An all-optical diffractive neural network based on microstructure arrays in optical fibers

Abstract

This invention discloses an all-optical diffraction neural network based on a microstructure array in an optical fiber, belonging to the field of optoelectronic computing technology. By modulating a coherent optical signal through a multilayer microstructure array at the speed of light, neural network-based functions are executed, creating an efficient, fast, and low-occupancy method for implementing machine learning tasks. It efficiently realizes large-scale neural networks using coherent light in a low-power and low-space manner.

Description

Technical Field

[0001] This invention belongs to the field of optoelectronic computing technology, and more specifically, relates to an all-optical diffraction neural network based on microstructure arrays in optical fibers. Background Technology

[0002] Artificial neural networks (ANNs) are simulations of the neural networks in the human brain, aiming to achieve machine learning techniques similar to artificial intelligence by mimicking the collaborative working methods of biological neurons to process data. With the advent of the big data era, ANNs have found widespread application in fields such as medical image analysis and semantic recognition. However, due to the susceptibility of electrical signals to interference, the proportional relationship between processing speed and energy consumption, and the footprint, as well as the limitations imposed by circuit-based neural networks on components such as CPUs and GPUs, traditional electrical neural networks still have significant room for improvement in terms of processing speed and energy consumption when dealing with the extremely high data volumes required by data centers. Currently, all-optical neural networks have been developed, which can use optical signals as carriers and achieve all-optical machine learning by designing appropriate spatial light modulators, phase plates, and other passive optical paths. Although this modulation method avoids the problems of high power consumption and slow processing speed of electrical neural networks, and can execute some functions of the neural network at the speed of light after the system is deployed, the construction of the optical path requires various components and places high demands on optical path design, layout space, and calibration, which is not conducive to the integration and industrialization of optical systems. Summary of the Invention

[0003] In view of the shortcomings of the prior art, the purpose of this invention is to provide an all-optical diffraction neural network based on microstructure arrays in optical fibers, which aims to solve the problems of low miniaturization and complex device structure of all-optical neural networks.

[0004] To achieve the above objectives, this invention provides an all-optical diffraction neural network based on a microstructure array in an optical fiber, comprising an optical input module, a neural network module, and an information acquisition module. The neural network module includes a computable optical fiber, the core of which includes a multi-layer microstructure array. Each microstructure array is used to independently perform phase modulation on the coherent optical signal propagating in the fiber core. The optical input module provides a coherent optical signal carrying input information. The coherent optical signal is phase-modulated by the neural network module, and the phase-modulated coherent optical signal is acquired by the information acquisition module. The features of the phase-modulated output signal are analyzed and decoded to obtain the information carried by the coherent light.

[0005] Preferably, the optical input module includes a light source and a spatial light modulator. The light source is used to generate coherent light, which is modulated by the spatial light modulator to carry input information, thereby obtaining a coherent light signal carrying the input information.

[0006] Preferably, the all-optical diffraction neural network further includes a microstructure fabrication system for writing, modifying, and eliminating multilayer microstructure arrays in a computable optical fiber.

[0007] Preferably, the microstructure array is a three-dimensional array composed of multiple uniformly distributed circular planar arrays of refractive index waveguides along the direction of light propagation in the optical fiber.

[0008] Preferably, the information acquisition module is a photodetector. Preferably, the microstructure processing module includes a femtosecond laser and a three-dimensional scanning galvanometer dynamic focusing system. The femtosecond laser is responsible for providing a 1030nm wavelength laser with a repetition rate of 100-500kHz, a maximum power of 10W, and a pulse width of 300fs-1ps. The three-dimensional scanning galvanometer dynamic focusing system is responsible for deflecting and focusing the laser at the processing site.

[0009] Furthermore, in one embodiment of the present invention, the microstructure fabrication system is implemented by a femtosecond laser and a three-dimensional scanning galvanometer dynamic focusing system, and the fabrication system optimizes the fabrication parameters of the microstructure array in the fiber core using deep learning methods.

[0010] Furthermore, the computable optical fiber is an OM1 standard multimode optical fiber, supporting multi-band communication at 850nm and 1300nm.

[0011] Furthermore, in one embodiment of the present invention, the distance and number of microstructure arrays in the multimode fiber core are obtained through pre-training, including: establishing a simulation model, inputting training and test sets according to task requirements, optimizing the structure during training through an error backpropagation algorithm, adjusting the processing parameters of the microstructure array and the physical distance between each module; after the simulation design is completed, physical manufacturing is performed using a femtosecond laser and a three-dimensional scanning galvanometer dynamic focusing system to execute specific neural network functions.

[0012] Compared with the prior art, the above technical solutions conceived in this invention can achieve the following results.

[0013] Beneficial effects:

[0014] 1. The all-optical diffraction neural network based on a microstructure array in an optical fiber, as described in this invention, modulates a coherent optical signal at the speed of light through a multilayer microstructure array in the optical fiber, performing neural network-based functions. This creates an efficient, fast, and low-occupancy method for implementing machine learning tasks, efficiently realizing large-scale neural networks with coherent light in a low-power, low-space manner. Using this neural network can enrich the functionality of existing optical systems while reducing the demand for electrical equipment, lowering response time, and increasing processing speed. For example, adding a computable optical fiber to the input surface of an existing photodetector allows for pre-extraction of feature values ​​from the input light field, simplifying subsequent analysis steps.

[0015] 2. This invention uses microstructure arrays to reduce or even completely replace the spatial light modulators required in optical paths other than optical fibers, while still maintaining the functionality of the neural network. This reduces the space required to deploy the all-optical neural network and lowers calibration requirements. This invention can improve the integration level of the all-optical neural network model, enabling lightweight and high-response devices such as photodetectors and modulators that require electrical control, making the application of purely optical systems in image recognition and optical sensing possible. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of an all-optical diffraction neural network based on a microstructure array in an optical fiber, according to an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram of the distribution of microstructures in the fiber core.

[0018] Figure 3 This is a schematic diagram illustrating an example of manufacturing computationally readable optical fibers.

[0019] Figure 4 This is a flowchart illustrating the method for implementing machine learning functionality using computable optical fibers.

[0020] Figure 5 This is an embodiment of the present invention, which describes the application of an all-optical diffraction neural network based on microstructure arrays in optical fibers in the design of photodetectors.

[0021] Figure 6 This is a schematic diagram illustrating the optical processing effect of a computable optical fiber on an input signal. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0023] The all-optical diffraction neural network based on microstructure array in optical fiber proposed according to an embodiment of the present invention will be described below with reference to the accompanying drawings.

[0024] Figure 1 This is a schematic diagram of an all-optical diffraction neural network based on a microstructure array in an optical fiber, according to an embodiment of the present invention.

[0025] like Figure 1 As shown, the all-optical diffraction neural network 100 based on microstructure array in optical fiber includes: an optical input module 110, a neural network module 120, and an information acquisition module 130.

[0026] The optical input module 110 provides the modulated optical signal. The neural network module 120 transforms, extracts, and compresses the coherent signal. The information acquisition module 130 receives the output signal from the neural network module and generates processing results based on the output signal.

[0027] The following section will elaborate on the all-optical diffraction neural network 100 based on microstructure arrays in optical fibers.

[0028] Furthermore, in one embodiment of the present invention, as Figure 1 As shown, the optical input module 110 includes: a light source 111 and a spatial light modulator 112.

[0029] The light source 111 is a single-longitudinal-mode solid-state laser with a working wavelength of 532nm, used to provide uniform coherent light. The spatial light modulator 112 is used to modulate the coherent light and convert it into coherent light carrying the input information.

[0030] Furthermore, in one embodiment of the present invention, the optical input module 110 is further configured to use the spatial light modulator 112 to convert the information to be processed into a spatial distribution of the light field.

[0031] Furthermore, in one embodiment of the present invention, the neural network module 120 includes: a computable optical fiber 121 and a microstructure fabrication system 122.

[0032] The computable fiber 121, before processing, is an OM1 multimode fiber with a core radius of 62.5 / 125 μm. OM1 multimode fiber can be considered a scattering medium for long-distance transmission. This invention utilizes multilayer phase modulation to extract the characteristic values ​​of the scattered light spot. The core of the computable fiber 121 contains a multilayer microstructure array, each layer independently modulating the phase of the coherent optical signal propagating in the core. The microstructure processing system 122 uses a 1030 nm femtosecond laser to write, modify, and eliminate the microstructures in the core of the computable fiber 121. The microstructure array is composed of refractive index waveguide-type structural units. At a specific location, the microstructure shape obtained by processing with specific parameters is an optical waveguide. The material of the optical waveguide has the same chemical composition as the unprocessed core, but due to structural changes causing an increase or decrease in molecular density within the processed area, the refractive index of the processed area changes compared to the unprocessed area, thus forming a waveguide. In the fiber core, an optical path difference is generated between light passing through the processed area and light passing through the unprocessed area. When the light propagates to the next modulation surface, the light with different optical paths becomes coherent. After modulation by multiple regions, the light field distribution at the input surface, after multiple interferences of the light passing through each pixel, achieves the target light field distribution at the output surface. Specific processing parameters are optimized during training using the difference between the normalized autocorrelation coefficients of the binarized output and target light spots of fiber 121 and one as the loss function, employing a stochastic gradient descent algorithm. The specific distribution of the waveguide-type microstructure in the fiber core can be found in [reference needed]. Figure 2 .

[0033] Furthermore, in one embodiment of the present invention, the information acquisition module 130 includes a photodetector 131.

[0034] The photodetector 131 is used to receive the output signal of the neural network module and convert the optical signal into an electrical signal.

[0035] Specifically, the photodetector 131 is used to convert the light field information of each pixel on the receiving surface of the photodetector into an electrical signal for data processing and display.

[0036] The all-optical diffraction neural network 100 of this invention can improve the integration of optical neural networks, expand the application scenarios of optical neural networks, and enable optical neural networks to better complete machine learning tasks in practice, especially natural scene image recognition, big data analysis and processing and computing tasks.

[0037] The all-optical diffraction neural network 100 of this invention aims to reduce the use of additional passive optical components such as spatial light modulators and waveplates, reduce the operation process of calibration between different components, thereby reducing the overall loss rate of the optical path, thereby reducing the space required for arranging the all-optical neural network and the calibration requirements, especially reducing the process accuracy requirements during assembly alignment.

[0038] The following is combined Figure 3 The working process of the all-optical diffraction neural network 100 based on microstructure array in optical fiber is further explained.

[0039] like Figure 3 As shown, the coherent optical signal is illuminated by the light source 111 onto the spatial light modulator 112. The modulator converts the coherent light into information-carrying coherent light. The coherent light is input into the neural network module 120, where it undergoes diffraction. It then enters the fiber core from the input end face of the computable fiber 121, passes through a multilayer microstructure array written by the microstructure fabrication system 122, and undergoes feature extraction and information compression processes through modulation of the spatial phase distribution. The output optical signal is received by the photodetector 132.

[0040] The following detailed description, through specific embodiments, will illustrate the establishment process of the all-optical diffraction neural network 100 based on microstructure arrays in optical fibers according to the present invention. The parameters of the all-optical diffraction neural network 100 in this embodiment are obtained by establishing a simulation model and optimizing it using deep learning methods. Specifically, Figure 4 The flowchart of a method for implementing machine learning functionality according to an embodiment of the present invention is as follows:

[0041] S1. Establish a numerical simulation model of an optical system based on an all-optical diffraction neural network with microstructure arrays in optical fibers.

[0042] In the optical input module, the all-optical diffraction neural network of this embodiment uses the coherent light emitted from the light source as the signal light input to the receiving surface of the spatial light modulator. After passing through the spatial light modulator, the output coherent signal light has the following relationship E out (x,y)=f(I in (x,y),x,y), where I in (x,y) is the coherent light intensity input to the spatial light modulator, E out (x,y) represents the output coherent light field at that point. f(I,x,y) is the correspondence between the output light field of the spatial light modulator at a selected location and the input light intensity at that point, determined based on the properties of the photorefractive material and the parameters of the recording and readout light. The output coherent signal light E... out (x, y) becomes the input of the neural network module.

[0043] In the neural network module, the input-output relationship of each fiber core and each microstructure array layer can be expressed as a complex transmittance matrix t(x,y). The output light field passing through the fiber core or microstructure array is the product of the input light field and the transmittance function. The transmittance function is calculated using the Fresnel-Kirchhoff diffraction theorem. The output of the last microstructure array layer is then acquired by the information acquisition module after propagation.

[0044] In the information acquisition module, the detector receives the intensity or phase information of the output light field and stores or processes it further.

[0045] S2. Utilize deep learning methods to simulate the optimal structure of an all-optical diffraction neural network based on microstructure arrays in optical fibers.

[0046] A deep learning network is built based on the simulation model described above. The image to be processed is used as input, and the correct result of the target task is used as the ground truth. Suitable training, validation, and test sets are constructed. In this step, the complex transmissivity matrix is ​​the optimization objective, and the optimizable variables are the size of individual microstructures and the relative distribution of multiple microstructures. The weighted average of the differences between the target complex transmissivity matrix and the actual complex transmissivity matrix is ​​used as the loss function. The gradient of the optimizable variables with respect to the input is calculated using the stochastic gradient descent algorithm. The next round is updated based on the gradient, and the optimal simulation result is obtained by adjusting hyperparameters such as the distance between arrays and the number of arrays.

[0047] S3. Perform preliminary processing on the optical fiber to be processed based on the simulated structure obtained from the simulation optimization.

[0048] By using a femtosecond laser and a three-dimensional scanning galvanometer dynamic focusing system to physically write microstructure arrays into the fiber core, and correctly writing microstructure arrays at various locations in the fiber core according to the simulation model, the function of an all-optical diffraction neural network based on microstructure arrays in optical fibers can be initially realized.

[0049] S4. Optimize the processing parameters of the pre-processed computable optical fiber.

[0050] The input light's optical field distribution is modulated using a spatial light modulator, and the output light's optical field distribution is acquired using a photodetector. The actual complex transmittance matrix is ​​then calculated. In this step, the complex transmittance matrix is ​​the optimization objective, and the optimizable variables are matrices describing the processing parameters. The weighted average of the differences between the objective and actual complex transmittance matrices is used as the loss function. After each processing step, the gradient of the optimizable variables with respect to the input is calculated using a stochastic gradient descent algorithm. The processing parameters for the next round are updated based on the gradient, and the optimal optimization result is obtained by adjusting hyperparameters such as the processing light source power and wavelength.

[0051] The following describes, with reference to the accompanying drawings, the application scenarios of the all-optical diffraction neural network based on microstructure arrays in optical fibers proposed according to embodiments of the present invention.

[0052] Figure 5 This is an embodiment of the present invention, which describes the application of an all-optical diffraction neural network based on microstructure arrays in optical fibers in the design of photodetectors.

[0053] like Figure 5As shown, this all-optical diffraction neural network based on microstructure array in optical fiber is applied to a photodetector system. The photodetector system includes: an input module, a computational optical fiber, a photodetector, and a data processing module.

[0054] The input module provides the coherent optical signal under test, carrying information, and serves as both the light source and signal for the system. The computable optical fiber is used for preprocessing the coherent optical signal under test or extracting feature values, such as... Figure 6 As shown, the complex input light field distribution can be converted into a focused light spot that is easy to analyze and process based on the learning results of the neural network. The photodetector receives the light field distribution after the signal has been processed by the computable fiber and converts it into an electrical signal. The data processing module is used for display or further analysis. The application of this invention can filter and discard useless information in the signal, allowing subsequent calculations to be rationally allocated to effective information, reducing the amount of data processed by electrical equipment, and improving the response speed of the photodetector system.

[0055] Furthermore, in the application of this embodiment of the invention, the computable optical fiber filters out useless information in the optical field distribution of the signal under test, and uses information compression and encoding to convert the actual optical field into an easily processed optical field, reducing the requirement for the accuracy of the photodetector and simplifying the computing power requirement of the entire system.

[0056] Specifically, the computationally readable fiber filters out useless information in the optical field distribution of the signal under test, classifying the complex actual optical field distribution into smaller focused spots based on the learning results of the neural network. This reduces the requirements of the photodetector used, simplifying the computational demands of the entire system and lowering the cost of the photodetector system, particularly in the task of identifying the focused spot location.

[0057] In summary, the all-optical diffraction neural network of this invention differs from traditional optical neural networks. By employing a microstructure array fabricated within the fiber core, the all-optical diffraction neural network of this invention reduces the number of traditional passive optical components used. This allows for simpler calibration and layout design tasks for all all-optical diffraction neural networks, while also reducing the equipment requirements of fixed-purpose high-precision measurement systems. This significantly improves system integration, expands application scope, and enables its widespread use in computing, sensing, modulation, communication, and other fields.

[0058] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An all-optical diffraction neural network based on a microstructure array in an optical fiber, characterized in that, include: The system comprises an optical input module, a neural network module, and an information acquisition module. The neural network module includes a computationally oriented optical fiber, the core of which contains a multi-layer microstructure array. Each microstructure array is used to independently perform phase modulation on the coherent optical signal propagating in the core. The optical input module provides a coherent optical signal carrying input information. The coherent optical signal is phase-modulated by the neural network module, and the phase-modulated coherent optical signal is acquired by the information acquisition module. The module analyzes the characteristics of the phase-modulated output signal and decodes it to obtain the information carried by the coherent light. It also includes a microstructure fabrication system for writing, modifying, and eliminating multilayer microstructure arrays in computable optical fibers.

2. The all-optical diffraction neural network according to claim 1, characterized in that, The optical input module includes a light source and a spatial light modulator. The light source generates coherent light, which is modulated by the spatial light modulator to carry input information, thereby obtaining a coherent light signal carrying the input information.

3. The all-optical diffraction neural network according to claim 1, characterized in that, The microstructure array is a three-dimensional array composed of multiple uniformly distributed circular planar arrays of refractive index waveguides along the direction of light propagation.

4. The all-optical diffraction neural network according to claim 1, characterized in that, The microstructure processing module includes a femtosecond laser and a three-dimensional scanning galvanometer dynamic focusing system. The femtosecond laser is used to provide laser light, and the three-dimensional scanning galvanometer dynamic focusing system is used to deflect and focus the laser light onto the processing site.

5. The all-optical diffraction neural network according to claim 1, characterized in that, The information acquisition module is a photodetector.

6. The all-optical diffraction neural network according to claim 1, characterized in that, The computable optical fiber is an OM1 standard multimode optical fiber.

Images