Three-dimensional heterogeneous integrated optical computing chip

Abstract

The invention discloses a three-dimensional heterogeneous integrated optical computing chip, and belongs to the field of optical computing. The computing chip consists of three parts: a spatial diffraction optical neural network core particle, a plane matrix reconfigurable optical neural network component and a plane switchable optical switch module, the single or multiple spatial diffraction optical neural network core particles, the plane matrix reconfigurable optical neural network assembly and the plane switchable optical switch module are combined together in a three-dimensional heterogeneous integration mode, and efficient multi-task optical calculation processing is achieved. The spatial diffraction optical neural network and the plane matrix reconfigurable optical neural network are combined with each other, a traditional optical calculation framework is broken through, the method has wide application prospects in the optical calculation field, and the blank of related technical fields is filled.

Description

Technical Field

[0001] The present invention belongs to the field of optical computing, and more specifically, relates to a three-dimensional heterogeneous integrated optical computing chip. Background Art

[0002] With the rapid development of artificial intelligence and deep learning, the demand for processing large-scale data and complex tasks is gradually increasing. Traditional computing platforms such as CPUs and GPUs have performance bottlenecks and excessive energy consumption when facing these demands, which has prompted people to seek new computing methods to meet the growing computing needs.

[0003] As an emerging computing method, optical computing can show great advantages in processing large-scale data with the help of the parallelism and high speed of optics. Compared with electronic computing, optical computing has superior characteristics such as high parallelism, low power consumption, anti-interference and large-capacity transmission. Optical computing has broad application prospects in various fields, including artificial intelligence, image processing, data mining, pattern recognition, etc.

[0004] With the development of optical computing, the current mainstream optical neural networks are spatial diffraction neural networks and planar linear neural networks. The spatial diffraction neural network uses the diffraction of light between multiple phase planes to achieve multi-layer full connection to achieve tasks such as image recognition. The network has a large number of neurons and can realize a larger neural network, but the network cannot be freely adjusted after training, has a single function, and has certain limitations. Planar linear neural networks can be flexibly controlled to achieve different matrix configurations and operations, but the scale of operations is often limited by the size of the network. In view of this, a three-dimensional heterogeneous integrated optical computing chip is proposed, which combines the spatial diffraction neural network and the planar linear matrix network through three-dimensional heterogeneous integration to achieve efficient multi-task optical computing processing. Summary of the invention

[0005] In view of the traditional optical neural network computing architecture, the present invention provides a three-dimensional heterogeneous integrated optical computing chip, which aims to provide a new optical neural network architecture to achieve efficient multi-task optical computing processing, and use the spatial diffraction optical neural network to extract features and reduce the dimension of the input task, and then use the plane matrix reconfigurable optical neural network to process the calculation, and finally detect the output calculation results through the detector array. Flexible switching of multi-tasks can be achieved by flexibly switching the plane matrix reconfigurable optical neural network and different spatial diffraction optical neural network core particles.

[0006] To achieve the above-mentioned object, the present invention provides a three-dimensional heterogeneous integrated optical computing chip, comprising a spatial diffraction optical neural network core particle, a planar matrix reconfigurable optical neural network component and a planar switchable optical switch module, wherein the spatial diffraction optical neural network core particle, the planar matrix reconfigurable optical neural network component and the planar switchable optical switch module are integrated by a three-dimensional heterogeneous integration method, wherein the spatial diffraction neural network core particle comprises at least one layer of multi-phase planes, and propagation layers are provided between the multi-phase planes; the planar matrix neural network component comprises a linear matrix network module and a detector array connected in sequence;

[0007] The spatial diffraction neural network core particle is used to perform feature extraction and dimensionality reduction on the task to be processed to obtain a feature output result, and the feature output result is input into a planar switchable optical switch module through coupling to perform state configuration; the planar matrix neural network component is used to perform calculations on the feature output result after state configuration.

[0008] Beneficial effects: The present invention realizes the combination of single or multiple spatial diffraction optical neural network core particles, planar matrix reconfigurable optical neural network components and planar switchable optical switch modules through heterogeneous integration, thereby realizing efficient multi-task optical computing processing. During the calculation process, the optical computing chip inputs the task to be processed into the spatial diffraction optical neural network core particle for feature extraction and dimensionality reduction; then, it inputs the planar switchable optical switch module through coupling for state configuration; finally, the task is input into the planar matrix reconfigurable optical neural network component for calculation processing, and the calculation result is detected and output through the detector array integrated on the planar matrix reconfigurable optical neural network component. During the optical calculation process, a nonlinear activation function can be inserted to realize nonlinear calculation, thereby improving the accuracy of task recognition. For data set tasks with similar characteristics, it can be achieved by reconfiguring the matrix parameters of the planar matrix reconfigurable optical neural network component; for data set tasks with significantly different characteristics, it can be achieved by adjusting the configuration state of the planar switchable optical switch module and the matrix configuration of the planar matrix reconfigurable optical neural network component; at the same time, it can also be combined with feedback control and intelligent algorithms, driven by tasks, to adaptively adjust the planar switchable optical switch module and the planar matrix reconfigurable optical neural network component, thereby realizing efficient and intelligent self-learning and self-configuration multi-task recognition.

[0009] Preferably, the core particle of the spatial diffraction optical neural network is based on the principle of spatial diffraction optics, through the optimization design of multi-layer phase planes, combined with the unique structure of the spatial diffraction optical neural network, the pixel points of the multi-layer phase planes are used as neurons, and the layers are approximately fully connected through optical diffraction, thereby realizing large-scale linear operations. By utilizing the diffraction characteristics of light waves during propagation, the propagation and superposition of the light field are accurately controlled by adjusting the phase plane distribution, thereby realizing the dimensionality reduction feature extraction of high-pixel image information.

[0010] Preferably, the spatial diffraction optical neural network core particle stacks multiple layers of phase planes in a three-dimensional integration manner to form a spatial diffraction optical neural network core particle, the number of layers is not less than 1 layer, and there is a propagation layer composed of a low refractive index material between the layers. The three-dimensional integration process can be achieved by transfer printing, direct bonding or femtosecond laser 3D direct writing and other technologies to integrate multiple layers of phase planes into a spatial diffraction optical neural network core particle. The phase plane can be realized based on the principles of resonance phase, transmission phase or geometric phase, and the component unit of each phase plane can be a metasurface unit or a diffraction optical unit.

[0011] Preferably, the planar matrix reconfigurable optical neural network component is composed of a linear matrix network module and a detector array. The linear matrix network module is used to realize arbitrary reconfigurable optical matrix calculation, linearly process the signal output by the planar switchable optical switch module, and then output the calculation result through the detector array.

[0012] Preferably, the linear matrix network comprises a network composed of Mach-Zehnder interferometers, micro-ring arrays or multi-mode interference waveguides. Any linear optical matrix configuration can be realized by reconfigurable unit control on the linear matrix network module to meet the needs of different tasks.

[0013] Preferably, the planar switchable optical switch module comprises an efficient coupler array and a switchable optical switch. The efficient coupler array is used to receive characteristic output results from different spatial diffraction optical neural network core particles and convert the spatial light mode into a waveguide mode; the switchable optical switch is used to implement the working state configuration of the planar switchable optical switch module and selectively output the characteristic output results from different spatial diffraction optical neural network core particles to the planar matrix reconfigurable optical neural network component.

[0014] Preferably, the processing materials of the spatial diffraction optical neural network core particles include silicon, silicon nitride, metal, silicon dioxide and polymers, etc.; the processing materials of the planar matrix reconfigurable optical neural network components and the planar switchable optical switch modules include silicon, germanium, silicon dioxide, indium phosphide, gallium arsenide, thin film lithium niobate, polymers or phase change materials, or a mixture of the above materials.

[0015] Preferably, the nonlinear activation function can be realized by inserting materials with nonlinear response (such as graphene, phase change materials, anti-saturated absorber materials, etc.) between layers of the spatial diffraction optical neural network core particle; the nonlinear activation function in the planar matrix neural network component can be composed of microrings, germanium silicon materials, SOA materials, etc.; or after completing the optical operation, the output result is subjected to nonlinear activation function operation in the electrical domain, thereby improving the recognition accuracy.

[0016] Preferably, the three-dimensional heterogeneous integration method can directly bond the spatial diffraction optical neural network core particles on a planar chip composed of a planar matrix reconfigurable optical neural network component and a planar switchable optical switch module through multi-core integration; different spatial diffraction optical neural network core particles, planar matrix reconfigurable optical neural network components and planar switchable optical switch modules can also be bonded to specific areas of an optical adapter plate, and the connection between the spatial diffraction optical neural network core particles and the planar matrix reconfigurable optical neural network components and the planar switchable optical switch modules can be achieved through the optical adapter plate.

[0017] Preferably, the task configuration method of the three-dimensional heterogeneous integrated optical computing chip dynamically switches the routing connection between different spatial diffraction optical neural network core particles and the plane matrix reconfigurable optical neural network components according to the task type through the plane switchable optical switch module, ensuring the flexibility and efficiency of the task processing path. At the same time, by dynamically adjusting the matrix weight configuration of the plane matrix reconfigurable optical neural network component to adapt to the computing requirements of different tasks, efficient and accurate feature extraction and result output are achieved. In addition, the method can also combine feedback control and intelligent algorithms to adaptively optimize and adjust the routing state of the plane switchable optical switch module and the matrix parameters of the plane matrix reconfigurable optical neural network component, realize the self-driven self-learning and self-configuration of multi-task configuration, thereby improving the efficiency and intelligence level in the switching and identification of different tasks. This configuration method can not only meet the rapid identification of similar feature tasks, but also realize the accurate processing of differentiated tasks through deep optimization, and provide self-driving, self-learning and self-configuration capabilities, and finally realize efficient intelligent recognition and calculation in multi-task scenarios.

[0018] Through the above technical solution conceived by the present invention, compared with the prior art, the present invention has the following beneficial effects:

[0019] 1. The present invention discloses a three-dimensional heterogeneous integrated optical computing chip, which fully utilizes the advantages of spatial diffraction neural network and planar matrix neural network to perform efficient multi-task parallel computing. Compared with the serial computing method of electronic computers, it provides a new idea for improving computing power.

[0020] 2. The present invention utilizes the strong processing capability of the spatial diffraction optical neural network to perform dimensionality reduction processing on the input task, realize feature extraction, reduce the processing difficulty of the plane matrix reconfigurable optical neural network, and realize efficient calculation processing.

[0021] 3. The present invention utilizes the flexibility of the planar matrix reconfigurable optical neural network to flexibly configure the input tasks, and performs linear operations on the features extracted by the spatial diffraction optical neural network, so as to realize the configuration under different tasks.

[0022] 4. The present invention utilizes the flexibility of the multi-core solution to flexibly configure tasks with large differences, connecting different spatial diffraction optical neural network cores to a planar matrix reconfigurable optical neural network to process different data tasks.

[0023] 5. The idea of ​​computing architecture used in the present invention is universal and can also be applied to other neural network information processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a three-dimensional heterogeneous integrated optical computing chip architecture.

[0025] Figure 2 It is the core particle working unit of the spatial diffraction optical neural network.

[0026] Figure 3 This is the working principle diagram of the spatial diffraction optical neural network core particle.

[0027] Figure 4 This is the schematic diagram of the planar switchable optical switch module.

[0028] Figure 5 Schematic diagram of a planar matrix reconfigurable optical neural network component.

[0029] Figure 6 This is a working principle diagram of the planar matrix reconfigurable optical neural network component.

[0030] Figure 7 Schematic diagram of a planar chip.

[0031] Figure 8 Schematic diagram of three-dimensional heterogeneous integrated optical computing chip.

[0032] Fig. 9 Schematic diagram of the working process of the three-dimensional heterogeneous integrated optical computing chip system. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0034] The present invention provides a three-dimensional heterogeneous integrated optical computing chip, which consists of three parts: a spatial diffraction optical neural network core, a plane matrix reconfigurable optical neural network component and a plane switchable optical switch module. Through heterogeneous integration, a single or multiple spatial diffraction optical neural network cores, a plane matrix reconfigurable optical neural network component and a plane switchable optical switch module are combined to achieve efficient multi-task optical computing processing. During the calculation process, the optical computing chip inputs the task to be processed into the spatial diffraction optical neural network core for feature extraction and dimensionality reduction; then, the task is input into the plane switchable optical switch module through coupling for state configuration; finally, the task is input into the plane matrix reconfigurable optical neural network component for calculation processing, and the calculation result is detected and output through the detector array integrated on the plane matrix reconfigurable optical neural network component. During the optical computing process, a nonlinear activation function can be inserted to realize nonlinear operation, thereby improving the accuracy of task recognition. For data set tasks with similar characteristics, it can be achieved by reconfiguring the matrix parameters of the planar matrix reconfigurable optical neural network component; for data set tasks with significantly different characteristics, it can be achieved by adjusting the configuration state of the planar switchable optical switch module and the matrix configuration of the planar matrix reconfigurable optical neural network component; at the same time, it can also be combined with feedback control and intelligent algorithms, driven by tasks, to adaptively adjust the planar switchable optical switch module and the planar matrix reconfigurable optical neural network component, thereby realizing efficient and intelligent self-learning and self-configuration multi-task recognition.

[0035] Specifically, the core particle of the spatial diffraction optical neural network is based on the principle of spatial diffraction optics. Through the optimization design of multi-layer phase planes and the unique structure of the spatial diffraction optical neural network, the pixel points of the multi-layer phase planes are used as neurons, and the layers are approximately fully connected through optical diffraction, thereby realizing large-scale linear operations. By utilizing the diffraction characteristics of light waves during propagation and adjusting the distribution of phase planes, the propagation and superposition of light fields can be precisely controlled, thereby realizing the dimensionality reduction feature extraction of high-pixel image information.

[0036] Specifically, the spatial diffraction optical neural network core particle stacks multiple layers of phase planes in a three-dimensional integration manner to form a spatial diffraction optical neural network core particle, the number of layers is not less than 1, and there is a propagation layer composed of low refractive index materials between the layers. The three-dimensional integration process can be achieved through transfer printing, direct bonding or femtosecond laser 3D direct writing and other technologies to integrate multiple layers of phase planes into a spatial diffraction optical neural network core particle. The phase plane can be realized based on the principles of resonance phase, transmission phase or geometric phase, and the component unit of each phase plane can be a metasurface unit or a diffraction optical unit.

[0037] Specifically, the planar matrix reconfigurable optical neural network component consists of a linear matrix network module and a detector array. The linear matrix network module is used to realize arbitrary reconfigurable optical matrix calculations, linearly process the signals output by the planar switchable optical switch module, and then output the calculation results through the detector array.

[0038] Specifically, the linear matrix network includes a network composed of Mach-Zehnder interferometers, micro-ring arrays or multi-mode interference waveguides. Through the reconfigurable unit control on the linear matrix network module, any linear optical matrix configuration can be realized to meet the needs of different tasks.

[0039] Specifically, the planar switchable optical switch module includes a high-efficiency coupler array and a switchable optical switch. The high-efficiency coupler array is used to receive characteristic output results from different spatial diffraction optical neural network core particles and convert the spatial light mode into a waveguide mode; the switchable optical switch is used to realize the working state configuration of the planar switchable optical switch module and selectively output the characteristic output results from different spatial diffraction optical neural network core particles to the planar matrix reconfigurable optical neural network component.

[0040] Specifically, the processing materials of spatial diffraction optical neural network core particles include silicon, silicon nitride, metal, silicon dioxide and polymers, etc.; the processing materials of planar matrix reconfigurable optical neural network components and planar switchable optical switch modules include silicon, germanium, silicon dioxide, indium phosphide, gallium arsenide, thin film lithium niobate, polymers or phase change materials, or a mixture of the above materials.

[0041] Specifically, the nonlinear activation function can be realized by inserting materials with nonlinear response (such as graphene, phase change materials, anti-saturated absorber materials, etc.) between layers of the spatial diffraction optical neural network core particles; the nonlinear activation function in the planar matrix neural network component can be composed of microrings, germanium silicon materials, SOA materials, etc.; or after completing the optical operation, the output result is subjected to nonlinear activation function operation in the electrical domain, thereby improving the recognition accuracy.

[0042] Specifically, the three-dimensional heterogeneous integration method can integrate multiple core particles to directly bond the spatial diffraction optical neural network core particles on a planar chip composed of a planar matrix reconfigurable optical neural network component and a planar switchable optical switch module; different spatial diffraction optical neural network core particles, planar matrix reconfigurable optical neural network components and planar switchable optical switch modules can also be bonded to specific areas of an optical adapter plate, and the connection between the spatial diffraction optical neural network core particles and the planar matrix reconfigurable optical neural network components and the planar switchable optical switch modules can be achieved through the optical adapter plate.

[0043] Specifically, the task configuration method of the three-dimensional heterogeneous integrated optical computing chip dynamically switches the routing connection between different spatial diffraction optical neural network core particles and the plane matrix reconfigurable optical neural network components according to the task type through the plane switchable optical switch module, ensuring the flexibility and efficiency of the task processing path. At the same time, by dynamically adjusting the matrix weight configuration of the plane matrix reconfigurable optical neural network component to adapt to the computing requirements of different tasks, efficient and accurate feature extraction and result output are achieved. In addition, the method can also combine feedback control and intelligent algorithms to adaptively optimize and adjust the routing state of the plane switchable optical switch module and the matrix parameters of the plane matrix reconfigurable optical neural network component, realize task self-driven self-learning and self-configuration multi-task configuration, thereby improving efficiency and intelligence in the switching and recognition of different tasks. This configuration method can not only meet the rapid recognition of similar feature tasks, but also realize accurate processing of differentiated tasks through deep optimization, and provide self-driving, self-learning and self-configuration capabilities, and finally realize efficient intelligent recognition and calculation in multi-task scenarios.

[0044] like Figure 1 As shown, the present invention provides a three-dimensional heterogeneous integrated optical computing chip architecture. The chip includes two core modules: a spatial diffraction optical neural network core particle, a planar switchable optical switch module, and a planar matrix reconfigurable optical neural network component. Through three-dimensional heterogeneous integration, the efficient feature extraction capability of the spatial diffraction optical neural network is combined with the flexible linear computing capability of the planar matrix reconfigurable optical neural network to achieve efficient and multi-task optical computing processing.

[0045] Specific working principle:

[0046] 1. Input processing:

[0047] When tasks from different data sets are input, the corresponding spatial diffraction optical neural network core is selected through an external control circuit. The task input is usually input in the form of an optical image, and the initial light field distribution is generated through modulation.

[0048] 2. Feature extraction and dimensionality reduction of spatial diffraction optical neural network:

[0049] The core particle of the spatial diffraction optical neural network is composed of multiple layers of phase planes, and each layer of phase planes is optimized and trained through previous simulation calculations. The trained phase plane can realize feature extraction and dimensionality reduction operations of input data according to the characteristics of the task. The phase plane modulates the wavefront information of the input light field to extract key features related to the task. Through the cascade effect of multiple layers of phase planes, the high-dimensional input data is projected into a low-dimensional feature space for subsequent efficient calculations.

[0050] 3. State configuration of the planar switchable optical switch module:

[0051] The optical signal (i.e., task characteristics) output from the spatial diffraction optical neural network enters the planar switchable optical switch module through the high-efficiency coupling array. The planar switchable optical switch selects the route suitable for the current spatial diffraction optical neural network core according to external control to ensure that the task characteristics are correctly transmitted to the corresponding matrix network module.

[0052] 4. Matrix operations of planar matrix reconfigurable optical neural network components:

[0053] For different data sets and tasks, the external control circuit is responsible for the working state of the planar matrix reconfigurable optical neural network. The weight configuration of the linear matrix network module is adjusted through the control circuit to achieve efficient support for multiple tasks.

[0054] like Figure 2 As shown in the figure, the spatial diffraction optical neural network core is formed by stacking multiple layers of phase planes in a three-dimensional integrated manner, and there is a propagation layer composed of low refractive index materials between the layers. The phase plane can be realized based on the principles of geometric phase, transmission phase or resonance phase.

[0055] like Figure 3 As shown in the figure, the working principle of the spatial diffraction optical neural network core particle is based on the principle of spatial diffraction. Through the optimization design of multi-layer phase planes and the unique structure of the spatial diffraction optical neural network, large-scale linear operations are realized. By using the diffraction characteristics of light waves during propagation and adjusting the phase plane distribution, the propagation and superposition of the light field are precisely controlled, thereby realizing the dimensionality reduction feature extraction of high-pixel image information. As shown in the figure, the digital image in the figure is extracted by dimensionality reduction into a combination of several light spots, which are provided to the subsequent planar chip processing.

[0056] like Figure 4 As shown, the planar switchable optical switch module is composed of a high-efficiency coupler array and a switchable optical switch. The high-efficiency coupler array is used to receive the characteristic output results from different spatial diffraction optical neural network core particles, and convert the spatial light mode into a waveguide mode through an array-shaped grating coupler to realize the conversion of the signal from space to plane. The switchable optical switch is used to realize the working state configuration of the planar switchable optical switch module, and selectively output the characteristic output results from different spatial diffraction optical neural network core particles to the planar matrix reconfigurable optical neural network component, such as the signals of different colors shown in the figure, for selective output.

[0057] like Figure 5As shown, the schematic diagram of the planar matrix reconfigurable optical neural network component is composed of a linear matrix network module and a detector array. The linear matrix network module is used to realize arbitrary reconfigurable optical matrix calculations, which can be composed of a network composed of a Mach-Zehnder interferometer, a microring array or a multi-mode interference waveguide, and linearly processes the signal output by the planar switchable optical switch module, and then outputs the calculation results through the detector array.

[0058] like Figure 6 As shown in the figure, the plane matrix reconfigurable optical neural network component works. The following is a detailed description of its working principle: The features of the spatial diffraction optical neural network core particles after extraction are as follows Figure 6 As shown in the task input, the spatial diffraction optical neural network core can reduce the dimensionality of complex image processing to different light spots, and select the input plane matrix reconfigurable optical neural network component in the form of an optical signal through a planar switchable optical switch module. The external power supply is connected to the reconfigurable unit of the on-chip network through an electrical control interface, and the working parameters of each module in the linear matrix network (such as phase adjustment) can be dynamically adjusted. According to the characteristics of the input task, the linear matrix network can configure different matrix weights in real time to adapt to different task requirements, thereby realizing flexible switching of multiple tasks. After the input features are calculated by the linear matrix network, as shown in Figure 6 As shown in the result output, it can be clearly seen that the original chaotic signal can be processed linearly to achieve the task recognition result output, and the recognition effect can be clearly seen, and the signals of different images are gathered at different positions. The optical signal enters the detector array for photoelectric conversion and outputs the corresponding task recognition result.

[0059] like Figure 7 As shown, the planar chip integrates and connects the planar switchable optical switch module with the planar matrix reconfigurable optical neural network component.

[0060] like Figure 8 As shown, the three-dimensional heterogeneous integrated optical computing chip is a three-dimensional heterogeneous integration of a spatial diffraction optical neural network core particle and a planar chip. Multiple spatial diffraction optical neural network core particles and planar chips are integrated into corresponding high-efficiency coupler arrays so that they correspond one to one.

[0061] like Fig. 9 As shown, the working diagram of the three-dimensional heterogeneous integrated optical computing chip system is shown. The following is a detailed description of its working principle:

[0062] 1. Data input

[0063] Input data type: Input data such as handwritten digital images (such as the MNIST dataset) are converted into optical signals through external devices (such as light sources or image projection devices). The input optical signals are initially modulated and encoded to adapt to the computational requirements of the subsequent spatial diffraction optical neural network.

[0064] 2. Feature extraction of spatial diffraction optical neural network core particles

[0065] The input optical signal is transmitted to the spatial diffraction optical neural network core, which is composed of multiple layers of phase planes. Through the action of these phase planes, the input signal is subjected to feature extraction and dimensionality reduction. Different data sets for different tasks correspond to different cores, and the phase design of each core is optimized and trained to ensure that the key features related to the task are extracted.

[0066] 3. State configuration of the planar switchable optical switch module:

[0067] The optical signal (i.e., task characteristics) output from the spatial diffraction optical neural network enters the planar switchable optical switch module through the high-efficiency coupling array. The planar switchable optical switch selects the route suitable for the current spatial diffraction optical neural network core according to external control to ensure that the task characteristics are correctly transmitted to the corresponding matrix network module.

[0068] 4. Task recognition of planar matrix reconfigurable optical neural network components

[0069] The planar matrix reconfigurable optical neural network component performs linear operations on the input feature signal through a dynamically configured linear matrix network. The system intelligently controls the planar switchable optical switch module and the planar matrix reconfigurable optical neural network component through external circuits, dynamically selects the network configuration that adapts to the current task, thereby achieving the flexibility of multi-tasking.

[0070] 5. Task output

[0071] The calculation results are output by the planar matrix neural network component and enter the detector array for photoelectric signal conversion. The converted electrical signal is output as the task processing result to the external device for further processing or direct display.

[0072] It will be easily understood by those skilled in the art that the above description is only 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 in the protection scope of the present invention.

Claims

1. A three-dimensional heterogeneous integrated optical computing chip, characterized in that: It includes a spatial diffraction optical neural network core particle, a planar matrix reconfigurable optical neural network component and a planar switchable optical switch module, wherein the spatial diffraction optical neural network core particle, the planar matrix reconfigurable optical neural network component and the planar switchable optical switch module are integrated in a three-dimensional heterogeneous integration manner, wherein the spatial diffraction optical neural network core particle includes at least one layer of multi-phase surfaces, and propagation layers are provided between the multi-phase surfaces; the planar matrix neural network component includes a linear matrix network module and a detector array connected in sequence; The spatial diffraction optical neural network core particle is used to perform feature extraction and dimensionality reduction on the task to be processed to obtain a feature output result, and the feature output result is input into a planar switchable optical switch module through coupling to perform state configuration; the planar matrix neural network component is used to operate on the feature output result after state configuration.

2. The three-dimensional heterogeneous integrated optical computing chip according to claim 1, characterized in that: The pixel points of the multi-layer phase surface are used as neurons, and approximately full connections are established between layers through optical diffraction, thereby realizing linear operations.

3. The three-dimensional heterogeneous integrated optical computing chip according to claim 1, characterized in that: The three-dimensional heterogeneous integration method includes transfer printing, direct bonding or femtosecond laser 3D direct writing.

4. The three-dimensional heterogeneous integrated optical computing chip according to claim 2, characterized in that: The multi-phase surface is a metasurface unit or a diffractive optical unit.

5. The three-dimensional heterogeneous integrated optical computing chip according to claim 1, characterized in that: Linear matrix network modules include Mach-Zehnder interferometers, microring arrays, or multimode interference waveguides.

6. The three-dimensional heterogeneous integrated optical computing chip according to claim 1, characterized in that: The planar switchable optical switch module includes a coupler array and an optical switch; the coupler array is used to receive characteristic light signals from different spatial diffraction optical neural network core particles and convert the spatial light mode into a waveguide mode; the optical switch is used to implement the working state configuration of the planar switchable optical switch module and selectively output the characteristic output results from different spatial diffraction optical neural network core particles to the planar matrix reconfigurable optical neural network component.

7. The three-dimensional heterogeneous integrated optical computing chip according to claim 1, characterized in that: The processing materials of spatial diffraction optical neural network core particles include silicon, silicon nitride, metal, silicon dioxide and polymer; the processing materials of planar matrix reconfigurable optical neural network components and planar switchable optical switch modules include silicon, germanium, silicon dioxide, indium phosphide, gallium arsenide, thin film lithium niobate, polymer, phase change material or a mixture of the above materials.

8. The three-dimensional heterogeneous integrated optical computing chip according to claim 1, characterized in that: The spatial diffraction optical neural network core also includes a first nonlinear activation function layer, which is located between layers of the multi-phase surface and is used to realize nonlinear operations. The material of the nonlinear activation function layer is graphene, phase change material, and anti-saturated absorber material; the planar matrix neural network component also includes a second nonlinear activation function layer, which is located between the matrix network module and the detector array. The material of the second nonlinear activation function layer is germanium silicon material and SOA material.

9. The three-dimensional heterogeneous integrated optical computing chip according to claim 3, characterized in that: The three-dimensional heterogeneous integration method is to directly bond the spatial diffraction optical neural network core particles on a planar chip composed of a planar matrix reconfigurable optical neural network component and a planar switchable optical switch module through multi-core integration; or different spatial diffraction optical neural network core particles, planar matrix reconfigurable optical neural network components and planar switchable optical switch modules are bonded to a preset area of ​​an optical adapter plate, and the spatial diffraction optical neural network core particles are connected with the planar matrix reconfigurable optical neural network components and the planar switchable optical switch module through the optical adapter plate.

10. The three-dimensional heterogeneous integrated optical computing chip according to claim 1, characterized in that: The planar switchable optical switch module dynamically switches the routing connections between different spatial diffraction optical neural network core particles and the planar matrix reconfigurable optical neural network components according to the task type; at the same time, the matrix weight configuration of the planar matrix reconfigurable optical neural network components is dynamically adjusted to adapt to the computing requirements of different tasks; it also includes combining feedback control and intelligent algorithms to adaptively optimize and adjust the routing state of the planar switchable optical switch module and the matrix parameters of the planar matrix reconfigurable optical neural network components to achieve task self-driven self-learning and self-configuration of multi-task configuration.

Images