A multi-task diffraction photoelectric neural network system and its usage method

Abstract

This invention discloses a multi-task diffraction photoelectric neural network system and its usage method, relating to the field of optoelectronic equipment technology for artificial intelligence computing. The system includes: a laser; a computer for converting received electronic signals into corresponding phase distribution data; a first spatial light modulator for receiving the phase distribution data to modulate the output light of the laser; an optical path decomposition unit for decomposing the phase-distributed modulated light; and multiple photoelectric detection units, each located on one branch of the optical path decomposition, for receiving light intensity distribution information and outputting processing results. This invention offers multiple data processing functions, reducing the amount of optoelectronic hardware required for processing different tasks.

Description

Technical Field

[0001] This invention relates to the field of optoelectronic devices for artificial intelligence computing, specifically to a multi-task diffraction optoelectronic neural network system and its usage method. Background Technology

[0002] The rapid development of artificial intelligence technology has led to a surge in demand for computing power, but the computing power of electronic computing chips is gradually approaching its physical limits. Therefore, using photons instead of electrons to perform all or part of the data processing and computation has become a way to address the current shortage of computing power and reduce computing energy consumption.

[0003] Currently, photonic computing hardware mainly falls into two categories: mature integrated optical chip devices and spatial light modulators. Compared to photonic computing hardware based on integrated optical chips, methods based on spatial light field modulation technology have advantages such as handling large amounts of computational data and high device fault tolerance. However, current optoelectronic neural networks based on spatial light field modulation technology, once trained, suffer from the problem of only being able to perform classification, recognition, and other computational processing on one type of data feature or type. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the first aspect of this invention provides a multi-task diffraction photoelectric neural network system, which has multiple data processing functions and can reduce the photoelectric hardware required to process different tasks.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A multi-task diffraction photoelectric neural network system, comprising:

[0007] Laser;

[0008] A computer, used to convert received electronic signals into corresponding phase distribution data;

[0009] A first spatial light modulator is used to receive the phase distribution data in order to modulate the received output light of the laser.

[0010] The optical path decomposition unit is used to decompose light after phase distribution modulation.

[0011] Multiple photoelectric detection units are provided, each of which is located on one branch of the optical path decomposition and is used to receive light intensity distribution information and output processing results.

[0012] In some embodiments, the optical path decomposition unit includes a plurality of first beam-splitting prisms disposed on the optical path.

[0013] In some embodiments, a Fourier lens is also provided between the first beam splitter prism and the corresponding photoelectric detection unit.

[0014] In some embodiments, a second spatial light modulator is also provided between the first beam splitter and the corresponding photoelectric detection unit.

[0015] In some embodiments, the photodetector unit is a standalone or arrayed photodetector.

[0016] In some embodiments, a collimator is also provided between the laser and the first spatial light modulator.

[0017] In some embodiments, a linear polarizer is further provided between the collimator and the first spatial light modulator.

[0018] In some embodiments, a second beam splitter is provided between the linear polarizer and the first spatial light modulator, the second beam splitter being used to separate the light incident on the first spatial light modulator from the light reflected by the first spatial light modulator.

[0019] In some embodiments, the laser is a single-wavelength laser.

[0020] The second aspect of the present invention provides a method for using a multi-task diffraction photoelectric neural network system, which has multiple data processing functions and can reduce the photoelectric hardware required when processing different tasks.

[0021] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0022] According to the above-described method of using a multi-task diffraction photoelectric neural network system, the method includes the following steps;

[0023] The received electronic signals are converted into corresponding phase distribution data using a computer;

[0024] The phase distribution data is received using a first spatial light modulator to modulate the output light of the received laser.

[0025] The light after phase distribution modulation is decomposed using an optical path decomposition unit;

[0026] By using photoelectric detection units set on each branch of the optical path decomposition, the corresponding light intensity distribution information is received and the processing results are output.

[0027] Compared with the prior art, the advantages of the present invention are as follows:

[0028] The multi-task diffraction photoelectric neural network system of this invention converts received electronic signals into corresponding phase distribution data using a computer; then, a first spatial light modulator receives the phase distribution data to modulate the output light of the received laser; next, an optical path decomposition unit decomposes the phase-distributed modulated light; finally, multiple photoelectric detection units are placed on one branch of the optical path decomposition to receive light intensity distribution information and output the processing results. This provides multiple data processing functions and reduces the amount of optoelectronic hardware required to handle different tasks. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the structure of the multi-task diffraction photoelectric neural network system in an embodiment of the present invention;

[0030] Figure 2 This is a computational network structure diagram of the multi-task diffraction photoelectric neural network in an embodiment of the present invention;

[0031] Figure 3 This is a flowchart illustrating the usage method of the multi-task diffraction photoelectric neural network system in this embodiment of the invention. Detailed Implementation

[0032] The technical solutions (including preferred technical solutions) of the present invention will be further described in detail below with reference to the accompanying drawings and by way of listing some optional embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0033] In the description of this application, it should be noted that the terms "upper," "lower," etc., indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. Unless otherwise expressly specified and limited, the terms "installed," "connected," and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication between two elements. For those skilled in the art, the specific meaning of the above terms in this application can be understood according to the specific circumstances.

[0034] Furthermore, in this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, without necessarily requiring or implying any such actual relationship or order between these entities or operations. Moreover, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0035] It is worth noting that compared to current AI computing hardware based on integrated electronic circuits, photonic computing has the potential to achieve high-speed, large-scale parallel, and low-power AI computing. However, most current optoelectronic deep neural networks can only perform classification, recognition, and other computational processing on one feature or one type of data. This invention proposes a multi-task diffractive optoelectronic neural network system based on spatial wavefront shaping technology. Specifically, it utilizes computational holographic multiplexing technology, enabling independent AI computing tasks to be implemented on different optical field output ports of a trained diffractive optoelectronic neural network system. The multiplexed AI system can increase processing power, computational scale, and parallelism.

[0036] Specifically, see Figure 1 As shown, this embodiment of the invention discloses a multi-task diffraction photoelectric neural network system, which includes a laser 1, a computer 9, a first spatial light modulator 5, an optical path decomposition unit, and multiple photoelectric detection units 7.

[0037] The computer 9 is used to convert the received electronic signals into corresponding phase distribution data; the first spatial light modulator 5 is used to receive the phase distribution data to modulate the output light of the received laser 1; the optical path decomposition unit is used to decompose the light after phase distribution modulation; each photoelectric detection unit 7 is set on one branch of the optical path decomposition and is used to receive light intensity distribution information and output the processing results.

[0038] In a specific implementation, laser 1 needs to maintain the same wavelength for output. Laser 1 can be a single-wavelength laser with any wavelength, or it can be a combination of multi-wavelength lasers and filters.

[0039] The optical path decomposition unit includes multiple first beam splitters 6 disposed on the optical path, see [link / reference]. Figure 1 As shown, it provides an example of setting up two first beam splitters 6, through which the optical path can be divided into three branches for propagation.

[0040] The photoelectric detection unit is a separate or arrayed photoelectric detector 7, still using Figure 1 For example, Figure 1 The optical path is divided into three branches for propagation. A photodetector 7 is set on each branch. The photodetector 7 is used to receive the light intensity and provide the data processing results.

[0041] In some embodiments, a Fourier lens and / or a second spatial light modulator are also provided between the first beam splitter and the corresponding photoelectric detection unit. This arrangement can increase the independence of each optical field information in each output port, thereby increasing the accuracy of the processing results of each task.

[0042] In some embodiments, a collimator 2 is also provided between the laser and the first spatial light modulator. After the collimator 2 is provided, it can be ensured that the light illuminating the first spatial light modulator 5 is collimated and approximates plane light, thereby ensuring the consistency between training calculation and light field transmission.

[0043] In addition, in some embodiments, a linear polarizer 3 is provided between the collimator 2 and the first spatial light modulator 5. The linear polarizer 3 is used to control the polarization of the laser, so that the spatial light modulator can control the corresponding linearly polarized light.

[0044] Furthermore, a second beam splitter 4 is provided between the linear polarizer 3 and the first spatial light modulator 5. The second beam splitter 4 is used to separate the light incident on the first spatial light modulator from the light reflected by the first spatial light modulator and guide the reflected light to the output port.

[0045] Figure 2 This is a computational network structure diagram of the multi-task diffraction photoelectric neural network in an embodiment of the present invention. A is the input layer, which is the electronic signal to be processed and stored on the computer; B is the first hidden layer, which is the phase distribution data on the first spatial light modulator 5; C is the second hidden layer, which is the light intensity signal measured by the photodetector 7; D is the output layer, whose data is transmitted to the computer for easy viewing by the user.

[0046] Figure 2 The networks in this example are trained entirely on a computer and then used in optoelectronic hardware. Networks E and G are trainable networks, and their weights are updated using a backpropagation algorithm. Either network E or G can be trained, leaving the weights of the other network with random values; or both can be trained simultaneously.

[0047] The network F represents the optical field diffraction characteristics, the value of which is determined by the distance the light travels in space and the specific optical elements in the optical path, and is derived from the classical diffraction formula. This network generally does not require training; if training is needed, the optical path must be adjusted accordingly.

[0048] Based on the above description, the multi-task diffraction photoelectric neural network system of the present invention first converts the electronic signal 8 to be processed at the input end into a specific wavefront distribution of laser light through a first spatial light modulator 5. Subsequently, the modulated light field propagates forward in free space. Due to the diffraction phenomenon of light, the light will form a specific intensity distribution after propagating a certain distance, and finally form a light intensity distribution that can represent the processing result at a specific position in the receiving surface. At the receiving end, multiple independent or arrayed photodetectors 7 are used to receive the light intensity and provide the data processing result. In the middle of the light path from the spatial light modulator to the detector, several first beam splitters 6 are placed. Due to the different distances of light propagation, the intensity distribution of the light split by each first beam splitter 6 is different, so different data processing results can be displayed.

[0049] Because the multi-task diffraction photoelectric neural network system in this invention has multiple data processing functions, taking image classification as an example, this invention can classify an image according to different rules. For example, an image of clothing can be classified according to its shape as T-shirt, shirt, jacket, etc., according to its color as black, blue, white, etc., and according to its material as cotton, fiber, silk, etc. It can also classify different images according to their categories. For example, when a handwritten number image is input into the input port, the number represented by the image is obtained at one output port of the system. When an animal image is input into the same input port, the animal represented by the image is obtained at another output port of the system.

[0050] In summary, the multi-task diffraction photoelectric neural network system of this invention converts the received electronic signals into corresponding phase distribution data using a computer 9; then, a first spatial light modulator 5 receives the phase distribution data to modulate the output light of the received laser 1; next, an optical path decomposition unit 6 decomposes the phase-distributed modulated light; finally, multiple photoelectric detection units 7 are placed on one branch of the optical path decomposition to receive light intensity distribution information and output the processing results. This provides multiple data processing functions and reduces the amount of optoelectronic hardware required to handle different tasks.

[0051] Meanwhile, see Figure 3 As shown in the figure, this invention also discloses a method for using a multi-task diffraction photoelectric neural network system, which includes the following steps;

[0052] S1. Use a computer to convert the received electronic signals into corresponding phase distribution data.

[0053] S2. The phase distribution data is received using a first spatial light modulator to modulate the output light of the received laser.

[0054] S3. Use the optical path decomposition unit to decompose the light after phase distribution modulation.

[0055] S4. Using the photoelectric detection unit set on each branch of the optical path decomposition, receive the corresponding light intensity distribution information and output the processing result.

[0056] In the specific implementation, laser 1 is a single-wavelength laser, and the wavelength can be any wavelength.

[0057] The optical path decomposition unit includes multiple first beam splitters 6 disposed on the optical path, see [link / reference]. Figure 1 As shown, it provides an example of setting up two first beam splitters 6, through which the optical path can be divided into three branches for propagation.

[0058] The photoelectric detection unit is a separate or arrayed photoelectric detector 7, still using Figure 1 For example, Figure 1 The optical path is divided into three branches for propagation. A photodetector 7 is set on each branch. The photodetector 7 is used to receive the light intensity and provide the data processing results.

[0059] In some embodiments, a Fourier lens and / or a second spatial light modulator are also provided between the first beam splitter and the corresponding photoelectric detection unit. This arrangement can increase the independence of each optical field information in each output port, thereby increasing the accuracy of the processing results of each task.

[0060] In some embodiments, a collimator 2 is also provided between the laser and the first spatial light modulator. After the collimator 2 is provided, it can be ensured that the light illuminating the first spatial light modulator 5 is collimated and approximates plane light, thereby ensuring the consistency between training calculation and light field transmission.

[0061] In addition, in some embodiments, a linear polarizer 3 is provided between the collimator 2 and the first spatial light modulator 5. The linear polarizer 3 is used to control the polarization of the laser, so that the spatial light modulator can control the corresponding linearly polarized light.

[0062] Furthermore, a second beam splitter 4 is provided between the linear polarizer 3 and the first spatial light modulator 5. The second beam splitter 4 is used to separate the light incident on the first spatial light modulator from the light reflected by the first spatial light modulator and guide the reflected light to the output port.

[0063] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A multi-task diffraction photoelectric neural network system, characterized in that, include: Laser; A computer, used to convert received electronic signals into corresponding phase distribution data; A first spatial light modulator is used to receive the phase distribution data in order to modulate the received output light of the laser. The optical path decomposition unit is used to decompose light after phase distribution modulation. Multiple photoelectric detection units, each of which is disposed on one branch of the optical path decomposition, are used to receive light intensity distribution information and output processing results; The optical path decomposition unit includes a plurality of first beam-splitting prisms disposed on the optical path; A Fourier lens and / or a second spatial light modulator are also provided between the first beam splitter and the corresponding photoelectric detection unit. The photoelectric detection unit is an independent or arrayed photoelectric detector.

2. The multi-task diffraction photoelectric neural network system according to claim 1, characterized in that: A collimator is also provided between the laser and the first spatial light modulator.

3. The multi-task diffraction photoelectric neural network system according to claim 2, characterized in that: A linear polarizer is also provided between the collimator and the first spatial light modulator.

4. The multi-task diffraction photoelectric neural network system according to claim 3, characterized in that: A second beam splitter is provided between the linear polarizer and the first spatial light modulator. The second beam splitter is used to separate the light incident on the first spatial light modulator from the light reflected by the first spatial light modulator.

5. The multi-task diffraction photoelectric neural network system according to claim 1, characterized in that: The laser is a single-wavelength laser.

6. The method of using a multi-task diffraction photoelectric neural network system according to claim 1, characterized in that, This method Includes the following steps; The received electronic signals are converted into corresponding phase distribution data using a computer; The phase distribution data is received using a first spatial light modulator to modulate the output light of the received laser. The light after phase distribution modulation is decomposed using an optical path decomposition unit; By using photoelectric detection units set on each branch of the optical path decomposition, the corresponding light intensity distribution information is received and the processing results are output.

Images