Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multimode fiber laser cavities as nonlinear optical processors

Dilem Eşlik, Bahadır Utku Kesgin, Fatma Nur Kılınç, Uğur Teğin

TL;DR

This work addresses the challenge of implementing nonlinear mappings for energy-efficient ML by using continuous-wave multimode fiber laser cavities as nonlinear optical processors. Inputs are encoded as phase patterns on an SLM and transformed by nonlinear modal dynamics and gain saturation into high-dimensional, class-discriminative features, enabling a simple linear readout to classify. Numerical and experimental results show accuracies in the range of $85-99\%$, with orders-of-magnitude fewer trainable parameters than CNNs, across diverse benchmarks. The findings suggest a compact, low-power photonic preprocessing approach with potential routes to higher throughput and chip-scale integration for scalable optical processing.

Abstract

Optical computing provides a promising path toward energy-efficient machine learning, yet implementing nonlinear transformations without complex electronics or high-power sources remains challenging. Here, we demonstrate that continuous-wave multimode fiber laser cavities can function as nonlinear optical processors. Input images encoded as phase patterns on a spatial light modulator undergo high-dimensional transformation through the interplay of multimode interference and gain saturation dynamics. The cavity maps input data into spatially stable, class-separable intensity distributions, enabling a simple linear classifier to achieve accuracies of 85--99\% across diverse benchmarks -- including medical imaging and remote sensing -- with orders of magnitude fewer trainable parameters than deep neural networks. Our results establish multimode fiber lasers as compact, low-power physical processors for scalable optical machine learning.

Multimode fiber laser cavities as nonlinear optical processors

TL;DR

This work addresses the challenge of implementing nonlinear mappings for energy-efficient ML by using continuous-wave multimode fiber laser cavities as nonlinear optical processors. Inputs are encoded as phase patterns on an SLM and transformed by nonlinear modal dynamics and gain saturation into high-dimensional, class-discriminative features, enabling a simple linear readout to classify. Numerical and experimental results show accuracies in the range of , with orders-of-magnitude fewer trainable parameters than CNNs, across diverse benchmarks. The findings suggest a compact, low-power photonic preprocessing approach with potential routes to higher throughput and chip-scale integration for scalable optical processing.

Abstract

Optical computing provides a promising path toward energy-efficient machine learning, yet implementing nonlinear transformations without complex electronics or high-power sources remains challenging. Here, we demonstrate that continuous-wave multimode fiber laser cavities can function as nonlinear optical processors. Input images encoded as phase patterns on a spatial light modulator undergo high-dimensional transformation through the interplay of multimode interference and gain saturation dynamics. The cavity maps input data into spatially stable, class-separable intensity distributions, enabling a simple linear classifier to achieve accuracies of 85--99\% across diverse benchmarks -- including medical imaging and remote sensing -- with orders of magnitude fewer trainable parameters than deep neural networks. Our results establish multimode fiber lasers as compact, low-power physical processors for scalable optical machine learning.
Paper Structure (9 sections, 3 equations, 5 figures, 1 table)

This paper contains 9 sections, 3 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Results
  3. Numerical validation
  4. Experimental demonstration
  5. Discussion
  6. Methods
  7. Cavity model and simulation
  8. Experimental setup
  9. Data encoding and classification

Figures (5)

  • Figure 1: Schematic of the multimode fiber laser computing platform. The cavity employs a linear geometry with a Sagnac loop, consisting of a Yb-doped gain fiber and a passive graded-index (GRIN) multimode fiber (MMF). A spatial light modulator (SLM) functions as a programmable end mirror to encode information, while a 50/50 coupler forms a feedback loop with output extraction.
  • Figure 2: Numerical simulation results. a) Representative SLM phase pattern (left) and corresponding laser output intensity (right). b,c) Confusion matrices for the UCI Glass dataset using raw features (b) and optically processed features (c).
  • Figure 3: Experimental classification results. a,b) Confusion matrices for TrashNet using raw images (a) and optically processed images (b). c,d) Confusion matrices for RSSCN7 using raw images (c) and optically processed images (d).
  • Figure 4: Medical imaging classification results. a,b) Confusion matrices for OCT MNIST using raw images (a) and optically processed images (b). c,d) Confusion matrices for HAM10000 using raw images (c) and optically processed images (d).
  • Figure 5: Feature space visualization via Linear Discriminant Analysis. a,b) LDA projections for OCT MNIST using raw images (a) and optically processed images (b). c,d) LDA projections for RSSCN7 using raw images (c) and optically processed images (d).