Multistage Space-Frequency Optical Neural Network

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional all-optical diffraction-depth neural networks are limited in complexity and performance for tasks requiring higher nonlinearity, as they primarily perform modulation in the spatial domain with lamination of diffraction layers.

Innovation Solution

A nonlinear all-optical deep-learning system with multistage space-frequency domain modulation, incorporating optical input, multistage space-frequency domain modulation, and information acquisition modules, utilizing frequency-domain and spatial-domain modulation alternately, along with nonlinear optical devices like photorefractive crystals for enhanced processing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If multistage space-frequency domain modulation is implemented, then task complexity and nonlinear processing capability are improved, but device complexity increases

Engineering Contradiction:

Improvetask complexityVSAvoiddevice complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system segments the modulation process into multiple stages, alternating between spatial domain and frequency domain operations. Each stage performs a specific function (spatial modulation, Fourier transformation, frequency modulation, inverse Fourier transformation), breaking down the complex nonlinear processing task into manageable sequential steps that can be implemented with individual optical components.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system transitions between different domain dimensions (spatial domain and frequency domain) through Fourier transformation operations. By switching between these dimensions, the system enhances its processing capability for complex tasks while using standard optical components, effectively adding a dimensional aspect to the computation without proportionally increasing physical device complexity.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

2Speed

If all-optical diffraction-depth neural network is used, then computation speed is improved, but nonlinear processing capability deteriorates

Engineering Contradiction:

Improvecomputation speedVSAvoidnonlinear processing capability

Core Design Contradiction:

SpeedVSAdaptability or versatility

Solution Approach 1:

The system maintains continuous all-optical processing throughout the entire computation pipeline, from input to output, without converting to electrical domain. This ensures computation speed remains at the speed of light while incorporating nonlinear optical devices that provide the necessary nonlinear processing capability for complex tasks.

Inventive Principle:

Principle #20Continuity of useful action

Solution Approach 2:

The system uses composite optical structures including photorefractive crystals and other nonlinear optical materials combined with standard diffraction elements. These composite structures provide both the speed of optical processing and the nonlinear processing capability needed for complex machine learning tasks.

Inventive Principle:

Principle #40Composite materials

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

This approach enables the completion of complex machine learning tasks and nonlinear computations at the speed of light, with improved performance and efficiency, allowing for large-scale neural network implementation with low power consumption and extensibility.

Implementation Method 1

a first lens is configured to perform a first Fourier transformation on the modulated optical information so as to transform the modulated optical information onto a Fourier plane

Methodology Applied

Scientific EffectFourier transformation:

Implementation Method 2

an all-optical diffraction-depth neural network provides an effective and unique all-optical machine-learning model for implementing a diffraction operation at the speed of light by using passive elements

Methodology Applied

Scientific EffectDiffraction: Diffraction

Implementation Method 3

incorporating optical input, multistage space-frequency domain modulation, and information acquisition modules, utilizing frequency-domain and spatial-domain modulation alternately, along with nonlinear optical devices like photorefractive crystals for enhanced processing

Methodology Applied

Scientific EffectNonlinear optical effect:

Data Source

PatentUS11600060B2Nonlinear all-optical deep-learning system and method with multistage space-frequency domain modulation

Publication Date: 2023.03.07 TSINGHUA UNIVERSITY

AI summary

The present disclosure discloses a nonlinear all-optical deep-learning system and method with multistage space-frequency domain modulation. The system includes an optical input module, configured to convert input information to optical information, a multistage space-frequency domain modulation module, configured to perform multistage space-frequency domain modulation on the optical information generated by the optical input module so as to generate modulated optical information, and an information acquisition module, configured to transform the modulated optical information onto a Fourier plane or an image plane, and to acquire the transformed optical information so as to generate processed optical information.