Optical computing method, device and system based on optical computing network structure and medium

By using a combination of micro-ring resonator and broadband phase shifter on the Mach-Zendel interferometer, the modulation efficiency and flexibility of photon calculation are improved, and the problems of low modulation efficiency and insufficient flexibility in traditional methods are solved, and the problem of low modulation efficiency and insufficient flexibility in traditional methods is achieved. Efficient and diversified computing capabilities.

CN119065451APending Publication Date: 2024-12-03GLITTERINTECH (XUZHOU) LTD
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202411080199.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-07
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The Mach-Zendel interferometer method has low modulation efficiency, large footprint, high power consumption in photon computing, and a single optical processor core can only realize single matrix and vector multiplication, which limits the flexibility of processing diversified computing tasks.

Method used

A micro-ring resonator is set up as a resonant phase shifter on the Mach-Zendel interferometer, and the narrowband and high-efficiency modulation characteristics of the micro-ring resonator are used for low-power phase adjustment, and cooperate with the broadband phase shifter to build a basic computing unit. Improve modulation efficiency, reduce floor area and power consumption, and realize programmable free switching between matrix and matrix multiplication and matrix multiplication.

Benefits of technology

It improves the modulation efficiency of Mach-Zendel interferometer, reduces floor area and power consumption, enhances the flexibility of computing scale and handles diversified computing tasks, and expands the applicable scenarios of optical computing.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119065451A_ABST
    Figure CN119065451A_ABST
Patent Text Reader

Abstract

The invention discloses an optical computing method, device and system based on an optical computing network structure and a medium, the optical computing network structure comprises a plurality of preset computing units, and each preset computing unit comprises a Mach-Zehnder interferometer, a broadband phase shifter and a plurality of micro-ring resonators. Determining a first modulation parameter for each micro-ring resonator from a preset micro-ring modulation parameter range; each pair of micro-ring resonators on the upper arm and the lower arm of the interferometer have the same first modulation parameter when the matrixes are multiplied; any two pairs of micro-ring resonators on the upper arm and the lower arm of the interferometer have different first modulation parameters during multiplication of different matrixes and vectors; the broadband phase shifter performs phase delay modulation on the light passing through the Mach-Zehnder interferometer according to the second modulation parameter, and each micro-ring resonator performs phase modulation on the light passing through the Mach-Zehnder interferometer according to the first modulation parameter; and inputting the optical signal to be calculated into the optical calculation network structure to obtain an optical calculation result output by the optical calculation network structure.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of optical computing technology, and in particular, to an optical computing method, device, system and medium based on an optical computing network structure. Background Art

[0002] Artificial Intelligence (AI) technology has developed rapidly in recent years and has been widely used in many fields such as language translation, image classification, audio recognition, autonomous driving and cancer diagnosis. As the scale of AI models continues to expand, the demand for high computing power, fast processing and energy efficiency is also increasing.

[0003] However, due to basic limitations of traditional electronic integrated circuits (EICs), such as Joule heating effect, electron tunneling effect, parasitic capacitance and radio frequency crosstalk, it is becoming increasingly difficult to improve their performance and energy efficiency through semiconductor technology.

[0004] Photonic computing has emerged as a promising solution, which fully utilizes the advantages of light such as low latency, low power consumption, no electromagnetic interference and parallel processing. It is particularly suitable for performing large-scale matrix operations, which are the basic operations of many AI algorithms such as feedforward neural networks, convolutional neural networks and pulse neural networks. Photonic computing uses the interference, diffraction or dispersion of light to achieve large-scale matrix operations with extremely low power consumption and latency.

[0005] At present, the implementation methods of photon computing can be mainly divided into the following three categories: spatial light diffraction method, Mach-Zehnder interferometer method, and wavelength division multiplexing method based on resonators (such as microrings). Among them, the Mach-Zehnder interferometer method has become an ideal method for photon matrix operations because of its advantages such as on-chip integration, efficient processing of real-valued and complex-valued matrix operations, and easy execution of singular value decomposition (SVD). However, the Mach-Zehnder interferometer method still has disadvantages such as low modulation efficiency, large footprint and power consumption. In addition, when implementing photon computing on an integrated platform, a single optical processor core can only implement a single matrix and vector multiplication, which greatly limits the flexibility of processing diversified computing tasks. Summary of the invention

[0006] The present application aims to solve one of the technical problems in the related art to a certain extent. To this end, the present application provides an optical computing method, device, system and medium based on an optical computing network structure.

[0007] As a first aspect of the present application, an optical computing method based on an optical computing network structure is provided, wherein the optical computing network structure includes a plurality of preset computing units,

[0008] The preset calculation unit includes a Mach-Zehnder interferometer, a broadband phase shifter, and a plurality of microring resonators, wherein the broadband phase shifter is arranged on an upper arm of the Mach-Zehnder interferometer, and at least part of the microring resonators are relatively arranged on the upper arm and on the lower arm of the Mach-Zehnder interferometer. The method includes:

[0009] According to the type to be calculated, a first modulation parameter is determined for each of the microring resonators from a preset microring modulation parameter range; wherein, when the type to be calculated is matrix-matrix multiplication, each pair of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer in the preset calculation unit has the same first modulation parameter; when the type to be calculated is different matrix-vector multiplication, any two pairs of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer in the preset calculation unit have different first modulation parameters;

[0010] Inputting the determined first modulation parameters into the corresponding microring resonators respectively, and inputting the preset second modulation parameters into each of the broadband phase shifters, so that each of the broadband phase shifters performs phase delay modulation on the light passing through the corresponding Mach-Zehnder interferometer according to the second modulation parameters, and each of the microring resonators performs phase modulation on the light passing through the corresponding Mach-Zehnder interferometer according to its own first modulation parameters;

[0011] The optical signal to be calculated is input into the optical calculation network structure to obtain the optical calculation result output by the optical calculation network structure.

[0012] Optionally, the optical computing network structure includes a cascaded input modulation module, a matrix multiplication module and an output detection module;

[0013] The input modulation module includes the preset computing units whose number is consistent with the number of input ports of the optical computing network structure;

[0014] The matrix multiplication module includes the preset computing units whose number is consistent with the square of the number of input ports of the optical computing network structure;

[0015] The output detection module includes addition and subtraction micro-ring filters and balanced detectors, the number of which is consistent with the number of input ports of the optical computing network structure, and the through end and the selected end of one of the addition and subtraction micro-ring filters are simultaneously connected to one of the balanced detectors.

[0016] Optionally, a nonlinear activation module is further cascaded between the matrix multiplication module and the output detection module, and the nonlinear activation module includes preset computing units whose number is consistent with the number of input ports of the optical computing network structure;

[0017] The microring resonator on the upper arm of each of the preset calculation units of the nonlinear activation module and the microring resonator on the lower arm of each of the preset calculation units of the nonlinear activation module are respectively in one and the other of an over-coupled state and an under-coupled state;

[0018] Each microring resonator on each preset calculation unit of the nonlinear activation module operates in a different nonlinear working area.

[0019] Optionally, the type of the optical computing network structure is any one of the following: a singular value decomposition structure for complex number computing, a mapping structure for complex number computing, a singular value decomposition structure for real number computing, and a mapping structure for real number computing;

[0020] In the singular value decomposition structure of the complex number calculation, the types of the preset calculation units in the input modulation module include a dual-port complex number calculation unit and / or a single-port complex number calculation unit, and the types of the preset calculation units in the matrix multiplication module include a dual-port real number calculation unit and / or a single-port real number calculation unit, and a dual-port complex number calculation unit;

[0021] In the mapping structure of the complex number calculation, the type of the preset calculation unit in the input modulation module includes a dual-port complex number calculation unit and / or a single-port complex number calculation unit, and the type of the preset calculation unit in the matrix multiplication module includes a single-port complex number calculation unit and / or a dual-port complex number calculation unit;

[0022] In the singular value decomposition structure of the real number calculation, the types of the preset calculation units in the input modulation module include dual-port real number calculation units and / or single-port real number calculation units, and the types of the preset calculation units in the matrix multiplication module include dual-port real number calculation units and / or single-port real number calculation units, and dual-port complex number calculation units;

[0023] In the mapping structure of the real number calculation, the type of preset calculation unit in the input modulation module includes a dual-port real number calculation unit and / or a single-port real number calculation unit, and the type of preset calculation unit in the matrix multiplication module includes a single-port real number calculation unit and / or a dual-port real number calculation unit.

[0024] Optionally, the dual-port real number calculation unit has dual input ports and dual output ports;

[0025] The dual-port complex number calculation unit has dual input ports and dual output ports, and also includes multiple pairs of micro-ring resonators on the output waveguide of the Mach-Zehnder interferometer relatively arranged therein, the number of the multiple pairs of micro-ring resonators is consistent with the number of multiple pairs of micro-ring resonators relatively arranged on the upper arm and the lower arm of the same Mach-Zehnder interferometer, and, when the type to be calculated is matrix-matrix multiplication, the multiple pairs of micro-ring resonators have the same first modulation parameter, and when the type to be calculated is different matrix-vector multiplication, any two pairs of micro-ring resonators in the multiple pairs of micro-ring resonators have different first modulation parameters;

[0026] The single-port real number calculation unit has a single input port and a single output port;

[0027] The single-port complex number calculation unit has a single input port and a single output port, and also includes a plurality of micro-ring resonators on an output waveguide of a Mach-Zehnder interferometer arranged therein, the plurality of micro-ring resonators is consistent in number with the plurality of micro-ring resonators arranged on an upper arm or a lower arm of the same Mach-Zehnder interferometer, and, when the type to be calculated is matrix-matrix multiplication, the plurality of micro-ring resonators have the same first modulation parameter, and when the type to be calculated is different matrix-vector multiplication, any two of the plurality of micro-ring resonators have different first modulation parameters.

[0028] As a second aspect of the present application, an electronic device is provided, wherein the electronic device includes:

[0029] one or more processors;

[0030] A memory having one or more computer programs stored thereon, when the one or more computer programs are executed by the one or more processors, the one or more processors implement the optical computing method based on the optical computing network structure provided in the first aspect of the present application.

[0031] As a third aspect of the present application, an optical computing system is provided, wherein the optical computing system includes an optical computing network structure and the electronic device provided in the second aspect of the present application,

[0032] The optical computing network structure includes a plurality of preset computing units, wherein the preset computing units include a Mach-Zehnder interferometer, a broadband phase shifter, and a plurality of micro-ring resonators, wherein the broadband phase shifter is arranged on the upper arm of the Mach-Zehnder interferometer, and at least part of the micro-ring resonators are relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer;

[0033] The broadband phase shifter is used to perform phase delay modulation on the light passing through the corresponding Mach-Zehnder interferometer according to a preset second modulation parameter; each of the microring resonators is used to perform phase modulation on the light passing through the corresponding Mach-Zehnder interferometer according to its own first modulation parameter, so as to perform optical calculation on the optical signal to be calculated; wherein, when the type to be calculated is matrix-matrix multiplication, each pair of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer in the preset calculation unit has the same first modulation parameter; when the type to be calculated is different matrix-vector multiplication, any two pairs of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer in the preset calculation unit have different first modulation parameters.

[0034] Optionally, the optical computing network structure includes a cascaded input modulation module, a matrix multiplication module, a nonlinear activation module and an output detection module;

[0035] The input modulation module includes the preset computing units whose number is consistent with the number of input ports of the optical computing network structure;

[0036] The matrix multiplication module includes the preset computing units whose number is consistent with the square of the number of input ports of the optical computing network structure;

[0037] The nonlinear activation module includes preset computing units whose number is consistent with the number of input ports of the optical computing network structure; the microring resonator on the upper arm of each preset computing unit of the nonlinear activation module and the microring resonator on the lower arm of each preset computing unit of the nonlinear activation module are respectively in one of an over-coupling state and an under-coupling state; each microring resonator on each preset computing unit of the nonlinear activation module operates in a different nonlinear working area;

[0038] The output detection module includes addition and subtraction micro-ring filters and balanced detectors, the number of which is consistent with the number of input ports of the optical computing network structure, and the through end and the selected end of one of the addition and subtraction micro-ring filters are simultaneously connected to one of the balanced detectors.

[0039] Optionally, the type of the optical computing network structure is any one of the following: a singular value decomposition structure for complex number computing, a mapping structure for complex number computing, a singular value decomposition structure for real number computing, and a mapping structure for real number computing;

[0040] In the singular value decomposition structure of the complex number calculation, the types of the preset calculation units in the input modulation module include a dual-port complex number calculation unit and / or a single-port complex number calculation unit, and the types of the preset calculation units in the matrix multiplication module include a dual-port real number calculation unit and / or a single-port real number calculation unit, and a dual-port complex number calculation unit;

[0041] In the mapping structure of the complex number calculation, the type of the preset calculation unit in the input modulation module includes a dual-port complex number calculation unit and / or a single-port complex number calculation unit, and the type of the preset calculation unit in the matrix multiplication module includes a single-port complex number calculation unit and / or a dual-port complex number calculation unit;

[0042] In the singular value decomposition structure of the real number calculation, the types of the preset calculation units in the input modulation module include dual-port real number calculation units and / or single-port real number calculation units, and the types of the preset calculation units in the matrix multiplication module include dual-port real number calculation units and / or single-port real number calculation units, and dual-port complex number calculation units;

[0043] In the mapping structure of the real number calculation, the type of the preset calculation unit in the input modulation module includes a dual-port real number calculation unit and / or a single-port real number calculation unit, and the type of the preset calculation unit in the matrix multiplication module includes a single-port real number calculation unit and / or a dual-port real number calculation unit;

[0044] Wherein, the dual-port real number calculation unit has dual input ports and dual output ports;

[0045] The dual-port complex number calculation unit has dual input ports and dual output ports, and also includes multiple pairs of micro-ring resonators on the output waveguide of the Mach-Zehnder interferometer relatively arranged therein, the number of the multiple pairs of micro-ring resonators is consistent with the number of multiple pairs of micro-ring resonators relatively arranged on the upper arm and the lower arm of the same Mach-Zehnder interferometer, and, when the type to be calculated is matrix-matrix multiplication, the multiple pairs of micro-ring resonators have the same first modulation parameter, and when the type to be calculated is different matrix-vector multiplication, any two pairs of micro-ring resonators in the multiple pairs of micro-ring resonators have different first modulation parameters;

[0046] The single-port real number calculation unit has a single input port and a single output port;

[0047] The single-port complex number calculation unit has a single input port and a single output port, and also includes a plurality of micro-ring resonators on an output waveguide of a Mach-Zehnder interferometer arranged therein, the plurality of micro-ring resonators is consistent in number with the plurality of micro-ring resonators arranged on an upper arm or a lower arm of the same Mach-Zehnder interferometer, and, when the type to be calculated is matrix-matrix multiplication, the plurality of micro-ring resonators have the same first modulation parameter, and when the type to be calculated is different matrix-vector multiplication, any two of the plurality of micro-ring resonators have different first modulation parameters.

[0048] As a fourth aspect of the present application, a computer-readable medium is provided, on which a computer program is stored, wherein when the computer program is executed by a processor, the optical computing method based on the optical computing network structure provided in the first aspect of the present application is implemented.

[0049] In the optical computing method based on the optical computing network structure provided in the embodiment of the present application, a microring resonator is set as a resonant phase shifter on the Mach-Zehnder interferometer, and the narrowband and efficient modulation characteristics of the microring resonator are used to realize the low-power phase adjustment of the optical signals of different wavelengths by the Mach-Zehnder interferometer. At the same time, a broadband phase shifter is set so that the broadband phase shifter provides common-mode phase shift / phase bias and cooperates with the microring resonator to overcome the loss caused by resonance. In this way, the Mach-Zehnder interferometer, microring resonator and broadband phase shifter are constructed as a basic computing unit, which can effectively improve the modulation efficiency of the Mach-Zehnder interferometer, thereby reducing the floor space and power consumption. Furthermore, the optical computing network structure is constructed using the basic computing unit, and according to the type to be calculated, a first modulation parameter is determined for each microring resonator from a preset microring modulation parameter range; when the type to be calculated is matrix-matrix multiplication, each pair of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer in the preset computing unit has the same first modulation parameter; when the type to be calculated is different matrix-vector multiplication, any two pairs of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer in the preset computing unit have different first modulation parameters; the determined first modulation parameter The numbers are respectively input into the corresponding microring resonators, and the preset second modulation parameters are input into each broadband phase shifter, so that each broadband phase shifter performs phase delay modulation on the light passing through the corresponding Mach-Zehnder interferometer according to the second modulation parameters, and each microring resonator performs phase modulation on the light passing through the corresponding Mach-Zehnder interferometer according to its own first modulation parameters, and the optical signal to be calculated is input into the optical computing network structure to obtain the optical computing result output by the optical computing network structure, and programmable free switching of matrix-matrix multiplication and matrix-vector multiplication can be realized, thereby improving the computing scale, the flexibility of handling diversified computing tasks, and expanding the applicable scenarios of optical computing. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The present application is further described below in conjunction with the accompanying drawings:

[0051] Figure 1 It is a flowchart of an implementation method of an optical computing method based on an optical computing network structure provided in an embodiment of the present application;

[0052] Figure 2 is a schematic diagram of a preset calculation unit provided in an embodiment of the present application;

[0053] Figure 3 It is a schematic diagram of an implementation of a preset calculation unit in a nonlinear activation module provided in an embodiment of the present application;

[0054] Figure 4aIt is a schematic diagram of an implementation of a singular value decomposition structure for complex number calculation provided in an embodiment of the present application;

[0055] Figure 4b It is a schematic diagram of an implementation of a mapping structure for complex number calculation provided in an embodiment of the present application;

[0056] Figure 4c It is a schematic diagram of an implementation of a singular value decomposition structure for real number calculation provided in an embodiment of the present application;

[0057] Figure 4d It is a schematic diagram of an implementation of a mapping structure for real number calculation provided in an embodiment of the present application;

[0058] Figure 5a It is a schematic diagram of an implementation of a dual-port real number calculation unit provided in an embodiment of the present application;

[0059] Figure 5b is a schematic diagram of an implementation of a dual-port complex calculation unit provided in an embodiment of the present application;

[0060] Figure 5c is a schematic diagram of an implementation of a single-port real number calculation unit provided in an embodiment of the present application;

[0061] Figure 5d is a schematic diagram of an implementation of a single-port complex calculation unit provided in an embodiment of the present application;

[0062] Figure 6 is a schematic diagram of an implementation of an electronic device provided in an embodiment of the present application;

[0063] Figure 7 It is a schematic diagram of a computer-readable medium provided in an embodiment of the present application.

[0064] Description of Reference Numerals

[0065] 101: Processor 102: Memory

[0066] 103: I / O interface 104: bus DETAILED DESCRIPTION

[0067] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments based on the embodiments are intended to be used to explain the present application and cannot be understood as limiting the present application.

[0068] References to "one embodiment" or "an example" or "an example" in this specification mean that a particular feature, structure, or characteristic described in conjunction with the embodiment itself may be included in at least one embodiment disclosed in the present application. The appearance of the phrase "in one embodiment" in various places in the specification does not necessarily refer to the same embodiment.

[0069] Although the Mach-Zehnder interferometer method is the current mainstream method for implementing photon computing, it has been successfully applied to many fields such as optical neural networks (ONNs), optical descramblers, modular linear optical circuits, optical convolutional neural networks, equalizers, digital to analog converters (DAC), Ising machines, and pattern analysis. However, on the one hand, each interferometer occupies an area of ​​more than 10,000 μm 2 , which is a significant obstacle to achieving ultra-large-scale integrated design. On the other hand, due to the low modulation efficiency, the modulation arm length of the interferometer can reach hundreds of μm. Therefore, when multiple Mach-Zehnder interferometers are cascaded, the footprint and power consumption will increase significantly. At the same time, when implementing photonic computing on an integrated platform, a single optical processor core can only implement a single matrix and vector multiplication, which greatly limits the flexibility of handling diverse computing tasks.

[0070] In view of this, the applicant of this application proposes to set a microring resonator as a resonant phase shifter on the Mach-Zehnder interferometer, and use the narrowband and efficient modulation characteristics of the microring resonator to realize the low-power phase adjustment of the Mach-Zehnder interferometer for optical signals of different wavelengths, and set a broadband phase shifter at the same time, so that the broadband phase shifter and the microring resonator work together to overcome the loss caused by resonance. In this way, the Mach-Zehnder interferometer, microring resonator and broadband phase shifter are constructed as a basic computing unit, which can effectively improve the modulation efficiency of the Mach-Zehnder interferometer, thereby reducing the floor space and power consumption. Furthermore, by using this basic computing unit to construct an optical computing network structure, it is also possible to realize programmable free switching of matrix-matrix multiplication and matrix-vector multiplication, thereby improving the calculation scale and the flexibility of handling diversified computing tasks.

[0071] As a first aspect of the present application, an optical computing method based on an optical computing network structure is provided, wherein the optical computing network structure includes a plurality of preset computing units, the preset computing units include a Mach-Zehnder interferometer, a broadband phase shifter, and a plurality of microring resonators, the broadband phase shifter is arranged on the upper arm of the Mach-Zehnder interferometer, and at least part of the microring resonators are relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer, such as Figure 1 As shown, the method may include:

[0072] In step S110, according to the type to be calculated, a first modulation parameter is determined for each of the microring resonators from a preset microring modulation parameter range; wherein, when the type to be calculated is matrix-matrix multiplication, each pair of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer in the preset calculation unit has the same first modulation parameter; when the type to be calculated is different matrix-vector multiplication, any two pairs of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer in the preset calculation unit have different first modulation parameters;

[0073] In step S120, the determined first modulation parameters are respectively input into corresponding microring resonators, and the preset second modulation parameters are input into each of the broadband phase shifters, so that each of the broadband phase shifters performs phase delay modulation on the light passing through the corresponding Mach-Zehnder interferometer according to the second modulation parameters, and each of the microring resonators performs phase modulation on the light passing through the corresponding Mach-Zehnder interferometer according to its own first modulation parameters;

[0074] In step S130, the optical signal to be calculated is input into the optical computing network structure to obtain the optical computing result output by the optical computing network structure.

[0075] like Figure 2 As shown, it is a schematic diagram of a preset calculation unit provided in an embodiment of the present application, wherein a plurality of pairs of microring resonators ( Figure 2 The microrings 1, microring 1'...microring N, microring N') shown in the figure are used to perform low-power phase adjustment on optical signals of different wavelengths. Of course, only one pair of microring resonators can be set to perform low-power phase adjustment on optical signals of a specific wavelength. In addition, a broadband phase shifter is also provided on the Mach-Zehnder interferometer to cooperate with the microring resonator to overcome the loss caused by resonance.

[0076] In the embodiment of the present application, a first modulation parameter and a preset second modulation parameter are provided to the microring resonator and the broadband phase shifter, respectively. When the preset calculation unit is working, first, each broadband phase shifter makes the light passing through the upper arm of the corresponding Mach-Zehnder interferometer have a phase delay reaching the second modulation parameter according to its own second modulation parameter, so that the entire Mach-Zehnder interferometer can be placed at an orthogonal working point.

[0077] Then, each pair of microring resonators arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer performs phase advance modulation and phase delay modulation on the light passing through the corresponding Mach-Zehnder interferometer according to its own first modulation parameter. For example, the multiple microring resonators arranged on the upper arm of the Mach-Zehnder interferometer gradually blueshift (i.e., move to the short-wave direction) their own resonance wavelength according to their own first modulation parameter, so that the light passing through the lower arm of the corresponding Mach-Zehnder interferometer has a phase advance reaching the first modulation parameter, and the multiple microring resonators arranged on the lower arm of the Mach-Zehnder interferometer gradually redshift (i.e., move to the long-wave direction) their own resonance wavelength according to their own first modulation parameter, so that the light passing through the upper arm of the corresponding Mach-Zehnder interferometer has a phase delay reaching the first modulation parameter. Conversely, the multiple microring resonators arranged on the upper arm of the Mach-Zehnder interferometer perform phase delay modulation, and the multiple microring resonators arranged on the lower arm of the Mach-Zehnder interferometer perform phase advance modulation.

[0078] When the microring resonator is near the resonance point, even if it slightly deviates from the resonance condition, it will cause a significant phase change, thereby achieving efficient modulation at extremely low power consumption. Each pair of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer are blue-shifted and red-shifted respectively, which can cause a phase difference change to obtain a significant total phase difference. Reversing the execution order of blue-shift and red-shift, that is, each pair of microring resonators are red-shifted and blue-shifted respectively, can also cause a phase difference change to obtain a significant reverse total phase difference. Coupled with the phase change generated by the broadband phase shifter, a wide range of absolute phase difference changes can be achieved at extremely low power consumption. In addition, when the resonant wavelength of a pair of microring resonators is blue-shifted and red-shifted around a certain working center wavelength, the phase of the working center wavelength can be arbitrarily adjusted. When the resonant wavelengths of multiple pairs of microring resonators are blue-shifted and red-shifted around multiple working center wavelengths, the phases of the multiple working center wavelengths can be arbitrarily adjusted. Due to the narrowband characteristics of the microring resonator, even if multiple wavelengths are modulated at the same time, the microring resonators will hardly affect each other. In short, within a preset calculation unit, multiple wavelengths of optical signals can be adjusted independently and arbitrarily. This method can be applied to multiple cascade modes such as multi-port Mach-Zehnder interferometers and multi-port Mach-Zehnder interferometers and cascades, single-port Mach-Zehnder interferometers and multi-port Mach-Zehnder interferometers and cascades, and single-port Mach-Zehnder interferometers and single-port Mach-Zehnder interferometers and cascades.

[0079] In the embodiment of the present application, the first modulation parameter is determined from a preset micro-ring modulation parameter range, and the preset micro-ring modulation parameter range is not particularly limited. For example, the preset micro-ring modulation parameter range may be [π / 4, 3π / 4]. There is no particular limitation on the preset second modulation parameter either. For example, the second modulation parameter may be preset to π / 2.

[0080] For example, when the preset calculation unit is working, it first uses a broadband phase shifter to make the phase delay of light passing through the corresponding Mach-Zehnder interferometer reach π / 2, and the entire Mach-Zehnder interferometer is placed at an orthogonal working point. Then, the microring resonators 1, 2, 3...N arranged on the upper arm of the Mach-Zehnder interferometer gradually blue-shift (i.e., move to the short-wave direction) their own resonant wavelengths, so that the phase advance of light passing through the lower arm of the corresponding Mach-Zehnder interferometer reaches π / 4 to 3π / 4, and the microring resonators 1', 2', 3'...N' arranged on the lower arm of the Mach-Zehnder interferometer gradually red-shift (i.e., move to the long-wave direction) their own resonant wavelengths, so that the phase delay of light passing through the upper arm of the corresponding Mach-Zehnder interferometer reaches π / 4 to 3π / 4. The phase change produced by the broadband phase shifter plus the phase change caused by the blue shift and red shift of each pair of microring resonators 1 and 1', 2 and 2', 3 and 3'...N and N' can obtain a total phase difference from π to 2π. Similarly, the phase change produced by the broadband phase shifter plus the phase change caused by the red shift and blue shift of each pair of microring resonators 1 and 1', 2 and 2', 3 and 3'...N and N' can obtain a total phase difference from -π to 0. Finally, the absolute phase difference can be changed in the range of [0,2π] at extremely low power consumption.

[0081] It should be noted that in the embodiment of the present application, the first modulation parameter is determined for each pair of microring resonators according to whether the optical signal to be calculated needs to perform matrix-matrix multiplication or matrix-vector multiplication. For example, for the first pair of microring resonators 1 and 1', the second pair of microring resonators 2 and 2', the third pair of microring resonators 3 and 3'... the Nth pair of microring resonators N and N', which are relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer, each pair of microring resonators, that is, each pair of microring resonators (for example, 1 and 1') arranged relatively have the same first modulation parameter, and when performing matrix-matrix multiplication, the total N pairs of microring resonators have the same first modulation parameter, and when performing matrix-vector multiplication, any two pairs of microring resonators in the total N pairs of microring resonators have different first modulation parameters.

[0082] Assume that the optical computing network structure has M input ports and M output ports. The light of each input port has a total of N wavelengths (M and N are both positive integers). The Nth wavelength of the input light of the Mth input port is denoted as λ. mn , preset calculation unit w ijThe modulation factor is denoted as [w ij (λ n )]. When each microring resonator in each preset computing unit modulates each wavelength in unison, w ij (λ n )=w ij , which can complete the input matrix [λ mn ] and the weight matrix [w ij ], which means that single-task calculations can be performed efficiently, and the calculation scale can be improved by using the two dimensions of spatial ports and wavelengths. When the N pairs of microring resonators in each preset computing unit modulate each wavelength differently, they can complete the M input vectors [λ n ] and N different weight matrices [w ij (λ n )], which means that parallel multi-task computing can be performed at the same time. If the optical computing network architecture is integrated on-chip and compared with the multi-core architecture of an electrical chip processor, it is equivalent to integrating multiple cores inside the optical chip that can execute different tasks simultaneously.

[0083] In the optical computing method based on the optical computing network structure provided in the embodiment of the present application, a microring resonator is set as a resonant phase shifter on the Mach-Zehnder interferometer, and the narrowband and efficient modulation characteristics of the microring resonator are used to realize the low-power phase adjustment of the optical signals of different wavelengths by the Mach-Zehnder interferometer. At the same time, a broadband phase shifter is set so that the broadband phase shifter provides common-mode phase shift / phase bias and cooperates with the microring resonator to overcome the loss caused by resonance. In this way, the Mach-Zehnder interferometer, microring resonator and broadband phase shifter are constructed as a basic computing unit, which can effectively improve the modulation efficiency of the Mach-Zehnder interferometer, thereby reducing the floor space and power consumption. Furthermore, the optical computing network structure is constructed using the basic computing unit, and according to the type to be calculated, a first modulation parameter is determined for each microring resonator from a preset microring modulation parameter range; when the type to be calculated is matrix-matrix multiplication, each pair of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer in the preset computing unit has the same first modulation parameter; when the type to be calculated is different matrix-vector multiplication, any two pairs of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer in the preset computing unit have different first modulation parameters; the determined first modulation parameter The numbers are respectively input into the corresponding micro-ring resonators, and the preset second modulation parameters are input into each broadband phase shifter, so that each broadband phase shifter performs phase delay modulation on the light passing through the corresponding Mach-Zehnder interferometer according to the second modulation parameters, and each micro-ring resonator performs phase modulation on the light passing through the corresponding Mach-Zehnder interferometer according to its own first modulation parameters, and the optical signal to be calculated is input into the optical computing network structure to obtain the optical computing result output by the optical computing network structure, and the programmable free switching of matrix-matrix multiplication and matrix-vector multiplication can be realized, thereby improving the computing scale, the flexibility of processing diversified computing tasks, and expanding the applicable scenarios of optical computing. .

[0084] In an embodiment of the present application, the optical computing network structure used may include a cascaded input modulation module, a matrix multiplication module and an output detection module. The number of preset computing units of the input modulation module and the matrix multiplication module, the number of add-drop ring filters and the number of balanced detectors of the output module are all related to the number of input ports of the optical computing network structure.

[0085] Accordingly, in some embodiments, the optical computing network structure includes a cascaded input modulation module, a matrix multiplication module, and an output detection module;

[0086] The input modulation module includes the preset computing units whose number is consistent with the number of input ports of the optical computing network structure;

[0087] The matrix multiplication module includes the preset computing units whose number is consistent with the square of the number of input ports of the optical computing network structure;

[0088] The output detection module includes addition and subtraction micro-ring filters and balanced detectors, the number of which is consistent with the number of input ports of the optical computing network structure, and the through end and the selected end of one of the addition and subtraction micro-ring filters are simultaneously connected to one of the balanced detectors.

[0089] Assuming that the optical computing network structure is designed to have M input ports and M output ports (M is a positive integer), the input modulation module includes M preset computing units, the matrix multiplication module includes M×M preset computing units, and the output module includes M addition and subtraction micro-ring filters and M balanced detectors. Based on this setting, programmable free switching between different matrices and matrix multiplications or programmable free switching between different vectors and matrix multiplications can be achieved.

[0090] As described above, when each preset calculation unit modulates each wavelength differently, it can complete M input vectors [λ n ] and N different weight matrices [w ij (λ n )], which means that multi-task computing can be performed simultaneously. If the on-chip integration of the optical computing network architecture is compared with the multi-core architecture of an electrical chip processor, it is equivalent to integrating multiple cores inside the optical chip that can perform different tasks simultaneously. Through the output detection module, the interaction between different cores can be further realized, that is, different output vectors represented by different wavelengths input from the same port to the output detection module are selected through a series of addition and subtraction micro-ring filters or directly passed to the balanced detector to complete the addition and subtraction operations of different output vectors, which greatly improves the computing efficiency and flexibility.

[0091] Furthermore, when it is necessary to detect coherent light, a preset calculation unit can be cascaded before each addition and subtraction micro-ring filter in the output module, and the type of the preset calculation unit is a dual-port real number calculation unit; it is understandable that when detecting incoherent light, it is not necessary to cascade a preset calculation unit before each addition and subtraction micro-ring filter in the output module. Accordingly, in some embodiments, in the output module, a preset calculation unit of the type of a dual-port real number calculation unit is also cascaded before each addition and subtraction micro-ring filter.

[0092] In addition to the defects mentioned above, the traditional optical computing method also has the defect that it is difficult to realize the all-optical nonlinear activation function integrated on-chip, which also limits the scope of application. In this regard, the applicant of this application proposes to cascade a preset computing unit with the same structure between the matrix multiplication module and the output detection module. Based on the preset computing unit with the same structure, the programmability of the all-optical nonlinear activation function can be further realized.

[0093] Accordingly, in some embodiments, a nonlinear activation module is further cascaded between the matrix multiplication module and the output detection module, and the nonlinear activation module includes preset computing units whose number is consistent with the number of input ports of the optical computing network structure;

[0094] The microring resonator on the upper arm of each of the preset calculation units of the nonlinear activation module and the microring resonator on the lower arm of each of the preset calculation units of the nonlinear activation module are respectively in one and the other of an over-coupled state and an under-coupled state;

[0095] Each microring resonator on each preset calculation unit of the nonlinear activation module operates in a different nonlinear working area.

[0096] When the resonator is adjusted to be in a resonant state, a very strong optical field intensity is generated on the resonator, which can induce a nonlinear effect in the silicon or silicon nitride waveguide. In order to highlight this nonlinear effect and use it for the all-optical nonlinear activation function, N pairs of microrings are required for nonlinear activation units with N input wavelengths. Figure 3As shown, a preset calculation unit including a Mach-Zehnder interferometer, a broadband phase shifter arranged on the upper arm of the Mach-Zehnder interferometer, and multiple pairs of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer is used again (the Mach-Zehnder interferometer also includes directional couplers or multi-mode interferometers on both sides, which are not operated in the embodiments of the present application and are only briefly mentioned), and the multiple microring resonators 1, 2, 3...N arranged on the upper arm of the Mach-Zehnder interferometer are placed in an over-coupled state, and the multiple microring resonators 1', 2', 3'...N' arranged on the lower arm of the Mach-Zehnder interferometer are placed in an under-coupled state (or vice versa), so that the phases of the light fields on the two arms caused by the nonlinear effect change in opposite directions, respectively, thereby making the nonlinear effect more significant. At the same time, optical pumping is used to make each microring resonator work in a different nonlinear working area (this can be achieved by adjusting the resonant wavelength of the microring resonator, or by setting optical bias ports 1 and 1', 2 and 2', 3 and 3'... N and N' in each microring resonator to inject laser light at a resonant point where the microring resonator is not at the working wavelength). Ultimately, the programmability of the all-optical nonlinear activation function can be achieved, and the nonlinear effect can be flexibly controlled, so that the best performance can be achieved in different application scenarios.

[0097] It should be noted that, when a nonlinear activation module is cascaded between the matrix multiplication module and the output detection module, the preset calculation unit included in the nonlinear activation module does not need to determine and input the first modulation parameter for its microring resonator, nor does it need to input the preset second modulation parameter to its broadband phase shifter. In other words, its microring resonator and broadband phase shifter do not need to modulate the light passing through the corresponding Mach-Zehnder interferometer according to the first modulation parameter and the preset second modulation parameter, respectively.

[0098] As mentioned above, the Mach-Zehnder interferometer can be divided into a dual-port Mach-Zehnder interferometer and a single-port Mach-Zehnder interferometer. Accordingly, different types of preset computing units can be constructed based on the Mach-Zehnder interferometer, and different types of optical computing network structures can be constructed based on different types of preset computing units.

[0099] Accordingly, in some embodiments, the type of the optical computing network structure is any of the following: a singular value decomposition structure for complex number computing, a mapping structure for complex number computing, a singular value decomposition structure for real number computing, and a mapping structure for real number computing;

[0100] In the singular value decomposition structure of the complex number calculation, the types of the preset calculation units in the input modulation module include a dual-port complex number calculation unit and / or a single-port complex number calculation unit, and the types of the preset calculation units in the matrix multiplication module include a dual-port real number calculation unit and / or a single-port real number calculation unit, and a dual-port complex number calculation unit;

[0101] In the mapping structure of the complex number calculation, the type of the preset calculation unit in the input modulation module includes a dual-port complex number calculation unit and / or a single-port complex number calculation unit, and the type of the preset calculation unit in the matrix multiplication module includes a single-port complex number calculation unit and / or a dual-port complex number calculation unit;

[0102] In the singular value decomposition structure of the real number calculation, the types of the preset calculation units in the input modulation module include dual-port real number calculation units and / or single-port real number calculation units, and the types of the preset calculation units in the matrix multiplication module include dual-port real number calculation units and / or single-port real number calculation units, and dual-port complex number calculation units;

[0103] In the mapping structure of the real number calculation, the type of preset calculation unit in the input modulation module includes a dual-port real number calculation unit and / or a single-port real number calculation unit, and the type of preset calculation unit in the matrix multiplication module includes a single-port real number calculation unit and / or a dual-port real number calculation unit.

[0104] For the above four different types of optical computing network structures, refer to Figure 4a , Figure 4b , Figure 4c and Figure 4d , taking the optical computing network structure including 4 input ports and 4 output ports as an example, a detailed description is given: Figure 4a As shown, it is a schematic diagram of an implementation of the singular value decomposition structure of complex number calculation provided by the embodiment of the present application; Figure 4b As shown, it is a schematic diagram of an implementation of the mapping structure of complex number calculation provided in the embodiment of the present application; Figure 4c As shown, it is a schematic diagram of an implementation of the singular value decomposition structure of real number calculation provided in an embodiment of the present application; Figure 4d , which is a schematic diagram of an implementation of a mapping structure for real number calculation provided in an embodiment of the present application. Figure 4a , Figure 4b , Figure 4c and Figure 4dIn the figure, each block marked with an Arabic numeral represents a preset computing unit, and different Arabic numerals represent different types of preset computing units, "1" represents a dual-port real number computing unit, "2" represents a dual-port complex number computing unit, "3" represents a single-port real number computing unit, and "4" represents a single-port complex number computing unit. It should be noted that since in the preset computing unit included in the nonlinear activation module, the microring resonator and the broadband phase shifter do not need to modulate the light passing through the corresponding Mach-Zehnder interferometer according to the first modulation parameter and the preset second modulation parameter respectively, in order to distinguish it from the preset computing units in the input modulation module, the matrix multiplication module and the output detection module, the Arabic numeral "5" is used to represent the type of the preset computing unit included in the nonlinear activation module, but in fact, there is a possibility that the structure of the preset computing unit "5" included in the nonlinear activation module is consistent with the structure of the preset computing unit "3".

[0105] It can be seen that in Figure 4a , Figure 4b , Figure 4c and Figure 4d In the method, after the optical signal to be calculated is input into the optical computing network structure, the input signal is first modulated by an input modulation module composed of 4 preset computing units, and then the matrix multiplication is completed by a matrix multiplication module composed of 16 preset computing units. Then, the nonlinear activation module composed of 4 preset computing units performs nonlinear activation. Then, the output detection module composed of 4 preset computing units, 4 addition and subtraction micro-ring filters and 4 balanced detectors detects the coherent light, and finally the optical computing result is obtained.

[0106] Since the optical computing network structure can independently modulate light of different wavelengths from different input ports, this programmable modulation greatly increases the flexibility and computing scale of optical computing, and can switch arbitrarily between different matrix-matrix multiplications and different matrix-vector multiplications. In addition, the nonlinear activation module can directly perform on-chip integrated all-optical nonlinear activation operations after linear operations, and can also realize all-optical deep learning inference functions.

[0107] It should be noted that Figure 4a , Figure 4b , Figure 4c and Figure 4d The four types of optical computing network structures shown are all illustrated by taking the example that the input modulation module has only one type of preset computing unit and the matrix multiplication module has only one type of real number computing unit or complex number computing unit. Figure 4aThe complex number calculation singular value decomposition structure shown in FIG. 1 can replace part or all of the "2" (i.e., the dual-port complex number calculation unit) in the input modulation module with "4" (i.e., the single-port complex number calculation unit), or part or all of the "1" (i.e., the dual-port real number calculation unit) in the matrix multiplication module with "3" (i.e., the single-port real number calculation unit), and still obtain the complex number calculation singular value decomposition structure. Figure 4b The complex number calculation mapping structure shown in FIG. 1 can replace part or all of the "4" (i.e., single-port complex number calculation units) in the input modulation module with "2" (i.e., dual-port complex number calculation units), or part or all of the "4" (i.e., single-port complex number calculation units) in the matrix multiplication module with "2" (i.e., dual-port complex number calculation units), and still obtain the complex number calculation mapping structure. Figure 4c The singular value decomposition structure of real number calculation shown in FIG. 1 can replace some or all of the "1" (i.e., dual-port real number calculation unit) in the input modulation module with "3" (i.e., single-port real number calculation unit), or some or all of the "1" (i.e., dual-port real number calculation unit) in the matrix multiplication module with "3" (i.e., single-port real number calculation unit), and still obtain the singular value decomposition structure of real number calculation. For example, Figure 4d The mapping structure of real number calculation shown in the figure can replace part or all of the "3" (that is, single-port real number calculation unit) in the input modulation module with "1" (that is, dual-port real number calculation unit), or can replace part or all of the "3" (that is, single-port real number calculation unit) in the matrix multiplication module with "1" (that is, dual-port real number calculation unit), and still obtain the mapping structure of real number calculation.

[0108] Accordingly, in some embodiments, the dual-port real number calculation unit has dual input ports and dual output ports;

[0109] The dual-port complex number calculation unit has dual input ports and dual output ports, and also includes multiple pairs of micro-ring resonators on the output waveguide of the Mach-Zehnder interferometer relatively arranged therein, the number of the multiple pairs of micro-ring resonators is consistent with the number of multiple pairs of micro-ring resonators relatively arranged on the upper arm and the lower arm of the same Mach-Zehnder interferometer, and, when the type to be calculated is matrix-matrix multiplication, the multiple pairs of micro-ring resonators have the same first modulation parameter, and when the type to be calculated is different matrix-vector multiplication, any two pairs of micro-ring resonators in the multiple pairs of micro-ring resonators have different first modulation parameters;

[0110] The single-port real number calculation unit has a single input port and a single output port;

[0111] The single-port complex number calculation unit has a single input port and a single output port, and also includes a plurality of micro-ring resonators on an output waveguide of a Mach-Zehnder interferometer arranged therein, the plurality of micro-ring resonators is consistent in number with the plurality of micro-ring resonators arranged on an upper arm or a lower arm of the same Mach-Zehnder interferometer, and, when the type to be calculated is matrix-matrix multiplication, the plurality of micro-ring resonators have the same first modulation parameter, and when the type to be calculated is different matrix-vector multiplication, any two of the plurality of micro-ring resonators have different first modulation parameters.

[0112] For the above four different types of preset computing units, refer to Figure 5a , Figure 5b , Figure 5c and Figure 5d Provide detailed explanation.

[0113] like Figure 5a As shown, it is a schematic diagram of an implementation method of a dual-port real number calculation unit provided in an embodiment of the present application. In the dual-port real number calculation unit, for N different wavelengths, it is necessary to use N pairs of microring resonators and 1 broadband phase shifter between the directional couplers or multimode interferometers on both sides of the Mach-Zehnder interferometer to perform phase adjustment on the light passing through the Mach-Zehnder interferometer. Specifically, the N pairs of microring resonators can be relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer, and the broadband phase shifter can be arranged on the upper arm of the Mach-Zehnder interferometer. Each pair of microring resonators performs one and the other of phase advance modulation and phase delay modulation on the light passing through the upper arm and the lower arm of the Mach-Zehnder interferometer, respectively, and the broadband phase shifter performs phase delay modulation on the light passing through the upper arm of the Mach-Zehnder interferometer.

[0114] like Figure 5b As shown, it is a schematic diagram of an implementation method of a dual-port complex computing unit provided in an embodiment of the present application. In the dual-port complex computing unit, for N different wavelengths, it is necessary to use N pairs of microring resonators and 1 broadband phase shifter between the directional couplers or multimode interferometers on both sides of the Mach-Zehnder interferometer to perform phase adjustment on the light passing through the Mach-Zehnder interferometer. Specifically, the N pairs of microring resonators can be relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer, and the broadband phase shifter can be arranged on the upper arm of the Mach-Zehnder interferometer. Each pair of microring resonators performs one and the other of phase advance modulation and phase delay modulation on the light passing through the upper arm and the lower arm of the Mach-Zehnder interferometer, respectively, and the broadband phase shifter performs phase delay modulation on the light passing through the upper arm of the Mach-Zehnder interferometer.

[0115] In addition, for the dual-port complex number calculation unit, it is also necessary to relatively set N pairs of microring resonators on the output waveguide of the Mach-Zehnder interferometer, wherein each pair of microring resonators also performs one and the other of phase advance modulation and phase delay modulation on the light passing through the Mach-Zehnder interferometer. It should be noted that the N pairs of microring resonators arranged on the output waveguide of the Mach-Zehnder interferometer are consistent with the N pairs of microring resonators arranged on the upper arm and the lower arm of the same Mach-Zehnder interferometer in that the number is the same (both are N pairs, i.e., 2N), and when the type to be calculated is matrix-matrix multiplication, the N pairs of microring resonators have the same first modulation parameters, and when the type to be calculated is different matrix-vector multiplication, any two pairs of microring resonators in the N pairs of microring resonators have different first modulation parameters. The difference is that the first modulation parameters can be the same or different, for example, Figure 5b The microrings M and M' ... P and P' shown in the figure may have the same or different first modulation parameters compared with the microrings 1 and 1' ... N and N'. After the N pairs of microring resonators arranged on the output waveguide of the Mach-Zehnder interferometer modulate the optical signal, an absolute phase difference change in the range of [0,2π] can also be achieved.

[0116] like Figure 5c As shown, it is a schematic diagram of an implementation of a single-port real number calculation unit provided in an embodiment of the present application. In the single-port real number calculation unit, for N different wavelengths, it is necessary to use N pairs of microring resonators and 1 broadband phase shifter between the directional couplers or multimode interferometers on both sides of the Mach-Zehnder interferometer to adjust the phase of the light passing through the Mach-Zehnder interferometer. Specifically, the N pairs of microring resonators can be relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer, and the broadband phase shifter can be arranged on the upper arm of the Mach-Zehnder interferometer. Each pair of microring resonators performs one and the other of phase advance modulation and phase delay modulation on the light passing through the upper arm and the lower arm of the Mach-Zehnder interferometer, respectively, and the broadband phase shifter performs phase delay modulation on the light passing through the upper arm of the Mach-Zehnder interferometer. The difference between the single-port real number calculation unit and the dual-port real number calculation unit is that the dual-port real number calculation unit has dual input ports and dual output ports, while the single-port real number calculation unit has a single input port and a single output port.

[0117] like Figure 5dAs shown, it is a schematic diagram of an implementation of a single-port complex number computing unit provided in an embodiment of the present application. In the single-port complex number computing unit, for N different wavelengths, it is necessary to use N pairs of microring resonators and 1 broadband phase shifter between the directional couplers or multimode interferometers on both sides of the Mach-Zehnder interferometer to perform phase adjustment on the light passing through the Mach-Zehnder interferometer. Specifically, the N pairs of microring resonators can be relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer, and the broadband phase shifter can be arranged on the upper arm of the Mach-Zehnder interferometer. Each pair of microring resonators performs one and the other of phase advance modulation and phase delay modulation on the light passing through the upper arm and the lower arm of the Mach-Zehnder interferometer, respectively, and the broadband phase shifter performs phase delay modulation on the light passing through the upper arm of the Mach-Zehnder interferometer.

[0118] In addition, for the single-port complex number calculation unit, N microring resonators need to be arranged on the output waveguide of the Mach-Zehnder interferometer, wherein each microring resonator performs phase advance modulation or phase delay modulation on the light passing through the Mach-Zehnder interferometer. It should be noted that the N microring resonators arranged on the output waveguide of the Mach-Zehnder interferometer are consistent with the N microring resonators arranged on the upper arm or the lower arm of the same Mach-Zehnder interferometer in that the number is the same (both are N) and when the type to be calculated is matrix-matrix multiplication, the N microring resonators have the same first modulation parameters, and when the type to be calculated is different matrix-vector multiplication, any two microring resonators in the N microring resonators have different first modulation parameters. The difference is that the first modulation parameters may be the same or different, for example, Figure 5d The microrings M ... M' and P' shown may have the same or different first modulation parameters compared to microrings 1 ... N. After the N microring resonators disposed on the output waveguide of the Mach-Zehnder interferometer modulate the optical signal, an absolute phase difference change in the range of [0,π] may also be achieved.

[0119] It can be seen that for N different wavelengths, the dual-port real number computing unit includes 2N microring resonators and 1 broadband phase shifter, the dual-port complex number computing unit includes 4N microring resonators and 1 broadband phase shifter, the single-port real number computing unit includes 2N microring resonators and 1 broadband phase shifter, and the single-port complex number computing unit includes 3N microring resonators and 1 broadband phase shifter. In the preset computing unit, the total number of microring resonators depends on the type of the preset computing unit and the number of wavelengths of the input light. In the optical computing network structure, the total number of microring resonators depends on the type of the preset computing unit, the type of the optical computing network structure, and the number of wavelengths of the input light.

[0120] The dual-port real number calculation unit and the single-port real number calculation unit can modulate the amplitude of the input optical signal as a whole through the N pairs of microring resonators arranged on the Mach-Zehnder interferometer. The dual-port complex number calculation unit and the single-port complex number calculation unit can modulate the amplitude of the input optical signal as a whole through the N pairs of microring resonators arranged on the Mach-Zehnder interferometer; the phase of the input optical signal can be modulated as a whole through the N pairs or N microring resonators arranged on the output waveguide; therefore, the amplitude and phase of the input optical signal can be modulated at the same time as a whole. Therefore, the singular value decomposition structure constructed based on the dual-port complex number calculation unit and / or the single-port complex number calculation unit and the dual-port real number calculation unit, and the mapping structure constructed based on the dual-port complex number calculation unit and / or the single-port complex number calculation unit can both modulate the amplitude and phase of the optical signal at the same time. The amplitude and phase of the optical signal correspond to the amplitude and phase angle of the complex number, respectively, so complex number calculation can be performed. If only real number calculations are needed, it is sufficient to replace the dual-port complex number calculation unit and the single-port complex number calculation unit in the input modulation module of the singular value decomposition structure of the complex number calculation with a dual-port real number calculation unit and a single-port real number calculation unit, respectively, and the obtained singular value decomposition structure of the real number calculation is obtained. Alternatively, it is sufficient to replace the dual-port complex number calculation unit and the single-port complex number calculation unit in the input modulation module of the mapping structure of the complex number calculation and in the matrix multiplication module with a dual-port real number calculation unit and a single-port real number calculation unit, respectively, and the obtained mapping structure of the real number calculation is obtained.

[0121] As a second aspect of the present application, an electronic device is provided, wherein: Figure 6 As shown, the electronic device includes:

[0122] One or more processors 101;

[0123] The memory 102 stores one or more computer programs. When the one or more computer programs are executed by the one or more processors 101, the one or more processors 101 implement an optical computing method based on an optical computing network structure provided in the first aspect of the present application.

[0124] The electronic device may further include one or more I / O interfaces 103 connected between the processor 101 and the memory 102 and configured to implement information interaction between the processor 101 and the memory 102 .

[0125] Among them, the processor 101 is a device with data processing capabilities, including but not limited to a central processing unit (CPU), etc.; the memory 102 is a device with data storage capabilities, including but not limited to random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory (FLASH); the I / O interface (read-write interface) is connected between the processor and the memory, and can realize information exchange between the processor and the memory, including but not limited to a data bus (Bus), etc.

[0126] In some embodiments, the processor 101 , the memory 102 , and the I / O interface 103 are connected to each other via a bus 104 , and further connected to other components of the computing device.

[0127] As a third aspect of the present application, an optical computing system is provided, wherein the optical computing system includes an optical computing network structure and the electronic device provided by the second aspect of the present application,

[0128] The optical computing network structure includes a plurality of preset computing units, wherein the preset computing units include a Mach-Zehnder interferometer, a broadband phase shifter, and a plurality of micro-ring resonators, wherein the broadband phase shifter is arranged on the upper arm of the Mach-Zehnder interferometer, and at least part of the micro-ring resonators are relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer;

[0129] The broadband phase shifter is used to perform phase delay modulation on the light passing through the corresponding Mach-Zehnder interferometer according to a preset second modulation parameter; each of the microring resonators is used to perform phase modulation on the light passing through the corresponding Mach-Zehnder interferometer according to its own first modulation parameter, so as to perform optical calculation on the optical signal to be calculated; wherein, when the type to be calculated is matrix-matrix multiplication, each pair of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer in the preset calculation unit has the same first modulation parameter; when the type to be calculated is different matrix-vector multiplication, any two pairs of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer in the preset calculation unit have different first modulation parameters.

[0130] Among them, the structure and working principle of the optical computing network structure have been described in detail above and will not be repeated here.

[0131] In the optical computing system provided in the embodiment of the present application, a microring resonator is arranged on the Mach-Zehnder interferometer as a resonant phase shifter, and the narrowband and efficient modulation characteristics of the microring resonator are used to realize the low-power phase adjustment of the optical signals of different wavelengths by the Mach-Zehnder interferometer, and a broadband phase shifter is arranged at the same time, so that the broadband phase shifter provides common-mode phase shift / phase bias and cooperates with the microring resonator to overcome the loss caused by resonance. In this way, the Mach-Zehnder interferometer, microring resonator and broadband phase shifter are constructed as a basic computing unit, which can effectively improve the modulation efficiency of the Mach-Zehnder interferometer, thereby reducing the floor space and power consumption. Furthermore, the optical computing network structure is constructed using the basic computing unit, and according to the type to be calculated, a first modulation parameter is determined for each microring resonator from a preset microring modulation parameter range; when the type to be calculated is matrix-matrix multiplication, each pair of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer in the preset computing unit has the same first modulation parameter; when the type to be calculated is different matrix-vector multiplication, any two pairs of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer in the preset computing unit have different first modulation parameters; the determined first modulation parameter The numbers are respectively input into the corresponding microring resonators, and the preset second modulation parameters are input into each broadband phase shifter, so that each broadband phase shifter performs phase delay modulation on the light passing through the corresponding Mach-Zehnder interferometer according to the second modulation parameters, and each microring resonator performs phase modulation on the light passing through the corresponding Mach-Zehnder interferometer according to its own first modulation parameters, and the optical signal to be calculated is input into the optical computing network structure to obtain the optical computing result output by the optical computing network structure, and programmable free switching of matrix-matrix multiplication and matrix-vector multiplication can be realized, thereby improving the computing scale, the flexibility of handling diversified computing tasks, and expanding the applicable scenarios of optical computing.

[0132] In the embodiments of the present application, there is no special limitation on the phase adjustment method of the broadband phase shifter and the microring resonator. For example, electrical adjustment, thermal adjustment, optical adjustment, etc. can be adopted.

[0133] In the embodiments of the present application, there is no special limitation on the optical integration platform on which the optical computing network structure is implemented. For example, the optical computing network structure can be implemented on optical integration platforms such as silicon-based (Silicon) optical platforms (pure silicon or silicon germanium platforms), silicon nitride (Silicon Nitride, SiN) optical platforms, indium phosphide (Indium Phosphide, InP) or indium arsenide (Indium Arsenide, InAs) and other III-V optical platforms, and lithium phosphate (LN, Lithium Niobate) optical platforms.

[0134] In some embodiments, the optical computing network structure includes a cascaded input modulation module, a matrix multiplication module, a nonlinear activation module, and an output detection module;

[0135] The input modulation module includes the preset computing units whose number is consistent with the number of input ports of the optical computing network structure;

[0136] The matrix multiplication module includes the preset computing units whose number is consistent with the square of the number of input ports of the optical computing network structure;

[0137] The nonlinear activation module includes the preset computing units whose number is consistent with the number of input ports of the optical computing network structure; the microring resonator on the upper arm of each of the preset computing units of the nonlinear activation module, and the microring resonator on the lower arm of each of the preset computing units of the nonlinear activation module, are respectively in one of an over-coupling state and an under-coupling state and the other; each microring resonator on each of the preset computing units of the nonlinear activation module operates in a different nonlinear working area;

[0138] The output detection module includes addition and subtraction micro-ring filters and balanced detectors, the number of which is consistent with the number of input ports of the optical computing network structure, and the through end and the selected end of one of the addition and subtraction micro-ring filters are simultaneously connected to one of the balanced detectors.

[0139] In some embodiments, the type of the optical computing network structure is any of the following: a singular value decomposition structure for complex number computing, a mapping structure for complex number computing, a singular value decomposition structure for real number computing, and a mapping structure for real number computing;

[0140] In the singular value decomposition structure of the complex number calculation, the types of the preset calculation units in the input modulation module include a dual-port complex number calculation unit and / or a single-port complex number calculation unit, and the types of the preset calculation units in the matrix multiplication module include a dual-port real number calculation unit and / or a single-port real number calculation unit, and a dual-port complex number calculation unit;

[0141] In the mapping structure of the complex number calculation, the type of the preset calculation unit in the input modulation module includes a dual-port complex number calculation unit and / or a single-port complex number calculation unit, and the type of the preset calculation unit in the matrix multiplication module includes a single-port complex number calculation unit and / or a dual-port complex number calculation unit;

[0142] In the singular value decomposition structure of the real number calculation, the types of the preset calculation units in the input modulation module include dual-port real number calculation units and / or single-port real number calculation units, and the types of the preset calculation units in the matrix multiplication module include dual-port real number calculation units and / or single-port real number calculation units, and dual-port complex number calculation units;

[0143] In the mapping structure of the real number calculation, the type of the preset calculation unit in the input modulation module includes a dual-port real number calculation unit and / or a single-port real number calculation unit, and the type of the preset calculation unit in the matrix multiplication module includes a single-port real number calculation unit and / or a dual-port real number calculation unit;

[0144] Wherein, the dual-port real number calculation unit has dual input ports and dual output ports;

[0145] The dual-port complex number calculation unit has dual input ports and dual output ports, and also includes multiple pairs of micro-ring resonators on the output waveguide of the Mach-Zehnder interferometer relatively arranged therein, the number of the multiple pairs of micro-ring resonators is consistent with the number of multiple pairs of micro-ring resonators relatively arranged on the upper arm and the lower arm of the same Mach-Zehnder interferometer, and, when the type to be calculated is matrix-matrix multiplication, the multiple pairs of micro-ring resonators have the same first modulation parameter, and when the type to be calculated is different matrix-vector multiplication, any two pairs of micro-ring resonators in the multiple pairs of micro-ring resonators have different first modulation parameters;

[0146] The single-port real number calculation unit has a single input port and a single output port;

[0147] The single-port complex number calculation unit has a single input port and a single output port, and also includes a plurality of micro-ring resonators on an output waveguide of a Mach-Zehnder interferometer arranged therein, the plurality of micro-ring resonators is consistent in number with the plurality of micro-ring resonators arranged on an upper arm or a lower arm of the same Mach-Zehnder interferometer, and, when the type to be calculated is matrix-matrix multiplication, the plurality of micro-ring resonators have the same first modulation parameter, and when the type to be calculated is different matrix-vector multiplication, any two of the plurality of micro-ring resonators have different first modulation parameters.

[0148] It is understandable that the connection between the two preset computing units can be either a dual-port interconnection or a single-port connection.

[0149] As a fourth aspect of the embodiment of the present application, Figure 7 As shown, a computer-readable medium is provided, on which a computer program is stored, wherein when the computer program is executed by a processor, an optical computing method based on an optical computing network structure provided in the first aspect of the present application is implemented.

[0150] It can be understood by a person of ordinary skill in the art that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program. Accordingly, the computer program can be stored in a non-volatile computer-readable storage medium, and the computer program can implement the method of any of the above-mentioned embodiments when executed. Among them, any reference to memory, storage, database or other media used in the embodiments provided in the embodiments of the present application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0151] The above are only specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Those skilled in the art should understand that the present application includes but is not limited to the contents described in the drawings and the above specific embodiments. Any modification that does not deviate from the functional and structural principles of the present application will be included in the scope of the claims.

Claims

1. An optical computing method based on an optical computing network structure, characterized in that: The optical computing network structure includes a plurality of preset computing units, the preset computing units include a Mach-Zehnder interferometer, a broadband phase shifter, and a plurality of microring resonators, the broadband phase shifter is arranged on an upper arm of the Mach-Zehnder interferometer, at least part of the microring resonators are relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer, and the method includes: According to the type to be calculated, a first modulation parameter is determined for each of the microring resonators from a preset microring modulation parameter range; wherein, when the type to be calculated is matrix-matrix multiplication, each pair of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer in the preset calculation unit has the same first modulation parameter; when the type to be calculated is different matrix-vector multiplication, any two pairs of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer in the preset calculation unit have different first modulation parameters; Inputting the determined first modulation parameters into the corresponding microring resonators respectively, and inputting the preset second modulation parameters into each of the broadband phase shifters, so that each of the broadband phase shifters performs phase delay modulation on the light passing through the corresponding Mach-Zehnder interferometer according to the second modulation parameters, and each of the microring resonators performs phase modulation on the light passing through the corresponding Mach-Zehnder interferometer according to its own first modulation parameters; The optical signal to be calculated is input into the optical calculation network structure to obtain the optical calculation result output by the optical calculation network structure.

2. The method according to claim 1, characterized in that: The optical computing network structure includes a cascaded input modulation module, a matrix multiplication module and an output detection module; The input modulation module includes the preset computing units whose number is consistent with the number of input ports of the optical computing network structure; The matrix multiplication module includes the preset calculation units whose number is consistent with the square of the number of input ports of the optical computing network structure; The output detection module includes addition and subtraction micro-ring filters and balanced detectors, the number of which is consistent with the number of input ports of the optical computing network structure, and the through end and the selected end of one of the addition and subtraction micro-ring filters are simultaneously connected to one of the balanced detectors.

3. The method according to claim 2, characterized in that A nonlinear activation module is also cascaded between the matrix multiplication module and the output detection module, and the nonlinear activation module includes preset computing units whose number is consistent with the number of input ports of the optical computing network structure; The microring resonator on the upper arm of each of the preset calculation units of the nonlinear activation module and the microring resonator on the lower arm of each of the preset calculation units of the nonlinear activation module are respectively in one and the other of an over-coupled state and an under-coupled state; Each microring resonator on each preset calculation unit of the nonlinear activation module operates in a different nonlinear working area.

4. The method according to claim 2, characterized in that: The type of the optical computing network structure is any one of the following: a singular value decomposition structure for complex number computing, a mapping structure for complex number computing, a singular value decomposition structure for real number computing, and a mapping structure for real number computing; In the singular value decomposition structure of the complex number calculation, the types of the preset calculation units in the input modulation module include a dual-port complex number calculation unit and / or a single-port complex number calculation unit, and the types of the preset calculation units in the matrix multiplication module include a dual-port real number calculation unit and / or a single-port real number calculation unit, and a dual-port complex number calculation unit; In the mapping structure of the complex number calculation, the type of the preset calculation unit in the input modulation module includes a dual-port complex number calculation unit and / or a single-port complex number calculation unit, and the type of the preset calculation unit in the matrix multiplication module includes a single-port complex number calculation unit and / or a dual-port complex number calculation unit; In the singular value decomposition structure of the real number calculation, the types of the preset calculation units in the input modulation module include dual-port real number calculation units and / or single-port real number calculation units, and the types of the preset calculation units in the matrix multiplication module include dual-port real number calculation units and / or single-port real number calculation units, and dual-port complex number calculation units; In the mapping structure of the real number calculation, the type of preset calculation unit in the input modulation module includes a dual-port real number calculation unit and / or a single-port real number calculation unit, and the type of preset calculation unit in the matrix multiplication module includes a single-port real number calculation unit and / or a dual-port real number calculation unit.

5. The method according to claim 4, characterized in that The dual-port real number calculation unit has dual input ports and dual output ports; The dual-port complex number calculation unit has dual input ports and dual output ports, and also includes multiple pairs of micro-ring resonators on the output waveguide of the Mach-Zehnder interferometer relatively arranged therein, the number of the multiple pairs of micro-ring resonators is consistent with the number of multiple pairs of micro-ring resonators relatively arranged on the upper arm and the lower arm of the same Mach-Zehnder interferometer, and, when the type to be calculated is matrix-matrix multiplication, the multiple pairs of micro-ring resonators have the same first modulation parameter, and when the type to be calculated is different matrix-vector multiplication, any two pairs of micro-ring resonators in the multiple pairs of micro-ring resonators have different first modulation parameters; The single-port real number calculation unit has a single input port and a single output port; The single-port complex number calculation unit has a single input port and a single output port, and also includes a plurality of micro-ring resonators on an output waveguide of a Mach-Zehnder interferometer arranged therein, the plurality of micro-ring resonators is consistent in number with the plurality of micro-ring resonators arranged on an upper arm or a lower arm of the same Mach-Zehnder interferometer, and, when the type to be calculated is matrix-matrix multiplication, the plurality of micro-ring resonators have the same first modulation parameter, and when the type to be calculated is different matrix-vector multiplication, any two of the plurality of micro-ring resonators have different first modulation parameters.

6. An electronic device, characterized in that: The electronic device comprises: one or more processors; A memory having one or more computer programs stored thereon, wherein when the one or more computer programs are executed by the one or more processors, the one or more processors implement an optical computing method based on an optical computing network structure according to any one of claims 1-5.

7. An optical computing system, characterized in that: The optical computing system comprises an optical computing network structure and the electronic device according to claim 6, wherein the optical computing network structure comprises a plurality of preset computing units, wherein the preset computing units comprise a Mach-Zehnder interferometer, a broadband phase shifter, and a plurality of microring resonators, wherein the broadband phase shifter is arranged on an upper arm of the Mach-Zehnder interferometer, and at least part of the microring resonators are relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer; The broadband phase shifter is used to perform phase delay modulation on the light passing through the corresponding Mach-Zehnder interferometer according to a preset second modulation parameter; Each of the microring resonators is used to phase modulate the light passing through the corresponding Mach-Zehnder interferometer according to its own first modulation parameter; To perform optical calculation on the optical signal to be calculated; wherein, when the type of calculation to be calculated is matrix-matrix multiplication, each pair of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer in the preset calculation unit has the same first modulation parameter; When the type to be calculated is different matrix and vector multiplication, any two pairs of microring resonators relatively arranged on the upper arm and the lower arm of the Mach-Zehnder interferometer in the preset calculation unit have different first modulation parameters.

8. The optical computing system according to claim 7, characterized in that The optical computing network structure includes a cascaded input modulation module, a matrix multiplication module, a nonlinear activation module and an output detection module; The input modulation module includes the preset computing units whose number is consistent with the number of input ports of the optical computing network structure; The matrix multiplication module includes the preset calculation units whose number is consistent with the square of the number of input ports of the optical computing network structure; The nonlinear activation module includes preset computing units whose number is consistent with the number of input ports of the optical computing network structure; The microring resonator on the upper arm of each of the preset calculation units of the nonlinear activation module and the microring resonator on the lower arm of each of the preset calculation units of the nonlinear activation module are respectively in one and the other of an over-coupled state and an under-coupled state; Each microring resonator on each of the preset calculation units of the nonlinear activation module operates in a different nonlinear working area; The output detection module includes addition and subtraction micro-ring filters and balanced detectors, the number of which is consistent with the number of input ports of the optical computing network structure, and the through end and the selected end of one of the addition and subtraction micro-ring filters are simultaneously connected to one of the balanced detectors.

9. The optical computing system according to claim 8, characterized in that The type of the optical computing network structure is any one of the following: a singular value decomposition structure for complex number computing, a mapping structure for complex number computing, a singular value decomposition structure for real number computing, and a mapping structure for real number computing; In the singular value decomposition structure of the complex number calculation, the types of the preset calculation units in the input modulation module include a dual-port complex number calculation unit and / or a single-port complex number calculation unit, and the types of the preset calculation units in the matrix multiplication module include a dual-port real number calculation unit and / or a single-port real number calculation unit, and a dual-port complex number calculation unit; In the mapping structure of the complex number calculation, the type of the preset calculation unit in the input modulation module includes a dual-port complex number calculation unit and / or a single-port complex number calculation unit, and the type of the preset calculation unit in the matrix multiplication module includes a single-port complex number calculation unit and / or a dual-port complex number calculation unit; In the singular value decomposition structure of the real number calculation, the types of the preset calculation units in the input modulation module include dual-port real number calculation units and / or single-port real number calculation units, and the types of the preset calculation units in the matrix multiplication module include dual-port real number calculation units and / or single-port real number calculation units, and dual-port complex number calculation units; In the mapping structure of the real number calculation, the type of the preset calculation unit in the input modulation module includes a dual-port real number calculation unit and / or a single-port real number calculation unit, and the type of the preset calculation unit in the matrix multiplication module includes a single-port real number calculation unit and / or a dual-port real number calculation unit; Wherein, the dual-port real number calculation unit has dual input ports and dual output ports; The dual-port complex number calculation unit has dual input ports and dual output ports, and also includes multiple pairs of micro-ring resonators on the output waveguide of the Mach-Zehnder interferometer relatively arranged therein, the number of the multiple pairs of micro-ring resonators is consistent with the number of multiple pairs of micro-ring resonators relatively arranged on the upper arm and the lower arm of the same Mach-Zehnder interferometer, and, when the type to be calculated is matrix-matrix multiplication, the multiple pairs of micro-ring resonators have the same first modulation parameter, and when the type to be calculated is different matrix-vector multiplication, any two pairs of micro-ring resonators in the multiple pairs of micro-ring resonators have different first modulation parameters; The single-port real number calculation unit has a single input port and a single output port; The single-port complex number calculation unit has a single input port and a single output port, and also includes a plurality of micro-ring resonators on an output waveguide of a Mach-Zehnder interferometer arranged therein, the plurality of micro-ring resonators is consistent in number with the plurality of micro-ring resonators arranged on an upper arm or a lower arm of the same Mach-Zehnder interferometer, and, when the type to be calculated is matrix-matrix multiplication, the plurality of micro-ring resonators have the same first modulation parameter, and when the type to be calculated is different matrix-vector multiplication, any two of the plurality of micro-ring resonators have different first modulation parameters.

10. A computer readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, an optical computing method based on an optical computing network structure according to any one of claims 1 to 5 is implemented.

Citation Information

Cited By

  • Intelligent optical computing non-linear convolution chip system, method and architecture

    CN119940421A

  • Intelligent optical computing nonlinear convolution chip system, method and architecture

    CN119940421B

  • Acquisition method of optical computing convolutional neural network and electronic equipment

    CN120930688A