Optical neural network general operator model, architecture and system

Abstract

The invention relates to the technical field of optical computing, in particular to an optical neural network general operator model, architecture and system. The model comprises a reconfigurable time sequence coding module which is used for obtaining an input optical signal and carrying out time sequence coding on the input optical signal based on a time sequence coding parameter corresponding to a target optical calculation task to obtain a time-interleaved signal; the reconfigurable linear optical calculation module is used for performing optical calculation on the time-interleaved signal based on an optical calculation matrix corresponding to the target optical calculation task to obtain a calculated optical signal; and the reconfigurable nonlinear activation module is used for performing nonlinear activation on the calculated optical signal based on a nonlinear activation function corresponding to the target optical calculation task to obtain and output an optical calculation result signal. By adopting the scheme, the flexibility and the practicability of optical calculation can be improved.

Description

Technical Field

[0001] The present disclosure relates to the field of optical computing technology, and in particular to a general operator model, architecture and system of an optical neural network. Background Art

[0002] With the rapid development of artificial intelligence and scientific computing, the complexity and scale of computing needs are also increasing. However, the existing electronic computing technology is limited by Moore's Law, and its performance is gradually approaching saturation, making it difficult to effectively cope with the increasingly stringent requirements of large-scale complex algorithms on computing power and power consumption. Light has natural advantages such as high throughput and low latency in the propagation process. Optical computing technology that uses photons instead of electrons as computing carriers is seen as the key to breaking the existing computing bottleneck. Summary of the invention

[0003] The present disclosure aims to solve one of the technical problems in the related art at least to some extent.

[0004] To this end, the first objective of the present disclosure is to propose a general operator model of optical neural networks to improve the flexibility and practicality of optical computing.

[0005] The second objective of the present disclosure is to propose a general operator architecture for optical neural networks.

[0006] The third objective of the present disclosure is to propose a universal operator system for optical neural networks.

[0007] To achieve the above-mentioned purpose, the first aspect of the present disclosure proposes a general operator model of an optical neural network, including: A reconfigurable timing coding module is used to obtain an input optical signal and perform timing coding on the input optical signal based on a timing coding parameter corresponding to a target optical computing task to obtain a time interleaved signal; A reconfigurable linear optical computing module, used for performing optical computing on the time-interleaved signal based on an optical computing matrix corresponding to the target optical computing task to obtain a calculated optical signal; The reconfigurable nonlinear activation module is used to perform nonlinear activation on the calculated optical signal based on the nonlinear activation function corresponding to the target optical computing task, and obtain and output the optical computing result signal.

[0008] Optionally, the timing coding parameters include the number of split beams and the delay time, and the reconfigurable timing coding module is used to perform timing coding on the input optical signal based on the timing coding parameters corresponding to the target optical computing task to obtain the time interleaved signal, specifically for: Based on the number of split beams, the input optical signal is split to obtain a plurality of split beams; Based on the delay duration, the split light beams in the multiple split light beams are delayed so that the delay between adjacent split light beams is the duration corresponding to a data symbol in the input optical signal, thereby obtaining a time-interleaved signal, wherein the time-interleaved signal includes multiple time-interleaved sub-signals.

[0009] Optionally, when the reconfigurable timing coding module is used to split the input optical signal, it is specifically used to: The beam splitter is controlled to split the input optical signal.

[0010] Optionally, the reconfigurable timing coding module is used to delay the split beams of the multiple split beams based on the delay duration, specifically for: Selecting a first microring resonator subset corresponding to the number of split beams from the first microring resonator set, and determining a control voltage corresponding to each first microring resonator in the first microring resonator subset; The split light is input to the first microring resonator subset to control the first microring resonator subset to delay the split light.

[0011] Optionally, the time-interleaved signal includes a plurality of time-interleaved sub-signals, and the reconfigurable linear optical computing module is used to perform optical computing on the time-interleaved signal based on the optical computing matrix corresponding to the target optical computing task, and when obtaining the calculated optical signal, is specifically used to: In the process of performing optical calculation on the time-interleaved signal based on the optical calculation matrix corresponding to the target optical calculation task, the columns in the matrix correspond one-to-one to the time-interleaved sub-signals, the multiple time-interleaved sub-signals input at the same time are multiplied one-to-one with the row elements in the rows corresponding to the multiple time-interleaved sub-signals input at the same time, and the multiple products obtained after the multiplication are added to obtain the calculated optical signal.

[0012] Optionally, the reconfigurable linear optical computing module is used to perform optical computing on the time-interleaved signal based on the optical computing matrix corresponding to the target optical computing task to obtain the calculated optical signal, specifically for: Based on the target optical computing task, determining a coupling coefficient corresponding to each modulator in a modulator array to obtain an optical computing matrix corresponding to the target optical computing task, wherein the modulator is at least one of a Mach-Zehnder interferometer or a waveguide coupler; The time-interleaved signal is input into the modulator array for optical calculation to obtain a calculated optical signal.

[0013] Optionally, the reconfigurable nonlinear activation module is used to perform nonlinear activation on the calculated optical signal based on the nonlinear activation function corresponding to the target optical computing task to obtain and output an optical computing result signal, including: Selecting a second microring resonator corresponding to the nonlinear activation function corresponding to the target optical computing task from the second microring resonator set to obtain a second microring resonator subset, and determining a control voltage corresponding to each second microring resonator in the second microring resonator subset; The calculated optical signal is input to the second microring resonator set to control the second microring resonator set to perform nonlinear activation on the calculated optical signal, thereby obtaining and outputting an optical calculation result signal.

[0014] To achieve the above-mentioned purpose, the second aspect of the present disclosure proposes a general optical neural network operator architecture, including: a data input module, the general optical neural network operator model shown in any one of the first aspects, a collection module and a control module; wherein, The data input module is used to obtain input data and encode the input data into a one-dimensional time optical signal to obtain an input optical signal; The control module is used to reconstruct the reconfigurable timing coding module, the reconfigurable linear optical computing module, and the reconfigurable nonlinear activation module in the optical neural network general operator model based on the target optical computing task, so as to control the optical neural network general operator model to perform optical computing on the input optical signal based on the target optical computing task to obtain an optical computing result signal; The acquisition module is used to acquire the optical calculation result signal.

[0015] Optionally, the control module is further used to: The working states of the data input module and the acquisition module are controlled.

[0016] To achieve the above-mentioned purpose, the third aspect of the present disclosure proposes an optical neural network general operator system, including: the optical neural network general operator architecture shown in any one of the aforementioned second aspects.

[0017] In summary, the general operator model, architecture and system of the optical neural network provided by the present invention can execute the specific optical computing task by reconstructing the parameters in the reconfigurable timing coding module, the reconfigurable linear optical computing module and the reconfigurable nonlinear activation module for the specific optical computing task. There is no need to design an independent computing model for the specific optical computing task, which can improve the flexibility and practicality of optical computing.

[0018] Additional aspects and advantages of the present disclosure will be given in part in the following description and in part will be obvious from the following description or learned through practice of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The above and / or additional aspects and advantages of the present disclosure will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which: Figure 1 A schematic diagram of the structure of a general operator model of an optical neural network provided in an embodiment of the present disclosure; Figure 2 A schematic diagram of the principle of a convolution operator implementation solution provided in an embodiment of the present disclosure; Figure 3 A schematic diagram of the operation of a general operator model of an optical neural network provided in an embodiment of the present disclosure; Figure 4 A schematic diagram of the structure of a general operator architecture for an optical neural network provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0020] Embodiments of the present disclosure 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 having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present disclosure, and should not be construed as limiting the present disclosure.

[0021] In recent years, the application of neural networks in artificial intelligence has achieved great success. However, traditional electronic computing faces performance bottlenecks when processing complex neural networks. Optical neural networks have the advantages of high speed, high throughput, and low power consumption, providing a new way for efficient neural network computing. However, most of the current optical neural network technologies are specially designed for specific tasks and perform dedicated computing functions, such as diffraction optical neural networks designed for image classification tasks.

[0022] Optical neural networks lacking universal operators are limited in many aspects: Low development efficiency: Different applications require the customization of different operators, which makes the development process cumbersome and time-consuming, and makes it impossible to quickly iterate and deploy new applications; High cost: Designing operators specifically for each task increases R&D costs and resource consumption, which is not conducive to economic efficiency and large-scale promotion; Unstable performance: A dedicated operator may perform well on a specific task, but its performance will fluctuate greatly when switching between different tasks, lacking stability and versatility. Lack of standardization: There is no universal operator, which makes it difficult to establish unified development and operation standards, affecting the interoperability and scalability of the system; Difficult to maintain and upgrade: The complexity of maintaining multiple dedicated operator systems increases, and upgrades require adjustment one by one, which is labor-intensive and prone to errors.

[0023] In summary, the lack of universal operators for optical neural networks has seriously affected the flexibility and practicality of optical neural networks, and restricted their widespread promotion and development in practical applications.

[0024] The present disclosure is described in detail below with reference to specific embodiments.

[0025] Figure 1 This is a schematic diagram of the structure of a general operator model of an optical neural network provided by an embodiment of the present disclosure. Figure 1 As shown, the optical neural network general operator model includes: A reconfigurable timing coding module is used to obtain an input optical signal and perform timing coding on the input optical signal based on a timing coding parameter corresponding to a target optical computing task to obtain a time interleaved signal; A reconfigurable linear optical computing module, used for performing optical computing on the time-interleaved signal based on an optical computing matrix corresponding to a target optical computing task, to obtain a calculated optical signal; The reconfigurable nonlinear activation module is used to perform nonlinear activation on the calculated optical signal based on the nonlinear activation function corresponding to the target optical computing task, and obtain and output the optical computing result signal.

[0026] According to some embodiments, the target optical computing task refers to an optical computing task that needs to be performed on the input optical signal. The target optical computing task does not specifically refer to a fixed task.

[0027] In some embodiments, the timing encoding parameter refers to a parameter used when performing timing encoding on the input light signal according to a target light computing task.

[0028] In some embodiments, the time-interleaved signal refers to an optical signal obtained by performing time-series encoding on an input optical signal.

[0029] According to some embodiments, the optical computation matrix refers to a computation matrix used when performing optical computation on an input optical signal according to a target optical computation task.

[0030] In some embodiments, the calculated optical signal refers to an optical signal obtained after optical calculation is performed on the input optical signal.

[0031] In some embodiments, the calculation method when performing optical calculation on the input optical signal includes but is not limited to matrix multiplication, convolution, etc.

[0032] According to some embodiments, the main function of the nonlinear activation function is to introduce nonlinear properties into the neural network of the optical neural network universal operator model, which enables the optical neural network universal operator model to learn and simulate complex data patterns and functions. Without a nonlinear activation function, no matter how many layers the neural network has, it is essentially still a linear regression model, which limits the expressive power and complexity of the network. The nonlinear activation function enables the neural network to perform nonlinear transformations, thereby solving problems that linear models cannot solve. For example, in classification problems, nonlinear activation functions can help neural networks divide data of different categories, even if the data is linearly inseparable in the original space. In addition, the nonlinear activation function can also simulate the transmission rules of human neurons, that is, signals are transmitted only under specific stimuli, which increases the biological interpretability of the network.

[0033] It should be noted that the general operator model of the optical neural network provided by the present invention can execute a specific optical computing task by reconstructing the parameters in the reconfigurable timing coding module, the reconfigurable linear optical computing module, and the reconfigurable nonlinear activation module. There is no need to design an independent computing model for the specific optical computing task, which can improve the flexibility and practicality of optical computing.

[0034] Optionally, the timing coding parameters include the number of split beams and the delay time. The reconfigurable timing coding module is used to perform timing coding on the input optical signal based on the timing coding parameters corresponding to the target optical computing task, and when the time interleaved signal is obtained, it is specifically used to: Based on the number of split beams, the input optical signal is split to obtain a plurality of split beams; Based on the delay time, the split beams in the multiple split beams are delayed so that the delay between adjacent split beams is the time length corresponding to a data symbol in the input optical signal, thereby obtaining a time interleaved signal, wherein the time interleaved signal includes multiple time interleaved sub-signals.

[0035] According to some embodiments, the reconfigurable timing encoding module may control a beam splitter or other beam splitting element to split the input optical signal.

[0036] In some embodiments, after the reconfigurable timing coding module obtains multiple split beams of light, it can copy them to multiple channels respectively, and delay the split beams input in different channels by different lengths, so that the delay between adjacent channels is ultimately equal to the duration of a data symbol in the input optical signal.

[0037] According to some embodiments, the reconfigurable timing coding module can utilize the group delay effect of the microring resonator to implement the delay operation, and different delay coding can be implemented by adjusting the control voltage of the microring resonator and the number of microring resonators. For example, by cascading multiple microring resonators, different degrees of delay function can be implemented.

[0038] That is, the reconfigurable timing coding module is used to delay the split light in the multiple split light beams based on the delay time, and can select a first microring resonator subset corresponding to the number of split light beams from the first microring resonator set, and determine the control voltage corresponding to each first microring resonator in the first microring resonator subset; the split light beam is input to the first microring resonator subset to control the first microring resonator subset to delay the split light beam. Therefore, the reliability and accuracy of the delay operation can be improved.

[0039] In some embodiments, the reconfigurable timing encoding module may also utilize a spiral waveguide delay line to implement a delay operation on the split beam.

[0040] Optionally, the reconfigurable linear optical computing module is used to perform optical computing on the time-interleaved signal based on the optical computing matrix corresponding to the target optical computing task, and when the calculated optical signal is obtained, it is specifically used to: In the process of performing optical calculation on the time-interleaved signal based on the optical calculation matrix corresponding to the target optical calculation task, the columns in the matrix correspond to the time-interleaved sub-signals one-to-one, and multiple time-interleaved sub-signals input at the same time are multiplied one-to-one with the row elements in the rows corresponding to the multiple time-interleaved sub-signals input at the same time, and the multiple products obtained after the multiplication are added to obtain the calculated optical signal.

[0041] It should be noted that Figure 2 The following is a schematic diagram of the principle of a convolution operator implementation scheme provided by the embodiment of the present disclosure. Figure 2 As shown in the figure, first, the input data channel expansion is realized through the beam splitting operation; then, the input data time sequence interleaving encoding is realized through the delay operation; finally, the multi-channel convolution is realized through the linear matrix operation. Therefore, the universal operator model of the optical neural network can use the time-space interleaving transformation method to realize the multi-channel convolution operator based on the linear matrix operation. In the process of realizing multi-channel convolution through linear matrix operation, by using the time-interleaved signal as a matrix operation operator or a convolution operation operator, multiple time-interleaved sub-signals input at the same time are multiplied one by one with the row elements in the rows corresponding to the multiple time-interleaved sub-signals input at the same time, so that the convolution operation between each row element of the optical calculation matrix and the input optical signal can be realized.

[0042] According to some embodiments, the reconfigurable linear optical computing module is used to perform optical computing on the time-interleaved signal based on the optical computing matrix corresponding to the target optical computing task. When the calculated optical signal is obtained, the coupling coefficient corresponding to each modulator in the modulator array can be determined based on the target optical computing task to obtain the optical computing matrix corresponding to the target optical computing task; the time-interleaved signal is input into the modulator array for optical computing to obtain the calculated optical signal.

[0043] In some embodiments, the modulator array forms a matrix for multiplication operations, and the coupling coefficient brought by each modulator corresponds to each element in the optical computing matrix, also known as a matrix multiplication operator. The modulators include but are not limited to Mach–Zehnder interferometers (MZIs), waveguide couplers, etc. Among them, the tunable MZI and waveguide couplers make the matrix multiplication operator in the optical computing matrix reconfigurable.

[0044] In the MZI array structure, each MZI has two input ports and two output ports, which can realize any 2×2 unitary matrix operation. By matrix decomposition, The unitary transformation of dimension can be obtained by Therefore, any unitary matrix can be realized using the MZI array.

[0045] In the waveguide coupler array structure, the input waveguide and the output waveguide cross vertically, and a curved waveguide is used at each intersection to couple the light in the horizontal input waveguide to the vertical output waveguide. There is a modulator on each curved waveguide to adjust its transmittance, thereby changing the coupling coefficient from the input waveguide to the output waveguide at each intersection.

[0046] Optionally, the reconfigurable nonlinear activation module is used to perform nonlinear activation on the calculated optical signal based on the nonlinear activation function corresponding to the target optical computing task, and obtain and output the optical computing result signal, including: Selecting a second microring resonator corresponding to a nonlinear activation function corresponding to a target optical computing task from the second microring resonator set to obtain a second microring resonator subset, and determining a control voltage corresponding to each second microring resonator in the second microring resonator subset; The calculated optical signal is input to the second microring resonator set to control the second microring resonator set to perform nonlinear activation on the calculated optical signal, thereby obtaining and outputting an optical calculation result signal.

[0047] It should be noted that the on-chip nonlinear activation function is realized by utilizing the Kerr effect and free carrier dispersion effect of the microring resonator. Near the resonant wavelength of the microring resonator, the Kerr effect and free carrier dispersion will cause phase changes related to the optical power. There is a nonlinear response relationship between the transmittance, phase transfer coefficient and input optical power, which can constitute a fast and dynamic all-optical nonlinear activation function.

[0048] In some embodiments, the complex light intensity in the microring resonator It can be expressed as:

[0049] Where t is time; is the resonant frequency of the resonator The detuned frequency, is the incident light frequency; is the refractive index; It is the change in refractive index caused by the Kerr effect; It is the refractive index change caused by free carrier dispersion; is a linear loss; is the two-photon absorption loss; is the free carrier absorption loss; is the energy coupled into the resonator, is the coupling efficiency, is the input energy. Therefore, by adjusting the control voltage and number of the second microring resonator, a reconfigurable and diversified nonlinear activation function can be achieved.

[0050] Take a scenario as an example. Figure 3 A schematic diagram of the operation of a general operator model of an optical neural network provided in an embodiment of the present disclosure. Figure 3 As shown in the figure, first, reconfigurable timing coding is achieved through the microring resonator group delay; then, reconfigurable linear optical computing is achieved through the MZI array or waveguide coupler array; finally, reconfigurable nonlinear activation is achieved through the microring resonator. Therefore, it is possible to efficiently perform basic neural network operations such as matrix multiplication, convolution, and nonlinear activation functions, which is suitable for a variety of application scenarios and neural network structures without the need to design and optimize them separately for each application.

[0051] In order to implement the above embodiments, the present disclosure also proposes a general operator architecture for optical neural networks.

[0052] like Figure 4 As shown, the optical neural network general operator architecture includes: a data input module, the optical neural network general operator model provided in the above embodiment, a collection module and a control module; wherein, A data input module, used for acquiring input data and encoding the input data into a one-dimensional time optical signal to obtain an input optical signal; A control module is used to reconstruct a reconfigurable timing coding module, a reconfigurable linear optical computing module, and a reconfigurable nonlinear activation module in the optical neural network general operator model based on a target optical computing task, so as to control the optical neural network general operator model to perform optical computing on an input optical signal based on the target optical computing task to obtain an optical computing result signal; The acquisition module is used to acquire the optical calculation result signal.

[0053] Optionally, the data input module is used to obtain input data and encode the input data into a one-dimensional time optical signal. When the input optical signal is obtained, it is specifically used to: The voltage of the electro-optic modulator is controlled by a high-speed arbitrary waveform generator to convert the input data into the intensity of the optical signal, thereby encoding the input data.

[0054] Optionally, when the acquisition module is used to acquire the optical calculation result signal, it is specifically used to: The photoelectric detector is controlled to collect the optical calculation result signal.

[0055] Optionally, the control module is used to reconstruct the reconfigurable temporal coding module, the reconfigurable linear optical computing module, and the reconfigurable nonlinear activation module in the optical neural network general operator model based on the target optical computing task, specifically for: Different timing coding parameters are achieved by adjusting the control voltage of the first microring resonator in the reconfigurable timing coding module; The optical computing matrix or convolution kernel weight is changed by adjusting the voltage of the modulator in the reconfigurable linear optical computing module; Different nonlinear activation functions are achieved by adjusting the control voltage of the second microring resonator in the reconfigurable nonlinear activation module.

[0056] Optionally, the control module is further configured to: Control the working status of the data input module and the acquisition module.

[0057] In summary, the architecture provided in this embodiment, by adopting a universal operator model for optical neural networks, can provide a standardized platform for optical neural networks, can more easily perform performance optimization and function expansion, can be better compatible with existing computing systems and architectures, and reduce the complexity and cost of system integration.

[0058] In order to implement the above embodiments, the present disclosure also proposes an optical neural network general operator system, including: the optical neural network general operator architecture provided by the above embodiments.

[0059] The collection, storage, use, processing, transmission, provision and disclosure of user personal information involved in this disclosure shall comply with the relevant laws and regulations and shall not violate public order and good morals.

[0060] It should be noted that personal information from users should be collected for legitimate and reasonable purposes and should not be shared or sold outside of these legitimate uses. In addition, such collection / sharing should be carried out after receiving the user's informed consent, including but not limited to notifying the user to read the user agreement / user notice and sign the agreement / authorization including authorization of relevant user information before the user uses the function. In addition, any necessary steps should be taken to protect and safeguard access to such personal information data and ensure that others who have access to personal information data comply with its privacy policy and procedures.

[0061] The present disclosure anticipates providing implementation schemes for users to selectively block the use or access of personal information data. That is, the present disclosure anticipates providing hardware and / or software to prevent or block access to such personal information data. Once the personal information data is no longer needed, risks can be minimized by limiting data collection and deleting the data. In addition, when applicable, such personal information is de-identified to protect the privacy of the user.

[0062] The acquisition, transmission, storage, use, and processing of data in the technical solution disclosed in this disclosure are in compliance with the relevant provisions of national laws and regulations.

[0063] It should be noted that in the embodiments of the present disclosure, certain software, components, models and other existing solutions in the industry may be mentioned, which should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the solution.

[0064] In the description of the aforementioned embodiments, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they contradict each other.

[0065] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the present disclosure, "plurality" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0066] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present disclosure includes additional implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present disclosure belong.

[0067] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or otherwise processing in a suitable manner if necessary, and then stored in a computer memory.

[0068] It should be understood that the various parts of the present disclosure can be implemented in hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0069] A person skilled in the art may understand that all or part of the steps in the above-mentioned embodiment method may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0070] In addition, each functional unit in each embodiment of the present disclosure may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0071] The storage medium mentioned above may be a read-only memory, a disk or an optical disk, etc. Although the embodiments of the present disclosure have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations of the present disclosure. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present disclosure.

Claims

1. A general operator model of an optical neural network, characterized in that: include: A reconfigurable timing coding module is used to obtain an input optical signal and perform timing coding on the input optical signal based on a timing coding parameter corresponding to a target optical computing task to obtain a time interleaved signal; A reconfigurable linear optical computing module, used for performing optical computing on the time-interleaved signal based on an optical computing matrix corresponding to the target optical computing task to obtain a calculated optical signal; The reconfigurable nonlinear activation module is used to perform nonlinear activation on the calculated optical signal based on the nonlinear activation function corresponding to the target optical computing task, and obtain and output the optical computing result signal.

2. The model according to claim 1, characterized in that The timing coding parameters include the number of split beams and the delay time. The reconfigurable timing coding module is used to perform timing coding on the input optical signal based on the timing coding parameters corresponding to the target optical computing task to obtain a time interleaved signal. Specifically, it is used to: Based on the number of split beams, the input optical signal is split to obtain a plurality of split beams; Based on the delay duration, the split light beams in the multiple split light beams are delayed so that the delay between adjacent split light beams is the duration corresponding to a data symbol in the input optical signal, thereby obtaining a time-interleaved signal, wherein the time-interleaved signal includes multiple time-interleaved sub-signals.

3. The model according to claim 2, characterized in that When the reconfigurable timing coding module is used to split the input optical signal, it is specifically used to: The beam splitter is controlled to split the input optical signal.

4. The model according to claim 2, characterized in that The reconfigurable timing coding module is used to delay the split beams of the multiple split beams based on the delay duration, specifically for: Selecting a first microring resonator subset corresponding to the number of split beams from the first microring resonator set, and determining a control voltage corresponding to each first microring resonator in the first microring resonator subset; The split light is input to the first microring resonator subset to control the first microring resonator subset to delay the split light.

5. The model according to claim 1, characterized in that The time-interleaved signal includes a plurality of time-interleaved sub-signals, and the reconfigurable linear optical computing module is used to perform optical computing on the time-interleaved signal based on the optical computing matrix corresponding to the target optical computing task to obtain the calculated optical signal, specifically for: In the process of performing optical calculation on the time-interleaved signal based on the optical calculation matrix corresponding to the target optical calculation task, the columns in the matrix correspond one-to-one to the time-interleaved sub-signals, the multiple time-interleaved sub-signals input at the same time are multiplied one-to-one with the row elements in the rows corresponding to the multiple time-interleaved sub-signals input at the same time, and the multiple products obtained after the multiplication are added to obtain the calculated optical signal.

6. The model according to claim 5, characterized in that The reconfigurable linear optical computing module is used to perform optical computing on the time-interleaved signal based on the optical computing matrix corresponding to the target optical computing task to obtain the calculated optical signal, specifically for: Based on the target optical computing task, determining a coupling coefficient corresponding to each modulator in a modulator array to obtain an optical computing matrix corresponding to the target optical computing task, wherein the modulator is at least one of a Mach-Zehnder interferometer or a waveguide coupler; The time-interleaved signal is input into the modulator array for optical calculation to obtain a calculated optical signal.

7. The model according to claim 1, characterized in that The reconfigurable nonlinear activation module is used to perform nonlinear activation on the calculated optical signal based on the nonlinear activation function corresponding to the target optical computing task, and obtain and output the optical computing result signal, including: Selecting a second microring resonator corresponding to the nonlinear activation function corresponding to the target optical computing task from the second microring resonator set to obtain a second microring resonator subset, and determining a control voltage corresponding to each second microring resonator in the second microring resonator subset; The calculated optical signal is input to the second microring resonator set to control the second microring resonator set to perform nonlinear activation on the calculated optical signal, thereby obtaining and outputting an optical calculation result signal.

8. A general operator architecture for optical neural networks, characterized in that: include: A data input module, a general operator model of an optical neural network according to any one of claims 1 to 7, a collection module and a control module; wherein, The data input module is used to obtain input data and encode the input data into a one-dimensional time optical signal to obtain an input optical signal; The control module is used to reconstruct the reconfigurable timing coding module, the reconfigurable linear optical computing module, and the reconfigurable nonlinear activation module in the optical neural network general operator model based on the target optical computing task, so as to control the optical neural network general operator model to perform optical computing on the input optical signal based on the target optical computing task to obtain an optical computing result signal; The acquisition module is used to acquire the optical calculation result signal.

9. The architecture according to claim 8, characterized in that The control module is also used for: The working states of the data input module and the acquisition module are controlled.

10. An optical neural network universal operator system, characterized in that: include: At least one optical neural network universal operator architecture as described in any one of claims 8 to 9.

Images