are promising for improved transmission capacity, connection flexibility, and security of data. However, the complex transmission characteristics of MMFs significantly hinder precise mode demultiplexing. Conventional approaches, including holographic measurements, phase retrieval algorithms, photonic lanterns, and multiplane light conversion, are limited by system complexity, size, and flexibility. In this paper, we demonstrate for the first time a purely optical, chip-scale AI solution for high-mode isolation, speed-of-light demultiplexing of MMF modes using a three-dimensional diffractive neural network (DNN). The DNN is trained with synthetic modal data and fabricated using two-photon nanolithography. It features a compact size of $120μm \times 120μm \times 80μm$ and a diffractive structure size of $1μm^{2}$ for the neurons at the hidden layers of the network. Experimentally, the DNN demultiplexer achieves a relative demultiplexing accuracy of over 80%. The AI approach of DNN allows for flexible design and overcomes the size and performance limitations of digital-optical demultiplexers. This work paves the way for compact, low-latency optical processors for high-performance demultiplexers and enables scalable, chip-integrated solutions for next-generation fiber optic networks.

Figures (4)

• Figure 1: (a) Schematic illustration of demultiplexing MMF using DNN. Pure eigenmodes are launched into the MMF, while the output becomes a speckle field due to coherent superposition. A six-mode example is shown here. The DNN separates the speckle field into individual mode channels based on their respective modal contributions. The modal energy can then be directly measured using photodetectors. (b) Schematic of a DNN-based mode demultiplexer. The input of the DNN is the superposed complex light field emerging from MMF. The DNN consists of several diffractive layers that shape the input into the corresponding detector area. The detector layer contained 6 different detection regions corresponding to 6 different modes. The relative intensities of light in different detection regions can be measured using either normal light detectors or single-photon detectors, facilitating the transmission of high-dimensional information. The forward propagation of the DNN is done by ASM, and the back propagation is used to train the diffractive layers.
• Figure 2: Performance evaluation of DNN on synthetic data. (a) Training process of DNN. (b) Phase distributions of three diffractive layers in the 3-layer DNN. (c) Visualization of the light propagation inside DNN for mode $LP_{11a}$. (d) The predicted output field of DNN for all six modes. (e) Mode isolation performance under different DNN depths. Here, the DNN varies from 2 to 5 layers. (f) Mode isolation across different optical wavelengths. Here, the wavelength varies from 1500nm to 1650nm.
• Figure 3: Performance evaluation of DNN on synthetic superposed data. (a) 3 different superposed light fields are used as inputs to the DNN. The output of the DNN and the corresponding label are also given. The bar chart depicts the amplitude difference across different modes. (b) Relative amplitude weights error of the DNN trained with different datasets under different DNN depths. (c) Relative amplitude weight error across different optical wavelengths.
• Figure 4: Experimental results of using 3D printed DNN. (a) Experimental fabrication of the DNN chip for MMF modes demultiplexing and detection. SEM images of the three-layer DNN and the microscopy image of MMF and DNN are shown. (b) A schematic diagram of the experimental setup. Abberations: MO, microscope objective, CAM. camera. (c),(d) Experimental results of two mutually perpendicular polarizations. The intensity value within each detection area represents the power ratio among different modes. The amplitude weights are then obtained by taking the square root of the intensity value. Each sample is independently normalized.