Multi-mode optical matrix calculation system and method, storage medium and equipment

Abstract

The embodiment of the invention discloses a multi-mode optical matrix calculation system, and the system comprises a waveguide input module which is used for receiving an optical signal of fundamental mode coding; the multi-mode demultiplexer is connected with the waveguide input module and is used for converting an input optical signal into M high-order mode optical signals; the multi-mode waveguide module is connected with the multi-mode demultiplexer and is used for transmitting the M high-order mode optical signals; the topological multimode interference coupling device is provided with N input ports and is used for receiving the M high-order mode optical signals from the multimode waveguide module and carrying out MN * MN unitary matrix operation on the M high-order mode optical signals through a topological density optimization algorithm to obtain the high-order mode optical signals; the multi-mode dedivision multiplexer is connected with the topology multi-mode interference coupling device and is used for converting the high-order mode optical signal into a fundamental mode optical signal; and the waveguide output module is connected with the multi-mode dedivision multiplexer and is used for outputting the converted fundamental mode optical signal.

Description

Technical Field

[0001] The present invention relates to the field of optical computing technology, and particularly to a multi-mode optical matrix computing system, method, storage medium, and device. Background Art

[0002] In the fields of high-speed optical communication, optical interconnection, and optical integration, optical modulation technology is widely used to implement complex matrix operations, thus promoting the development of optical computing. Photons can carry information among multiple dimensions of space, phase, frequency, and mode, and thus can achieve large-scale parallel computing.

[0003] Traditional optical computing is implemented by an optical waveguide network, and matrix multiplication operations are completed by using the coupling between waveguides. The typical optical computing architectures applicable to on-chip are mainly two types: those based on Mach-Zehnder interferometer devices and those based on micro-ring resonator devices. Among them, for the Mach-Zehnder architecture, in order to ensure the coupling efficiency, an interference arm with sufficient length is required, so the limitation in device size restricts the integration of this architecture; while the micro-ring architecture is based on the resonance principle and uses lights of different wavelengths for computing, so the coherence and stability among multiple light sources in the architecture need to be considered, which increases the complexity and cost of the system.

[0004] Due to the above limitations, traditional optical matrix computing units face problems of low integration and limited precision, and it is difficult to meet the requirements of future high-performance computing. Summary of the Invention

[0005] Based on this, in view of the above problems, it is necessary to propose a multi-mode optical matrix computing system.

[0006] A multi-mode optical matrix computing system, the system includes:

[0007] A waveguide input module, configured to receive an optical signal encoded in a fundamental mode;

[0008] A multi-mode demultiplexer, connected to the waveguide input module, and configured to convert the input optical signal into M high-order mode optical signals;

[0009] A multi-mode waveguide module, connected to the multi-mode demultiplexer, and configured to transmit the M high-order mode optical signals;

[0010] A topological multi-mode interference coupling device, having N input ports, configured to receive the M high-order mode optical signals from the multi-mode waveguide module, and perform an MN×MN unitary matrix operation on the M high-order mode optical signals through a topological density optimization algorithm to obtain high-order mode optical signals;

[0011] A multi-mode demultiplexer, connected to the topological multi-mode interference coupling device, and configured to convert the high-order mode optical signals into fundamental mode optical signals;

[0012] A waveguide output module, connected to the multi-mode demultiplexer, for outputting the converted fundamental mode optical signal.

[0013] In the above solution, the size of the multi-mode waveguide is larger than that of the input waveguide, and the optical signal of the input waveguide is a silicon-based optical signal.

[0014] In the above solution, the M high-order mode optical signals correspond to M ports. When performing a unitary matrix operation of MN×MN on the M high-order mode optical signals, the number of N input ports > the number of M ports.

[0015] In the above solution, the number of ports of the multi-mode demultiplexer and the multi-mode demultiplexer corresponds to the number of ports of the topological multi-mode interference coupling device.

[0016] In the above solution, the spacing of the multi-mode waveguides of the topological multi-mode interference coupling device is determined according to the arrangement mode of the demultiplexing device, and the arrangement mode of the demultiplexing device includes columnar arrangement and mirror arrangement.

[0017] In the above solution, the spacing of the multi-mode waveguides of the topological multi-mode interference coupling device is determined according to the arrangement mode of the demultiplexing device:

[0018] When N is an even number, the spacing between waveguides is compressed by mirror arrangement;

[0019] For the case of M×N>K, the arrangement mode of the demultiplexing device is M×(N - 1)+X = K, where X is a single-mode waveguide, and is interspersed between the back-to-back arranged demultiplexing devices.

[0020] In the above solution, the conversion of the input optical signal into M high-order mode optical signals specifically includes:

[0021] According to the sequence numbers of the input channels, the corresponding input optical signals are respectively converted into M modes of mutually orthogonal TE0, TE1, TE2 to TE (M-1) M.

[0022] The present application also proposes a multi-mode optical matrix calculation method, and the method includes:

[0023] Receiving an optical signal encoded in the fundamental mode;

[0024] Converting the input optical signal into M high-order mode optical signals;

[0025] Performing a unitary matrix operation of MN×MN on the M high-order mode optical signals through a topological density optimization algorithm to obtain high-order mode optical signals;

[0026] Converting the high-order mode optical signals into fundamental mode optical signals.

[0027] The present application also provides a readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the following steps:

[0028] Receive a fundamental mode encoded optical signal;

[0029] Convert the input optical signal into M high-order mode optical signals;

[0030] Perform an MN×MN unitary matrix operation on the M high-order mode optical signals through a topological density optimization algorithm to obtain high-order mode optical signals;

[0031] Convert the high-order mode optical signals into fundamental mode optical signals.

[0032] The present application also provides a computer device, including a memory and a processor, where the memory stores a computer program, which, when executed by the processor, causes the processor to perform the following steps:

[0033] Receive a fundamental mode encoded optical signal;

[0034] Convert the input optical signal into M high-order mode optical signals;

[0035] Perform an MN×MN unitary matrix operation on the M high-order mode optical signals through a topological density optimization algorithm to obtain high-order mode optical signals;

[0036] Convert the high-order mode optical signals into fundamental mode optical signals.

[0037] Adopting the embodiments of the present invention has the following beneficial effects: The present invention converts a fundamental mode optical signal into a high-order mode optical signal, realizes a larger amount of calculation under a similar device size, thereby significantly improving the integration of the device; then optimizes the structure of the multimode interference coupling device through a topological optimization algorithm, effectively reducing optical loss and mode crosstalk, thereby improving the accuracy and stability of the calculation. Finally, the high-order mode optical signals are reconverted into fundamental mode optical signals through a multimode demultiplexer, avoiding the high-loss problem in the multimode waveguide and improving the signal transmission efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0039] Among them:

[0040] Figure 1This is a structural schematic diagram of the present invention.

[0041] Figure 2 This is a schematic diagram of the placement manner of the demultiplexing device in the calculation unit in the present invention.

[0042] Figure 3 This is a system structure diagram in the application scenario of an optical neural network.

[0043] Figure 4 This is a network architecture diagram of a TE0TE1 dual-mode topological optical matrix calculation unit for implementing the 3x3 convolution kernel calculation function.

[0044] Figure 5 This is a network architecture diagram of a cascaded structure of Mach-Zehnder interferometers of the same scale;

[0045] Figure 6 This is a flowchart of a multi-mode optical matrix calculation method in an embodiment.

[0046] Explanation of Reference Numerals

[0047] 1: Waveguide input module; 2: Multi-mode demultiplexer; 3: Multi-mode waveguide module; 4: Topological multi-mode interference coupling device; 5: Multi-mode demultiplexer; 6: Waveguide output module. 7: Side-by-side arrangement. 8: Mirror arrangement. 9: Topological linear matrix calculation unit. 10: Pooling layer. 11: Fully connected layer. 12: Schematic diagram of a 3x3 convolution kernel topological optical matrix network. 13: Dual-mode topological optical matrix unit. 14: Schematic diagram of a 3x3 convolution kernel Mach-Zehnder interferometer device network. 15: Mach-Zehnder interferometer device unit. Detailed implementation manners

[0048] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts belong to the scope of protection of the present invention.

[0049] In the following description, a large number of specific details are given to provide a more thorough understanding of the present invention; however, it is obvious to those skilled in the art that the present invention can be implemented without one or more of these details; in other examples, in order to avoid confusion with the present invention, some technical features well known to the art are not described. It should be understood that the present invention can be implemented in different forms and should not be construed as limited to the embodiments presented here; on the contrary, providing these embodiments will make the disclosure thorough and complete, and will fully convey the scope of the present invention to those skilled in the art.

[0050] The purpose of the terms used herein is only to describe specific embodiments and is not a limitation of the present invention. As used herein, the singular forms "a", "an" and "the" are also intended to include the plural forms unless the context clearly dictates otherwise. It should also be understood that the terms "comprising" and / or "including", when used in this specification, identify the presence of the stated features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups. As used herein, the term "and / or" includes any and all combinations of the related listed items.

[0051] To thoroughly understand the present invention, detailed structures will be presented in the following description to illustrate the technical solutions proposed by the present invention; the optional embodiments of the present invention are described in detail below. However, in addition to these detailed descriptions, the present invention may also have other embodiments.

[0052] The present invention provides a multi-mode optical matrix calculation system and its method, aiming to achieve efficient matrix calculation through the processing of multi-mode optical signals. The system includes a waveguide input module, a multi-mode demultiplexer, a multi-mode waveguide module, a topological multi-mode interference coupling device, a multi-mode demultiplexer, and a waveguide output module. The system can receive a fundamental-mode encoded optical signal, convert it into a high-order mode optical signal for processing, and finally restore it to a fundamental-mode optical signal for output.

[0053] As Figure 1 shown, in one embodiment, a multi-mode optical matrix calculation system is provided. The multi-mode optical matrix calculation system includes:

[0054] (1) A waveguide input module for receiving a fundamental-mode encoded optical signal;

[0055] This module is responsible for receiving a fundamental-mode encoded optical signal. The fundamental mode is the most stable transmission mode in an optical waveguide, having the lowest loss and the longest transmission distance. Through the waveguide input module, external optical signals can be efficiently introduced into the system, providing a stable input source for subsequent processing.

[0056] Preferably, a silicon waveguide fabricated using an SOI (Silicon on Insulator) process can be used, and its width is typically 0.5 μm.

[0057] (2) A multi-mode demultiplexer connected to the waveguide input module for converting the input optical signal into M high-order mode optical signals;

[0058] As Figure 2As shown, the multi-mode demultiplexer is connected to the waveguide input module. The function of the demultiplexer is to convert the input fundamental mode optical signal into M high-order mode optical signals. The high-order mode optical signals have a more complex field distribution in the waveguide and can carry more information. Through this conversion, the system can make full use of the transmission capacity of the optical waveguide to improve the capacity and efficiency of information transmission.

[0059] Preferably, optimization based on the topological level set algorithm can ensure that the conversion rates of different modes are nearly equal, so as to maintain the relative optical intensity relationship between channels during output.

[0060] (3) Multi-mode waveguide module, connected to the multi-mode demultiplexer, for transmitting M high-order mode optical signals;

[0061] Since the transmission characteristics of high-order mode optical signals in the waveguide are different from those of the fundamental mode, the multi-mode waveguide module needs to be designed wide and long enough to ensure that these modes can be stably transmitted without serious mode coupling or loss.

[0062] Preferably, this application uses a silicon waveguide (such as 0.9um) with a width greater than that of the fundamental mode waveguide to ensure the effective propagation of high-order modes.

[0063] (4) Topological multi-mode interference coupling device, having N input ports, for receiving M high-order mode optical signals from the multi-mode waveguide module, and performing an MN×MN unitary matrix operation on the M high-order mode optical signals through the topological density optimization algorithm to obtain high-order mode optical signals;

[0064] Specifically, the device has N input ports for receiving M high-order mode optical signals from the multi-mode waveguide module. The core part adopts an irregular design area to cause strong scattering and interference of the optical information of different ports and modes, realizing vector matrix multiplication. Its core function is to perform an MN×MN unitary matrix operation on the M high-order mode optical signals through the topological density optimization algorithm. The unitary matrix operation can maintain the energy conservation and phase relationship of the optical signals, realizing complex mode interference and conversion. Through this processing, the system can perform arbitrary linear combinations and regulations on the input high-order mode optical signals to realize specific optical signal processing functions.

[0065] (5) Multi-mode demultiplexer, connected to the topological multi-mode interference coupling device, for converting high-order mode optical signals into fundamental mode optical signals;

[0066] It is connected to a topological multimode interference coupling device. This demultiplexer is responsible for converting the high-order mode optical signal after interference processing back into a fundamental mode optical signal. This process is the inverse process of mode conversion, aiming to restore an optical signal form that is easy to transmit and detect. Through the multimode demultiplexer, the system can convert the complex interference result into a simple fundamental mode output, facilitating subsequent processing and applications.

[0067] (6) Waveguide output module, connected to the multimode demultiplexer, for outputting the converted fundamental mode optical signal.

[0068] This module is responsible for outputting the converted fundamental mode optical signal. Through the waveguide output module, the system can efficiently transmit the processed optical signal to external devices or systems, realizing the complete process of optical signal reception, processing, and transmission.

[0069] In some embodiments, the size of the multimode waveguide is larger than that of the input waveguide, and the optical signal of the input waveguide is a silicon-based optical signal.

[0070] In some embodiments, M high-order mode optical signals correspond to M ports. When performing a unitary matrix operation of MN×MN on M high-order mode optical signals, the number of N input ports > M ports.

[0071] In some embodiments, the number of ports of the multimode demultiplexer and the multimode demultiplexer corresponds to the number of ports of the topological multimode interference coupling device.

[0072] For the K×K unitary matrix calculation unit to be implemented, it is necessary to first satisfy M×N≥K to provide sufficient calculation channels. When selecting M and N within the satisfied range, try to make N>M as much as possible within the allowable range. Considering that M corresponds to the total number of modes involved in the calculation, the larger M is, the more complex the objective function of optimizing the demultiplexing devices (2)(5) is, and the corresponding device performance is also relatively poor. At the same time, when the number of N is too large, since the number of N ports is related to the size of the central topological multimode interference coupling device (4), too many N ports require a larger central design area to carry, which may lead to waste of the design area.

[0073] In some embodiments, the pitch of the multimode waveguide of the topological multimode interference coupling device is determined according to the arrangement mode of the demultiplexing device, and the arrangement mode of the demultiplexing device includes columnar arrangement and mirror arrangement.

[0074] The function of the demultiplexing device is to convert the optical information of the fundamental mode into mode light of different orders in the same channel. Since the encoded information is reflected in the optical intensity of different input signals in optical computing, and the final information output is a normalization calculation of all detected optical powers. If the conversion rate difference of the demultiplexing device for different modes is too high, it will seriously affect the relative relationship of the optical intensity between the original channels, resulting in calculation errors. If only the conversion rates of different modes are considered to be similar, but the optical propagation loss is too high, then it is difficult to detect the light after passing through several computing units.

[0075] Therefore, based on topology optimization, the problems to be considered can be used as optimization objectives, and a more balanced optimization result can be found among different objectives, so as to approximate the most ideal structure as much as possible. In addition, for a specified wavelength, the optimized demultiplexing device has universality and can be used for any unitary matrix calculation with the same number of ports.

[0076] In some embodiments, the spacing of the multimode waveguides of the topological multimode interference coupling device is determined according to the arrangement mode of the demultiplexing device:

[0077] When N is even, the spacing between waveguides is compressed by the mirror-type arrangement;

[0078] For the case of M×N>K, the arrangement mode of the demultiplexing device is M×(N - 1)+X = K, where X is a single-mode waveguide, which is inserted between the demultiplexing devices arranged in the opposite direction.

[0079] The size and structure of the multimode interference coupling device (4) are limited by the shapes of the demultiplexing devices (2)(5) at the front and rear ends. Since the demultiplexing device is usually an asymmetric device, the arrangement mode of the device will affect the spacing of the multimode waveguides (3). There are usually two modes: the parallel arrangement (7) and the mirror-type arrangement (8). When N is even, the spacing between waveguides can be further compressed by the mirror-type arrangement; for the case of M×N>K, since there are some unused mode channels, and due to the asymmetric structure of the demultiplexing device, there will be a paired adjacent structure during the mirror arrangement, but the spacing between adjacent paired structures is large. It can be flexibly adjusted to the arrangement mode of M×(N - 1)+X = K as appropriate. Where X is a single-mode waveguide, which is inserted between the demultiplexing devices arranged in the opposite direction. This special distribution mode can further compress the size of the overall system in a specific scenario.

[0080] In some embodiments, the input optical signal is converted into M high-order mode optical signals, specifically including:

[0081] According to the sequence numbers of the input channels, the corresponding input optical signals are respectively converted into M modes of mutually orthogonal TE0, TE1, TE2 to TE (M-1) of.

[0082] Preferably, according to singular value decomposition, any real-valued matrix A can be simplified by decomposing it into the following form:

[0083] A = UΣV *

[0084] where Σ is a diagonal matrix without negative numbers. In the implementation of optical devices, the numerical encoding of each row corresponds to an optical channel. Therefore, the physical meaning of Σ is to perform independent intensity modulation on the optical information of each encoding, which can be achieved by a Mach-Zehnder interferometer or an optical attenuator / amplifier. And U and V are unitary matrices, and the optical information between different encoding channels will interact with each other. This part of optical calculation can be achieved by the interference between optical signals.

[0085] To achieve the above purpose of constructing an arbitrary unitary matrix calculation, the optical calculation unit module of the present invention is composed of a multi-mode demultiplexer for information encoding at both ends and a topological multi-mode interference coupling device for mode power distribution in the middle part.

[0086] Among them, the fundamental mode optical signal carrying information after encoding enters the calculation unit through the input waveguide module (1). The input signal is usually encoded and decoded only with amplitude information, but in some special cases, amplitude and phase information are also used simultaneously for optical calculation in the complex domain. The matrix calculations of both encoding rules are applicable to the present invention. The fundamental mode optical signals input from M channels are combined by the multi-mode demultiplexer (2), and the corresponding input signals are respectively converted into M modes of TE0, TE1, TE2 to TE(M-1) that are orthogonal to each other according to the sequence number of the input channels. The multi-mode demultiplexer (2) is optimized and iterated by the level set algorithm in the topological algorithm. In each round of iteration, according to the gradient, the initial structure contour is evolved under certain constraints to achieve local inward contraction or outward expansion, and finally the optical loss of each output mode in the device can be reduced. These modes will enter the multi-mode waveguide module (3) and propagate simultaneously. The size of the multi-mode waveguide needs to be larger than the input fundamental mode waveguide because the higher-order modes have a larger cut-off width. Taking a silicon-based waveguide as an example, the width of a single-mode waveguide is usually 0.5um, while a dual-mode waveguide including the fundamental mode and the first-order mode requires a width of 0.9um. This port also serves as an input port in the topological multi-mode interference coupling device (4) and performs unitary calculation together with multiple modes of the remaining ports.

[0087] The topological multimode interference coupling device (4) located at the core position has N multimode waveguides as input ports, enabling this computing unit to accommodate up to MN input channels, thereby realizing the MN×MN unitary matrix operation. After the optical information of different ports and modes passes through the irregular design area obtained by the topological density optimization algorithm, strong scattering and interference occur among them, realizing the redistribution of the optical power on different modes and different ports. This process completes the vector matrix multiplication. Finally, the optical signal also needs to pass through the multimode demultiplexer structure (5) symmetrical to the front end to re-decode the optical information of different modes into the form of the fundamental mode and output it to MN waveguides.

[0088] In this process, the mode light is only used to save the design space in matrix calculation to carry a larger amount of data. During the propagation of the optical signal from the input waveguide on the left to the output waveguide module (6) on the rightmost end, its encoding, decoding, and signal modulation are ultimately reflected in the fundamental mode. Therefore, this computing unit has good scalability and can achieve good coupling with other single-mode optical devices or computing modules. The reason for demultiplexing the mode through structures (2) and (5) instead of using multimode waveguides to carry signals throughout is that multimode optical signals mainly propagate in the optical waveguide in the form of total internal reflection, which will cause dissipation and distortion of the optical signal due to multiple reflection losses, while single-mode optical signals propagate in a straight line. Therefore, using the fundamental mode waveguide when realizing matrix calculation and propagating to the next unit can have lower losses.

[0089] Preferably, as Figure 3 shown, as the convolution kernel of a specific optical neural network, starting from the first layer of the network trained in the optical network simulation model, layer-by-layer training is carried out to obtain the specific parameters of each layer of the linear computing unit. Due to the error caused by optical loss, the actual unitary matrix optimized in the previous layer needs to be reused to correct the weights of the convolution kernel in the subsequent layer for retraining. For the convolutional optical neural network structure, the computing unit after topological optimization serves as the convolution kernel of linear calculation - the topological linear matrix computing unit (9). After downsampling through the pooling layer (10), it enters the convolution kernel of the next level. Taking (9) and (10) as a whole, iterative calculation is carried out from front to back, using the actual optimized matrix vector of the previous layer as the updated weight to calibrate the parameters of the next layer, which can reduce the large deviation of the training result caused by cumulative error. Finally, through the fully connected layer (11), the features extracted from the optical network are classified and mapped to the corresponding classification targets. This optimization method can keep the actual optical network with a high accuracy rate.

[0090] In summary, based on the coding method of multimode interference coupling, the present invention introduces multimode source coding at the ports and uses reverse design to deduce the topological structure of the unitary matrix calculation part, greatly compressing the size of the calculation unit. For the traditional cascaded system of Mach-Zehnder interference devices, to implement matrix calculations of size NM×NM through rectangular network structures and triangular network structures, respectively, it requires (NM) 2 and NM(NM - 1) / 2 Mach-Zehnder interference devices. However, the cascading method of the present invention only requires 2M + 1 topological structure units, further increasing the integration of device units.

[0091] As Figure 4 shown, in some embodiments, the present invention uses a 3×3 convolution kernel to implement the calculation of a 9×9 unitary matrix. In this design, a dual-mode demultiplexing device is used to encode 9 fundamental modes into a dual-mode interference coupling device with five input-output channels for matrix calculation.

[0092] Waveguide input module: The waveguide input module (1) is fabricated using the silicon-on-insulator process and is a silicon waveguide with a width of 0.5um. The thickness of the following devices is defaulted to 220nm.

[0093] Multimode demultiplexer: The design of the multimode demultiplexer (2) is optimized based on the topological level set algorithm to ensure that the conversion rates of different modes are nearly equal, thereby maintaining the relative light intensity relationship between channels during output. The structure of this demultiplexer will optimize TE0→TE0 and TE0→TE1 to minimize optical loss and reduce crosstalk between the two modes.

[0094] Multimode waveguide module: A silicon waveguide with a width of 0.9um to ensure the effective propagation of two modes.

[0095] Topological multimode interference coupling device: The topological multimode interference coupling device (4) located at the core position has 5 input-output ports, where 4 are dual-mode waveguides and 1 is a single-mode waveguide. It adopts a mirror-type structure arrangement, with the two dual-mode demultiplexers on both sides arranged facing each other, and a single-mode waveguide is inserted into the gap arranged back-to-back in the middle. This arrangement fully utilizes the design space of the calculation unit, making the structure very compact. After the optical information of different ports and modes passes through the irregular design area obtained by the topological density optimization algorithm, strong scattering and interference occur between them, completing the vector matrix multiplication.

[0096] Output demultiplexer: Finally, the optical signal passes through a multimode demultiplexing structure (5) symmetric to the front end to re-decode the optical information of different modes into 9 fundamental mode waveguides, ensuring the accuracy and effectiveness of the output signal.

[0097] The network structure (12) of the 3x3 convolution kernel computing device (13) designed as above has much lower complexity than the network structure (14) based on Mach-Zehnder interferometer devices of the same scale. As Figure 5 shown, the Mach-Zehnder interferometer network (14) requires (n2-1)! units of device (15) to implement an nxn convolution kernel. While the topologically optimized unit device only requires (2n+1) devices, having a great advantage in integration.

[0098] The above-optimized optical convolution computing unit can be applied to specific optical neural network scenarios for batch training and optimization. Through layer-by-layer optimization of the trained first-layer network, the error caused by optical loss is reduced, thus maintaining a high accuracy.

[0099] As Figure 6 shown, the present application also proposes a multi-mode optical matrix calculation method, which includes:

[0100] Receiving a base-mode encoded optical signal;

[0101] Converting the input optical signal into M high-order mode optical signals;

[0102] Performing an MN×MN unitary matrix operation on the M high-order mode optical signals through a topological density optimization algorithm to obtain high-order mode optical signals;

[0103] Converting the high-order mode optical signals into base-mode optical signals.

[0104] The present application also proposes a readable storage medium storing a computer program, which when executed by a processor, causes the processor to perform the following steps:

[0105] Receiving a base-mode encoded optical signal;

[0106] Converting the input optical signal into M high-order mode optical signals;

[0107] Performing an MN×MN unitary matrix operation on the M high-order mode optical signals through a topological density optimization algorithm to obtain high-order mode optical signals;

[0108] Converting the high-order mode optical signals into base-mode optical signals.

[0109] A computer device includes a memory and a processor. When the computer program stored in the memory is executed by the processor, the processor is caused to perform the following steps:

[0110] Receiving a base-mode encoded optical signal;

[0111] Converting the input optical signal into M high-order mode optical signals;

[0112] Perform the unitary matrix operation of MN×MN on M high-order mode optical signals through the topological density optimization algorithm to obtain high-order mode optical signals;

[0113] Convert the high-order mode optical signals into fundamental mode optical signals.

[0114] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in this application can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can 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 (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0115] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as these combinations of technical features do not conflict, they should be considered to be within the scope described in this specification.

[0116] The above-described embodiments merely represent several implementation manners of this application. Their descriptions are relatively specific and detailed, but they should not be construed as limiting the patent scope of this application. It should be noted that for those of ordinary skill in the art, without departing from the concept of this application, several modifications and improvements can still be made, and these all belong to the protection scope of this application. The above-disclosed is only the preferred embodiment of the present invention, and of course, it cannot be used to limit the scope of the rights of the present invention. Therefore, equivalent changes made according to the claims of the present invention still fall within the scope covered by the present invention.

Claims

1. A multi-mode optical matrix calculation system, characterized in that, The system includes: A waveguide input module for receiving an optical signal encoded in a fundamental mode; A multimode demultiplexer connected to the waveguide input module for converting the input optical signal into M high-order mode optical signals; A multimode waveguide module connected to the multimode demultiplexer for transmitting the M high-order mode optical signals; A topological multimode interference coupling device having N input ports for receiving the M high-order mode optical signals from the multimode waveguide module and performing an MN×MN unitary matrix operation on the M high-order mode optical signals through a topological density optimization algorithm to obtain high-order mode optical signals; A multimode demultiplexer connected to the topological multimode interference coupling device for converting the high-order mode optical signals into fundamental mode optical signals; A waveguide output module connected to the multimode demultiplexer for outputting the converted fundamental mode optical signals.

2. The multi-mode optical matrix calculation system according to claim 1, wherein The size of the multimode waveguide is larger than that of the input waveguide, and the optical signal of the input waveguide is a silicon-based optical signal.

3. The multi-mode optical matrix computing system according to claim 1, wherein The M high-order mode optical signals correspond to M ports. When performing an MN×MN unitary matrix operation on the M high-order mode optical signals, N input ports > M ports.

4. The multi-mode optical matrix calculation system according to claim 1, wherein The number of ports of the multimode demultiplexer and the multimode demultiplexer corresponds to the number of ports of the topological multimode interference coupling device.

5. The multi-mode optical matrix calculation system according to claim 1, characterized in that, Determine the pitch of the multimode optical waveguides of the topological multimode interference coupling device according to the arrangement mode of the demultiplexing device. The arrangement mode of the demultiplexing device includes columnar arrangement and mirror arrangement.

6. The multi-mode optical matrix calculation system according to claim 1, characterized in that, Determine the pitch of the multimode optical waveguides of the topological multimode interference coupling device according to the arrangement mode of the demultiplexing device: When N is an even number, compress the pitch between the waveguides through mirror arrangement; For the case of M×N>K, arrange the demultiplexing device in the manner of M×(N - 1)+X=K, where X is a single-mode waveguide and is interspersed between the demultiplexing devices arranged back to back.

7. The multi-mode optical matrix calculation system according to claim 1, characterized in that, The conversion of the input optical signal into M high-order mode optical signals specifically includes: According to the number of sequences of the input channels, the corresponding input optical signals are respectively converted into M modes of mutually orthogonal TE0, TE1, TE2 to TE (M-1) of.

8. A multi-mode optical matrix calculation method, characterized in that, The method includes: Receiving an optical signal encoded in a fundamental mode; Converting the input optical signal into M high-order mode optical signals; Performing an MN×MN unitary matrix operation on the M high-order mode optical signals through a topological density optimization algorithm to obtain high-order mode optical signals; Converting the high-order mode optical signals into fundamental mode optical signals.

9. A readable storage medium storing a computer program, which when executed by a processor causes the processor to execute the steps of the method according to any one of claims 7.

10. A computer device including a memory and a processor, the memory storing a computer program, which when executed by the processor causes the processor to execute the steps of the method according to any one of claims 7.

Images