Matrix calculation method, device and system based on optical calculation architecture and medium

Abstract

The invention discloses a matrix calculation method, device and system based on an optical calculation architecture and a medium, the optical calculation architecture comprises an input waveguide, light splitting units, an output waveguide and a photoelectric detector, and the light splitting units are arranged in multiple rows and multiple columns and each comprise an input port and two output ports. Two adjacent light splitting units in each row of light splitting units are connected, the first light splitting unit is connected with an input waveguide through an input port, each light splitting unit is connected to an output waveguide, and the output waveguides connected with each column of light splitting units are connected with the same photoelectric detector; determining the light splitting weight of each light splitting unit according to the light splitting ratio design parameter of each light splitting unit; respectively inputting each light splitting weight into a corresponding light splitting unit and respectively providing a to-be-calculated optical signal to each input waveguide; the light splitting unit inputs the light passing through the light splitting unit to the output waveguide connected with the light splitting unit according to the light splitting weight part; and obtaining a detection result of each photoelectric detector, and determining a matrix operation result of the to-be-calculated optical signal.

Description

Technical Field

[0001] The present application relates to the technical field, and in particular, to a matrix computing method, device, system and medium based on an optical computing architecture. Background Art

[0002] In recent years, artificial intelligence (AI) technology has developed rapidly 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 high energy efficiency is also increasing.

[0003] However, due to the existence of basic limitations such as Joule heating effect, electron tunneling effect, parasitic capacitance and radio frequency crosstalk, traditional electronic integrated circuits face increasing challenges in improving performance and energy efficiency, and it is becoming increasingly difficult to achieve this goal through semiconductor technology. Photonic computing has emerged as a promising solution. Photonic computing technology is an emerging technology that uses light instead of electricity for information processing. It uses physical phenomena such as interference, diffraction and polarization of light to achieve logical operations and mathematical calculations. It makes full use of 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.

[0004] Currently, large-scale cross-bar architecture is considered to be a method to easily construct large-scale computing matrices in photonic computing, such as using optical couplers and phase change materials to achieve a simple and smaller computing array. However, there are many problems such as high loss and poor signal-to-noise ratio when constructing large-scale computing architectures using optical couplers and phase change materials. Summary of the invention

[0005] 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 a matrix calculation method, device, system and medium based on an optical computing architecture.

[0006] As a first aspect of the present application, a matrix computing method based on an optical computing architecture is provided, wherein the optical computing architecture includes an input waveguide, a splitting unit, an output waveguide and a photodetector, each of the splitting units is arranged in multiple rows and columns, each of the splitting units includes an input port and two output ports, the first splitting unit in each row of the splitting units is connected to the input waveguide through the input port, two adjacent splitting units in each row of the splitting units are connected to the input port of the splitting unit cascaded behind through one of the output ports of the splitting unit cascaded in front, each of the splitting units is connected to the output waveguide through another output port, and the end of the output waveguide connected to the splitting units in each column is connected to the same photodetector; the method includes:

[0007] Determining the splitting weight of each of the splitting units according to the splitting ratio design parameters of each of the splitting units;

[0008] Input each of the splitting weights to the corresponding splitting unit, and provide the optical signal to be calculated to each of the input waveguides; wherein the splitting unit inputs the light passing through itself to the output waveguide connected to itself according to the splitting weight part;

[0009] Obtaining detection results of each of the photoelectric detectors;

[0010] The matrix operation result of the optical signal to be calculated is determined according to the detection results of each of the photoelectric detectors.

[0011] Optionally, the splitting weight of each splitting unit is determined by the following formula:

[0012]

[0013] c ij =b i a ij (2);

[0014]

[0015] In formulas (1), (2) and (3), K ij represents the splitting ratio design parameter of the splitting unit in the i-th column and the j-th row, a ij represents the spectral coefficient of the spectral unit in the i-th column and the j-th row, c ij represents the splitting weight of the splitting unit in the i-th column and the j-th row, b i represents the compensation coefficient of the i-th column of the splitter unit, m represents the total number of columns of the splitter unit, and “=const=” represents taking a constant value.

[0016] Optionally, the optical computing architecture includes a plurality of chips arranged in multiple rows and columns, the spectroscopic units are arranged in multiple rows and columns on each of the chips, the chips are connected via a bus waveguide, and the matrix computing scale of the optical computing architecture is hm×kn, wherein h represents the total number of columns of the chip, k represents the total number of rows of the chip, m represents the total number of columns of the spectroscopic units in the chip, and n represents the total number of rows of the spectroscopic units in the chip, and h, k, m, and n are all positive integers.

[0017] Optionally, providing the optical signal to be calculated to each of the input waveguides respectively comprises:

[0018] The optical frequency comb is input to the demultiplexer DeMUX, so that the demultiplexer DeMUX performs optical splitting processing on the optical frequency comb to obtain the optical signal to be calculated, and provides the optical signal to be calculated to each of the input waveguides respectively.

[0019] As a second aspect of the present application, an electronic device is provided, wherein the electronic device includes:

[0020] one or more processors;

[0021] A memory having one or more computer programs stored thereon, which, when executed by the one or more processors, enables the one or more processors to implement the matrix calculation method based on the optical computing architecture provided according to the first aspect of the present application.

[0022] As a third aspect of the present application, a matrix computing system is provided, wherein the matrix computing system comprises the electronic device and the optical computing architecture provided according to the second aspect of the present application, the optical computing architecture comprises an input waveguide, a splitting unit, an output waveguide and a photodetector, each of the splitting units is arranged in multiple rows and columns, each of the splitting units comprises an input port and two output ports, the first splitting unit in each row of the splitting units is connected to the input waveguide through the input port, two adjacent splitting units in each row of the splitting units are connected to the input port of the splitting unit cascaded in the rear through one of the output ports of the splitting unit cascaded in the front, each of the splitting units is connected to the output waveguide through another output port, and the end of the output waveguide connected to the splitting units in each column is connected to the same photodetector;

[0023] The spectrometer unit is used to input the light passing through itself into the output waveguide connected to itself according to the received spectrometer weight part, and each photodetector is used to detect the output light of the corresponding column of the spectrometer unit to obtain a detection result, and each detection result is used to determine the matrix operation result of the optical signal to be calculated input into the optical computing architecture.

[0024] Optionally, the output waveguides connected to any two light-splitting units in each column of the light-splitting units are different, and the ends of the output waveguides connected to the light-splitting units in each column are connected in parallel to the photodetector.

[0025] Optionally, the output waveguide is a bus waveguide, and the bus waveguide includes a plurality of incoherent combiners that are different in pairs, and the incoherent combiners connected to any two splitting units in each column of the splitting units are different.

[0026] Optionally, a beam splitter, a phase shifter and a coupler are cascaded in sequence between the one input port and the two output ports of the optical splitting unit.

[0027] 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 matrix calculation method based on the optical computing architecture provided according to the first aspect of the present application is implemented.

[0028] In the matrix computing method based on the optical computing architecture provided in the embodiment of the present application, an optical computing architecture including an input waveguide, a splitting unit, an output waveguide and a photodetector is first constructed, wherein each of the splitting units is arranged in multiple rows and columns, each of the splitting units includes an input port and two output ports, the first splitting unit in each row of the splitting units is connected to the input waveguide through the input port, and the two adjacent splitting units in each row of the splitting units are connected to the input port of the splitting unit cascaded behind through one of the output ports of the preceding splitting unit cascaded, and each of the splitting units is connected to the output waveguide through another output port, and the end of the output waveguide to which the splitting units in each column are connected is connected to the same photodetector. Furthermore, according to the splitting ratio design parameters of each splitting unit, the splitting weight of each splitting unit is determined; each splitting weight is input into the corresponding splitting unit, and the optical signal to be calculated is provided to each input waveguide; wherein the splitting unit inputs the light passing through itself into the output waveguide connected to itself according to the splitting weight part; the detection result of each photodetector is obtained; according to the detection result of each photodetector, the matrix operation result of the optical signal to be calculated is determined. By adjusting the row and column arrangement of the splitting units in the optical computing architecture and controlling the splitting weight of each splitting unit, large-scale matrix computing can be flexibly realized, and the light utilization rate of the entire optical computing architecture can be effectively improved, the signal-to-noise ratio can be improved, and the loss can be reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The present application is further described below in conjunction with the accompanying drawings:

[0030] Figure 1It is a flowchart of an implementation of a matrix calculation method based on an optical computing architecture provided in an embodiment of the present application;

[0031] Figure 2 is a schematic diagram of an implementation of an optical computing architecture provided in an embodiment of the present application;

[0032] Figure 3 is a schematic diagram of an implementation of a light splitting unit provided in an embodiment of the present application;

[0033] Figure 4a It is a schematic diagram of setting a bus waveguide provided in an embodiment of the present application;

[0034] Figure 4b is a schematic diagram of another implementation of the optical computing architecture provided in an embodiment of the present application;

[0035] Figure 5a is a schematic diagram of using multiple independent light sources to provide optical signals to be calculated provided by an embodiment of the present application;

[0036] Figure 5b is a schematic diagram of using an optical frequency comb to provide an optical signal to be calculated provided by an embodiment of the present application;

[0037] Figure 6 This is a flow chart of an implementation method of providing the optical signal to be calculated to each input waveguide respectively provided in an embodiment of the present application;

[0038] Figure 7 It is a module diagram of an implementation of an electronic device provided in an embodiment of the present application;

[0039] Figure 8 is a schematic diagram of using a beam combiner to transmit the output light of a light splitting unit to a photodetector provided in an embodiment of the present application;

[0040] Fig. 9 It is a schematic diagram of a computer-readable medium provided in an embodiment of the present application.

[0041] Description of Reference Numerals

[0042] 101: Processor 102: Memory

[0043] 103: I / O interface 104: bus DETAILED DESCRIPTION

[0044] 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.

[0045] 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.

[0046] Currently, the use of optical couplers and phase change materials to construct large-scale computing architectures has many problems such as high loss and poor signal-to-noise ratio. In view of this, the applicant of this application proposes that by using a splitter unit to construct a computing matrix and controlling the splitting weight of the splitter unit, the light utilization rate of the entire optical computing architecture can be effectively improved, the signal-to-noise ratio can be improved, and the loss can be reduced.

[0047] As a first aspect of an embodiment of the present application, a matrix computing method based on an optical computing architecture is provided, wherein the optical computing architecture includes an input waveguide, a splitting unit, an output waveguide and a photodetector, each of the splitting units is arranged in multiple rows and columns, each of the splitting units includes an input port and two output ports, the first splitting unit in each row of the splitting units is connected to the input waveguide through the input port, two adjacent splitting units in each row of the splitting units are connected to the input port of the splitting unit cascaded behind through one of the output ports of the preceding splitting unit cascaded, each of the splitting units is connected to the output waveguide through another output port, and the end of the output waveguide connected to the splitting units in each column is connected to the same photodetector; Figure 1 As shown, the method may include:

[0048] In step S110, the splitting weight of each splitting unit is determined according to the splitting ratio design parameters of each splitting unit;

[0049] In step S120, each of the splitting weights is input into a corresponding splitting unit, and the optical signal to be calculated is provided to each of the input waveguides; wherein the splitting unit inputs the light passing through itself into the output waveguide connected to itself according to the splitting weight part;

[0050] In step S130, the detection results of each of the photoelectric detectors are obtained;

[0051] In step S140, a matrix operation result of the optical signal to be calculated is determined according to the detection results of each of the photodetectors.

[0052] like Figure 2 FIG. 1 is a schematic diagram of an implementation of an optical computing architecture provided in an embodiment of the present application. In the optical computing architecture, each rectangle represents a light splitting unit. 11 、a 12 、a13 、a 14 、a 21 、a 22 、a 23 、a 24 、a 31 、a 32 、a 33 、a 34 、a 41 、a 42 、a 43 、a 44 The light splitting units are arranged in multiple rows and columns. The first light splitting unit in each row has a light splitting weight of a. 11 、a 12 、a 13 、a 14 The input ports of the four splitting units are all connected to input waveguides. In each row of splitting units, two adjacent splitting units are connected to the input port of the splitting unit cascaded in the front through one of the output ports of the splitting unit cascaded in the front, and the other output port of each splitting unit is connected to the output waveguide. For example, the splitting weight is a 11 、a 21 、a 31 、a 41 In this row of splitter units, the first splitter unit has a splitter weight a. 11 This splitter unit, through one of its output ports, is connected to the splitter with a splitting weight of a. 21 The end of the output waveguide connected to each column of the splitting unit is connected to the same photodetector. For example, the splitting weight is a 11 、a 12 、a 13 、a 14 The end of the output waveguide to which this column of splitting units is connected is connected to the same photodetector.

[0053] It should be noted that in order to distinguish the input waveguide and output waveguide in the optical splitting unit from the input waveguide and output waveguide in the optical computing architecture, the input waveguide and output waveguide in the optical splitting unit are respectively referred to as input ports and output ports, that is, the input ports and output ports of the optical splitting unit are actually input waveguides and output waveguides. When referring to input waveguides and output waveguides in the following text, they specifically refer to input waveguides and output waveguides in the optical computing architecture.

[0054] For the last column of splitting units, the splitting weight is a 41 、a 42 、a 43 、a44 One of the output ports of this column of spectroscopic units is connected to the output waveguide, and for the output light of the other output port, measures can be taken to make it be absorbed or directed out to space loss, thereby preventing the output light of the other output port from affecting the matrix calculation result due to reflection / refraction, etc.

[0055] by Figure 2 For example, the light intensity provided to the four input waveguides is represented as x1, x2, x3, and x4 respectively, and the light intensity detected by the photodetector at the end of the output waveguide connected to the first column of the splitting unit can be expressed as: y1 = a 11 *x1+a 12 *x2+a 13 *x3+a 14 *x4. Then for the entire optical computing architecture, the following matrix calculations can be achieved:

[0056] Herein, m and n represent the total number of columns and rows of the light splitting units, respectively.

[0057] The splitting weight matrix implemented by the optical computing architecture Represented as A, the matrix composed of the light intensity of the light signal to be calculated Expressed as X, the entire optical computing architecture implements Y=AX.

[0058] It is understandable that the embodiment of the present application does not impose any specific limitation on the number of rows and columns of the light splitting unit in the optical computing architecture, which needs to be determined based on the matrix A required in the actual application.

[0059] Matrix operation results for the entire optical computing architecture It needs to be determined based on the detection results of each photodetector. The detection results of each photodetector are usually expressed as electrical signals. How to process them to obtain matrix operation results There are already feasible ways to do this, and the embodiments of the present application do not impose any specific limitations on this.

[0060] In the matrix computing method based on the optical computing architecture provided in the embodiment of the present application, an optical computing architecture including an input waveguide, a splitting unit, an output waveguide and a photodetector is first constructed, wherein each of the splitting units is arranged in multiple rows and columns, each of the splitting units includes an input port and two output ports, the first splitting unit in each row of the splitting units is connected to the input waveguide through the input port, and the two adjacent splitting units in each row of the splitting units are connected to the input port of the splitting unit cascaded behind through one of the output ports of the preceding splitting unit cascaded, and each of the splitting units is connected to the output waveguide through another output port, and the end of the output waveguide to which the splitting units in each column are connected is connected to the same photodetector. Furthermore, according to the splitting ratio design parameters of each splitting unit, the splitting weight of each splitting unit is determined; each splitting weight is input into the corresponding splitting unit, and the optical signal to be calculated is provided to each input waveguide; wherein the splitting unit inputs the light passing through itself into the output waveguide connected to itself according to the splitting weight part; the detection result of each photodetector is obtained; according to the detection result of each photodetector, the matrix operation result of the optical signal to be calculated is determined. By adjusting the row and column arrangement of the splitting units in the optical computing architecture and controlling the splitting weight of each splitting unit, large-scale matrix computing can be flexibly realized, and the light utilization rate of the entire optical computing architecture can be effectively improved, the signal-to-noise ratio can be improved, and the loss can be reduced.

[0061] The optical splitter unit in the optical computing architecture can have multiple implementations, but generally includes an input port, a beam splitter, a phase shifter, a coupler, and an output port. Accordingly, in some embodiments, a beam splitter, a phase shifter, and a coupler are cascaded in sequence between the one input port and the two output ports of the optical splitter unit.

[0062] like Figure 3 As shown, a schematic diagram of an implementation of a light splitting unit provided in an embodiment of the present application is shown, wherein the light splitting unit includes an input port, a 1x2 beam splitter, two phase shifters, a 2x2 coupler and two output ports. Light enters the light splitting unit through the input port, and the 1x2 beam splitter splits the light into two beams, which are transmitted to the 2x2 coupler through the phase shifter. The phase shifter can adjust the relative phase of the two beams so that the two beams eventually produce an interference effect at the output port, thereby obtaining different optical power distribution effects. For example, the light splitting weight is a 11 This spectroscopic unit will be a 11 The light is output to the output waveguide to which it is connected, and the ratio is (1-a 11 ) is given a splitting weight of a 21 This spectroscopic unit.

[0063] In another embodiment, the optical splitter unit may include an input port, a 2x2 beam splitter, two phase shifters, a 2x2 coupler, and two output ports. Figure 3 The 1x2 beam splitter between the input port and the phase shifter in the optical splitter unit shown is replaced by a 2x2 beam splitter.

[0064] In the embodiment of the present application, after the determined splitting weight is input into the splitting unit, the phase shifter in the splitting unit adjusts the splitting weight from the initial value to the determined value through thermal adjustment, electrical adjustment or other adjustment methods.

[0065] In the optical computing architecture provided in the embodiment of the present application, the splitting ratio of each splitting unit can be expressed as Among them, a ij Represents the splitting coefficient of the splitting unit in the i-th column and the j-th row. It can be seen that the splitting ratio of the splitting unit is related to its own splitting weight and the splitting weight of the splitting unit cascaded behind it. The closer the splitting unit is to the input waveguide, the smaller the splitting ratio it has, and the more difficult it is to achieve precise control. The applicant of the present application further proposes that the splitting ratio of each splitting unit can be precisely controlled by precisely controlling the splitting weight of each splitting unit. Specifically, the matrix composed of the splitting weights of each splitting unit is Each row (i.e., the splitting weight of each column of the splitting unit in the optical computing architecture) is multiplied by a compensation coefficient to obtain:

[0066] Among them, b1, b2, ..., b n represents the compensation coefficient. The compensation coefficient of each column of the optical computing architecture is the same. Then the splitting ratio of the splitting unit becomes Furthermore, by combining b1, b2, ..., b n Set to meet b1>b2>...>b n , the splitting ratio of the splitting unit close to the input waveguide can be improved, and precise splitting ratio control can be achieved.

[0067] Furthermore, considering that the output light of one of the output ports of the last column of splitting units is usually lost, it can be calculated in the splitting weight matrix Fill in a line in the middle and get: Finally, by making the sum of the elements in each column of the matrix the same and the same as the maximum value of the sum of the elements in the previous column before the complement row, the optimal splitting weight setting can be obtained to ensure that the relative size of the elements in the splitting weight matrix remains unchanged and the energy loss of the entire optical computing architecture is minimized.

[0068] Accordingly, in some embodiments, the light splitting weight of each light splitting unit is determined by the following formula:

[0069]

[0070] c ij =b i a ij (2);

[0071]

[0072] In formulas (1), (2) and (3), K ij represents the splitting ratio design parameter of the splitting unit in the i-th column and the j-th row, a ij represents the spectral coefficient of the spectral unit in the i-th column and the j-th row, c ij represents the splitting weight of the splitting unit in the i-th column and the j-th row, b i represents the compensation coefficient of the i-th column of the splitter unit, m represents the total number of columns of the splitter unit, and “=const=” represents taking a constant value.

[0073] Among them, a ij represents the spectral coefficient, i.e. the initial spectral weight, c ij The matrix calculation method provided in the embodiment of the present application indicates the splitting weight after control, that is, the splitting weight that each splitting unit needs to achieve in the end. Formula (3) achieves that in the matrix after row filling, the sum of the elements in each column is the same and the maximum value of the sum of the elements in the column before row filling is the same.

[0074] In the embodiment of the present application, the compensation coefficient is used to accurately control the splitting weight of each splitting unit, so as to improve the splitting ratio of the splitting unit close to the input waveguide and realize accurate control of the splitting ratio. In addition, through the row filling operation, the sum of the elements of each column in the matrix after the row filling is the same and the maximum value of the sum of the elements of the column before the row filling is the same, which can ensure that the relative size of the elements in the matrix of the splitting weight remains unchanged and the energy loss of the entire optical computing architecture is minimized.

[0075] It is understandable that, since the light splitting weight of each light splitting unit is adjusted by the compensation coefficient, accordingly, the influence of the compensation coefficient can be removed by using numerical value, circuit or program processing at the photodetector end.

[0076] In semiconductor manufacturing, the chip size is limited, usually about 20mm*20mm. It may not be possible to construct a large-scale matrix computing array only through a cascade structure. The applicant of this application further proposes that a bus waveguide can be used to splice and interconnect the chips, making it easier to construct a large-scale matrix computing array. Accordingly, in some embodiments, the optical computing architecture includes a plurality of chips arranged in multiple rows and columns, and the spectroscopic units are arranged in multiple rows and columns on each of the chips. The chips are connected by a bus waveguide. The matrix computing scale of the optical computing architecture is hm×kn, where h represents the total number of columns of the chip, k represents the total number of rows of the chip, m represents the total number of columns of the spectroscopic units in the chip, and n represents the total number of rows of the spectroscopic units in the chip. h, k, m, and n are all positive integers.

[0077] Among them, in the embodiments of the present application, there is no specific limitation on how many chips are spliced ​​and interconnected, which depends on the actual computing needs.

[0078] In the embodiments of the present application, there is no specific limitation on how to use bus waveguides to splice and interconnect chips. For example, Figure 4a As shown, a bus waveguide may be arranged on one side of the output port of each column of the optical splitting unit in each chip from the vertical direction, that is, from the column direction of the optical splitting unit, to splice and interconnect the chips.

[0079] like Figure 4b As shown, it is a schematic diagram of another implementation of the optical computing architecture provided in an embodiment of the present application, which includes a total of 4 chips arranged in 2 rows and 2 columns, and a single chip includes a total of 16 splitting units arranged in 4 rows and 4 columns. The 4 chips are spliced ​​and interconnected through the bus waveguide in the column direction of the splitting unit, so as to realize the expansion of the matrix computing scale from 4×4 to 8×8.

[0080] In this splicing mode, one of the output ports of the last column of splitter units is connected to the output waveguide, while the other output port, since there is no input port of the subsequent splitter unit connected to it, measures can be taken to make the output light be absorbed or exported to space loss, thereby preventing the output light from affecting the matrix calculation results due to reflection / refraction, etc.

[0081] It can be seen that the matrix calculation scale finally achieved by this method of splicing and interconnecting chips with the help of bus waveguides is related to the number of rows and columns of the chip and the number of rows and columns of the optical splitting units in a single chip. The matrix calculation method based on the optical computing architecture provided in the embodiment of the present application is expected to achieve a large matrix calculation scale of hundreds or even thousands of ports by connecting the chips through bus waveguides.

[0082] It is understandable that, in addition to the square array arrangement of multiple rows and columns, the spectroscopic units in the chip may also be arranged in a non-square array of m×n, which can also achieve the expansion of the matrix calculation scale.

[0083] There are many implementations for how to provide the optical signal to be calculated to each input waveguide respectively, for example, Figure 5a As shown, a plurality of independent light sources, namely, light source 1, light source 2, light source 3 and light source 4, corresponding to a plurality of input waveguides are used, and each independent light source is provided to its corresponding input waveguide, so that there is no coherence between the outputs of each column of the light splitting unit. Figure 5b As shown, a multi-wavelength light source is first realized by an optical frequency comb, and then a demultiplexer DeMUX is used to perform light splitting, and finally the split results are provided to each input waveguide respectively.

[0084] Accordingly, in some embodiments, Figure 6 As shown, the providing of the optical signal to be calculated to each of the input waveguides (i.e., involved in step S120) may include:

[0085] In step S121, the optical frequency comb is input to the demultiplexer DeMUX, so that the demultiplexer DeMUX performs optical splitting processing on the optical frequency comb to obtain optical signals to be calculated, and provides the optical signals to be calculated to each of the input waveguides respectively.

[0086] As a second aspect of the embodiments of the present application, an electronic device is provided, wherein: Figure 7 As shown, the electronic device includes:

[0087] One or more processors 101;

[0088] 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 the matrix calculation method based on the optical computing architecture provided in the first aspect of the embodiment of the present application.

[0089] 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 .

[0090] 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.

[0091] 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.

[0092] As a third aspect of the embodiments of the present application, a matrix computing system is provided, wherein the matrix computing system comprises the electronic device and the optical computing architecture provided according to the second aspect of the embodiments of the present application, wherein the optical computing architecture comprises an input waveguide, a splitting unit, an output waveguide and a photodetector, wherein each of the splitting units is arranged in multiple rows and columns, each of the splitting units comprises an input port and two output ports, the first splitting unit in each row of the splitting units is connected to the input waveguide via the input port, two adjacent splitting units in each row of the splitting units are connected to the input port of the splitting unit cascaded in the rear via one of the output ports of the splitting unit cascaded in the front, each of the splitting units is connected to the output waveguide via another output port, and the end of the output waveguide to which the splitting units in each column are connected is connected to the same photodetector;

[0093] The spectrometer unit is used to input the light passing through itself into the output waveguide connected to itself according to the received spectrometer weight part, and each photodetector is used to detect the output light of the corresponding column of the spectrometer unit to obtain a detection result, and each detection result is used to determine the matrix operation result of the optical signal to be calculated input into the optical computing architecture.

[0094] The composition of the optical computing architecture and how to implement matrix calculations have been described in detail above and will not be repeated here.

[0095] There are many implementations for how the output light of each light splitting unit is transmitted to the photodetector through the output waveguide, for example: Figure 2 As shown, in each column of splitting units, the output waveguides connected to the splitting units are different in pairs, and the output waveguides are connected to the photodetector in parallel, which is equivalent to the output light of each splitting unit being input to the photodetector in parallel and detected uniformly by the photodetector in the photodetector.

[0096] Correspondingly, in some embodiments, the output waveguides connected to any two light-splitting units in each column of the light-splitting units are different, and the ends of the output waveguides connected to the light-splitting units in each column are connected in parallel to the photodetectors.

[0097] In addition, a mode multiplexing beam combining method can also be used to combine the output light of each optical splitting unit and then input it into the photodetector. Accordingly, in some embodiments, the output waveguide is a bus waveguide, and the bus waveguide includes a plurality of incoherent beam combiners that are different in pairs, and the incoherent beam combiners connected to any two optical splitting units in each column of the optical splitting units are different.

[0098] like Figure 8 As shown, the optical splitter unit is connected to an incoherent combiner in the bus waveguide through one of its output ports, and the incoherent combiners connected to any two optical splitters in each column of the optical splitter units are different. For example, the output light between the optical splitter unit in the first column and the first row is combined with the output light of the optical splitter unit in the first column and the second row through the first type of incoherent combiner, the first combining result is combined with the output light of the optical splitter unit in the first column and the third row through the second type of incoherent combiner, and the second combining result is combined with the output light of the optical splitter unit in the first column and the fourth row through the third type of incoherent combiner, and the final combining result is obtained and input into the photodetector.

[0099] The incoherent combiners connected to each column of optical splitters are different from each other. Therefore, after the output light of each optical splitter is coupled into the bus waveguide, the modes exhibited in the bus waveguide are different from each other, so there is no interference between them, and the beam combining does not produce additional losses. For example, in a column of optical splitters, the output light of the optical splitter in the first row remains unchanged after being coupled into the bus waveguide in TE0 mode, the output light of the optical splitter in the second row is coupled into the bus waveguide in TE0 mode and undergoes mode conversion to TE1 mode, the output light of the optical splitter in the third row is coupled into the bus waveguide in TE0 mode and undergoes mode conversion to TE2 mode... and so on. Light from different output ports can be coupled into the same bus waveguide, and will not interfere with each other in the same bus waveguide, and beam combining will not produce additional losses. It should be noted that the above is only an exemplary description. In fact, the mode exhibited by the output light of the optical splitter in the bus waveguide is not limited to the TE mode, but can also be the TM mode.

[0100] In some embodiments, a beam splitter, a phase shifter and a coupler are cascaded in sequence between the one input port and the two output ports of the light splitting unit.

[0101] As a fourth aspect of the embodiment of the present application, Fig. 9As shown, a computer-readable medium is provided, on which a computer program is stored, wherein when the computer program is executed by a processor, the matrix calculation method based on the optical computing architecture provided in the first aspect of the embodiment of the present application is implemented.

[0102] 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.

[0103] 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. A matrix calculation method based on an optical computing architecture, characterized in that: The optical computing architecture includes an input waveguide, a light splitting unit, an output waveguide and a photodetector, wherein the light splitting units are arranged in multiple rows and columns, and each light splitting unit includes an input port and two output ports. The first light splitting unit in each row of the light splitting units is connected to the input waveguide through the input port, and two adjacent light splitting units in each row of the light splitting units are connected to the input port of the light splitting unit cascaded behind through one of the output ports of the light splitting unit cascaded in front, and each light splitting unit is connected to the output waveguide through another output port, and the end of the output waveguide connected to the light splitting units in each column is connected to the same photodetector; the method includes: Determining the splitting weight of each of the splitting units according to the splitting ratio design parameters of each of the splitting units; Input each of the splitting weights to the corresponding splitting unit, and provide the optical signal to be calculated to each of the input waveguides; wherein the splitting unit inputs the light passing through itself to the output waveguide connected to itself according to the splitting weight part; Obtaining detection results of each of the photoelectric detectors; The matrix operation result of the optical signal to be calculated is determined according to the detection results of each of the photoelectric detectors.

2. The method according to claim 1, characterized in that The splitting weight of each splitting unit is determined by the following formula: c ij =b i a ij (2); In formulas (1), (2) and (3), K ij represents the splitting ratio design parameter of the splitting unit in the i-th column and the j-th row, a ij represents the spectral coefficient of the spectral unit in the i-th column and the j-th row, c ij represents the splitting weight of the splitting unit in the i-th column and the j-th row, b i represents the compensation coefficient of the i-th column of the splitter unit, m represents the total number of columns of the splitter unit, and "=const=" represents taking a constant value.

3. The method according to claim 1, characterized in that The optical computing architecture includes multiple chips arranged in multiple rows and columns, the spectroscopic units are arranged in multiple rows and columns on each of the chips, and the chips are connected by a bus waveguide. The matrix calculation scale of the optical computing architecture is hm×kn, wherein h represents the total number of columns of the chip, k represents the total number of rows of the chip, m represents the total number of columns of the spectroscopic units in the chip, and n represents the total number of rows of the spectroscopic units in the chip, and h, k, m, and n are all positive integers.

4. The method according to any one of claims 1 to 3, characterized in that Providing the optical signal to be calculated to each of the input waveguides respectively comprises: The optical frequency comb is input to the demultiplexer DeMUX, so that the demultiplexer DeMUX performs optical splitting processing on the optical frequency comb to obtain the optical signal to be calculated, and provides the optical signal to be calculated to each of the input waveguides respectively.

5. 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 the matrix calculation method based on the optical computing architecture according to any one of claims 1-4.

6. A matrix computing system, characterized in that: The matrix computing system comprises the electronic device and the optical computing architecture according to claim 5, wherein the optical computing architecture comprises an input waveguide, a splitting unit, an output waveguide and a photodetector, wherein each of the splitting units is arranged in multiple rows and columns, and each of the splitting units comprises an input port and two output ports, wherein the first splitting unit in each row of the splitting units is connected to the input waveguide via the input port, and the two adjacent splitting units in each row of the splitting units are connected to the input port of the splitting unit cascaded in the rear via one of the output ports of the splitting unit cascaded in the front, and each of the splitting units is connected to the output waveguide via another output port, and the end of the output waveguide connected to the splitting units in each column is connected to the same photodetector; The spectrometer unit is used to input the light passing through itself into the output waveguide connected to itself according to the received spectrometer weight part, and each photodetector is used to detect the output light of the corresponding column of the spectrometer unit to obtain a detection result, and each detection result is used to determine the matrix operation result of the optical signal to be calculated input into the optical computing architecture.

7. The matrix computing system according to claim 6, characterized in that: The output waveguides connected to any two light-splitting units in each column of the light-splitting units are different, and the ends of the output waveguides connected to the light-splitting units in each column are connected in parallel to the photodetector.

8. The matrix computing system according to claim 6, characterized in that: The output waveguide is a bus waveguide, and the bus waveguide includes a plurality of incoherent beam combiners that are different in pairs, and the incoherent beam combiners connected to any two optical splitting units in each column of the optical splitting units are different.

9. The matrix computing system according to any one of claims 6 to 8, characterized in that: A beam splitter, a phase shifter and a coupler are cascaded in sequence between the one input port and the two output ports of the light splitting unit.

10. A computer readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the matrix calculation method based on the optical computing architecture according to any one of claims 1 to 4 is implemented.

Images