Single Modulation Layer Diffractive Neural Network for Compact Optical Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep neural networks (DNNs) with multiple modulation layers are complex, physically large, suffer from optical power loss, and finding it difficult to introduce non-linearities due to their enclosed 3D design, which complicates dynamic reconfiguration and alignment.
Innovation Solution
A D2NN system utilizing a single 2D modulation layer with multiple modulation regions, a diffraction layer, and optional non-linearity and Fresnel lenses, allowing for easy reconfiguration and introduction of non-linearities without power loss, using static or dynamic modulation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple modulation layers are used in DNN, then the neural network can perform complex computations, but the device becomes physically large and suffers from optical power loss
Solution Approach 1:
The patent combines multiple modulation layer functions into a single modulation layer by using a cyclic optical path with mirrors. The light beam passes through the same modulation layer multiple times in sequence, performing multiple modulation operations that would traditionally require multiple separate layers, thereby eliminating optical power loss between layers while maintaining computational capability.
Solution Approach 2:
The patent implements continuous modulation by reflecting the light beam back through the modulation layer multiple times without interruption. The cyclic path ensures the light continuously interacts with the modulation layer, performing repeated computational operations without the gaps and losses associated with traditional multi-layer architectures.
2Adaptability or versatility
If multiple modulation layers are used in DNN, then deeper networks can be implemented, but the device complexity and physical size increase
Solution Approach 1:
The patent merges multiple modulation layer functions into a single physical layer that is reused through cyclic reflection. Instead of stacking multiple separate modulation layers vertically, the system uses one modulation layer that the light passes through repeatedly in a cyclic path defined by mirrors, significantly reducing structural complexity.
Solution Approach 2:
The patent transforms the traditional vertical stacking of multiple modulation layers into a cyclic temporal sequence using mirrors. The light beam travels through the modulation layer multiple times in sequence along a reflected path, converting spatial complexity into temporal reuse, thereby achieving deep network functionality with reduced physical footprint.
3Volume of moving object
If enclosed 3D design is used for DNN, then compact structure is achieved, but introducing non-linearities becomes difficult
Solution Approach 1:
The patent extracts the non-linearity introduction function from the enclosed 3D modulation layer structure and implements it separately using optical nonlinear materials or devices positioned in the optical path. This separation allows non-linearities to be introduced without complicating the modulation layer design or requiring complex 3D integration.
Solution Approach 2:
The patent introduces optical nonlinear materials or devices as intermediary elements in the cyclic optical path. These intermediaries enable non-linear optical operations without requiring the modulation layer itself to be complex or three-dimensional, maintaining manufacturing simplicity while achieving the desired non-linear functionality.
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
Simplifies the design, reduces optical power loss, facilitates easy reconfiguration, and enables straightforward introduction of non-linearities, resulting in a more efficient and compact D2NN system.
Implementation Method 1
a diffraction layer configured to diffract the first reflected signal to obtain a diffracted signal
Implementation Method 2
the first modulation region configured to perform a first spatial modulation on the light beam to obtain a first modulation signal
Implementation Method 3
the first spatial modulation and the second spatial modulation are any combination of a phase modulation or an amplitude modulation
Implementation Method 4
a first mirror configured to reflect the first modulated signal at a first non-zero incident angle to obtain a first reflected signal
Implementation Method 5
Fresnel lenses positioned between the diffraction layer and the second mirror
Data Source
AI summary
An apparatus comprises a first mirror; a second mirror; a modulation layer positioned between the first mirror and the second mirror and comprising a plurality of modulation regions; a diffraction layer positioned between the modulation layer and the second mirror, and an input port admitting a light beam into the apparatus. The light beam passes through the diffraction layer and is modulated by the modulation layer to create a first modulated beam before being reflected by the first mirror, the first mirror reflecting the first modulated beam toward the second mirror, the second mirror reflecting the first modulated beam toward the modulation layer to be modulated for at least a second time.