Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Optical Spiking Neural Networks via Rogue-Wave Statistics

Bahadır Utku Kesgin, Gülsüm Yaren Durdu, Uğur Teğin

TL;DR

Optical computing aims to reduce energy use by exploiting light's speed and parallelism, but nonlinear activation in low-power optics remains challenging. The authors introduce an optical spiking neural network in which free-space diffraction performs synaptic integration and rogue-wave (caustic) statistics provide a physically grounded firing nonlinearity, with a phase-mask acting as trainable synaptic weights and a differentiable digital twin enabling end-to-end co-design. The system fires when the diffracted intensity satisfies $|E|^2 \ge I_{RW}$ with $I_{RW}=2 I_{sig}$, and demonstrates competitive classification on BreastMNIST (82.45%) and Olivetti Faces (95.00%) using a free-space, SLM-based setup. This work establishes rogue-wave physics as a scalable, energy-efficient computational primitive and suggests paths to temporal-domain and on-chip photonic implementations for ultrafast neuromorphic processing.

Abstract

Optical computing could reduce the energy cost of artificial intelligence by leveraging the parallelism and propagation speed of light. However, implementing nonlinear activation, essential for machine learning, remains challenging in low-power optical systems dominated by linear wave physics. Here, we introduce an optical spiking neural network that uses optical rogue-wave statistics as a programmable firing mechanism. By establishing a homomorphism between free-space diffraction and neuronal integration, we demonstrate that phase-engineered caustics enable robust, passive thresholding: sparse spatial spikes emerge when the local intensity exceeds a significant-intensity rogue-wave criterion. Using a physics-informed digital twin, we optimize granular phase masks to deterministically concentrate energy into targeted detector regions, enabling end-to-end co-design of the optical transformation and a lightweight electronic readout. We experimentally validate the approach on BreastMNIST and Olivetti Faces, achieving accuracies of 82.45\% and 95.00\%, respectively, competitive with standard digital baselines. These results demonstrate that extreme-wave phenomena, often treated as deleterious fluctuations, can be harnessed as structural nonlinearity for scalable, energy-efficient neuromorphic photonic inference.

Optical Spiking Neural Networks via Rogue-Wave Statistics

TL;DR

Optical computing aims to reduce energy use by exploiting light's speed and parallelism, but nonlinear activation in low-power optics remains challenging. The authors introduce an optical spiking neural network in which free-space diffraction performs synaptic integration and rogue-wave (caustic) statistics provide a physically grounded firing nonlinearity, with a phase-mask acting as trainable synaptic weights and a differentiable digital twin enabling end-to-end co-design. The system fires when the diffracted intensity satisfies with , and demonstrates competitive classification on BreastMNIST (82.45%) and Olivetti Faces (95.00%) using a free-space, SLM-based setup. This work establishes rogue-wave physics as a scalable, energy-efficient computational primitive and suggests paths to temporal-domain and on-chip photonic implementations for ultrafast neuromorphic processing.

Abstract

Optical computing could reduce the energy cost of artificial intelligence by leveraging the parallelism and propagation speed of light. However, implementing nonlinear activation, essential for machine learning, remains challenging in low-power optical systems dominated by linear wave physics. Here, we introduce an optical spiking neural network that uses optical rogue-wave statistics as a programmable firing mechanism. By establishing a homomorphism between free-space diffraction and neuronal integration, we demonstrate that phase-engineered caustics enable robust, passive thresholding: sparse spatial spikes emerge when the local intensity exceeds a significant-intensity rogue-wave criterion. Using a physics-informed digital twin, we optimize granular phase masks to deterministically concentrate energy into targeted detector regions, enabling end-to-end co-design of the optical transformation and a lightweight electronic readout. We experimentally validate the approach on BreastMNIST and Olivetti Faces, achieving accuracies of 82.45\% and 95.00\%, respectively, competitive with standard digital baselines. These results demonstrate that extreme-wave phenomena, often treated as deleterious fluctuations, can be harnessed as structural nonlinearity for scalable, energy-efficient neuromorphic photonic inference.
Paper Structure (4 sections, 7 equations, 4 figures, 1 table)

This paper contains 4 sections, 7 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Results
  3. Discussion
  4. Methods

Figures (4)

  • Figure 1: Experimental schematic of rogue wave-based optical spiking neural network. The collimated beam illuminates a reflective phase-only Spatial Light Modulator (SLM), which encodes the complex-valued input data and synaptic weights using a macropixel double-phase encoding scheme. A second calibrated 4-f relay system (L3-L4) demagnifies the diffracted speckle pattern to establish a 1-to-1 spatial correspondence between the SLM computation window and the CMOS detector array, ensuring accurate readout of the rogue wave events.
  • Figure 2: Rogue waves in the presence of amplitude-encoded data. a, Input amplitude distribution encoding the information. b, Phase modulation pattern applied to the SLM. c, Resulting optical intensity distribution at the detector plane after propagation. d, Histogram of the probability density function for the recorded intensity values, illustrating the statistical distribution.e, Optical spikes generated after rogue wave thresholding.
  • Figure 3: Results of binary classification with BreastMNIST dataset.textbfa, Input amplitude distribution encoding the information. b, Optimized phase modulation pattern applied to the SLM. c, Simulated optical spikes generated by the corresponding complex amplitude. d, Confusion matrix of the simulation results.e, Confusion matrix of the experimental results. f, Experimentally measured optical spikes generated by the corresponding complex amplitude.
  • Figure 4: Results of binary classification with Olivetti Faces dataset.textbfa, Input amplitude distribution encoding the information. b, Optimized phase modulation pattern applied to the SLM. c, Simulated optical spikes generated by the corresponding complex amplitude. d, Confusion matrix of the simulation results.e, Confusion matrix of the experimental results. f, Experimentally measured optical spikes generated by the corresponding complex amplitude.