Positioning and assembling device and method for stacking and cascading spatial diffraction optical neural network arithmetic units

Abstract

The invention provides a positioning and assembling device and method for spatial diffraction optical neural network arithmetic unit stacking and cascading, and relates to the technical field of spatial diffraction optical neural network arithmetic unit stacking and cascading positioning and assembling. Comprising a laser light source; a laser beam expanding unit; a first beam splitter prism; the spatial light modulator is used for sending the test phase pattern to the spatial diffraction optical neural network operation unit through the measurement light beam; the first photoelectric detector is used for receiving the measuring light beam operated by the spatial diffraction optical neural network operation unit and generating light intensity data; the upper computer is used for collecting the light intensity data generated by the first photoelectric detector and determining the spatial position where the spatial diffraction optical neural network operation unit is placed according to the light intensity data; and the motion console is used for receiving a control instruction of the upper computer and adjusting the positions of the spatial diffraction optical neural network arithmetic unit and the first photoelectric detector according to the control instruction so as to change the measurement light beam acquired by the first photoelectric detector.

Description

Technical Field

[0001] This application relates to the field of positioning and assembly technology for stacked and cascaded spatial diffractive optical neural network computing units, and more specifically, to a positioning and assembly device and method for stacked and cascaded spatial diffractive optical neural network computing units. Background Technology

[0002] Diffractive optical neural networks (DONs) utilize the coherence and superposition properties of light to simulate neuronal interconnections through cascaded, layered computational units with optical neurons configured in different modes. Once trained, a Diffractive optical neural network requires only one forward propagation of light to obtain the inference result.

[0003] Spatial diffractive optical neural network (SOLN) computing units are typically fabricated on transparent substrates using techniques such as photolithography or two-photon 3D printing. These units consist of metamaterials composed of subwavelength artificial microstructures, with each microstructure acting as an optical neuron. Each layer of computing units contains a large number of optical neurons with dimensions ranging from micrometers to hundreds of nanometers. The spatial alignment accuracy of these optical neurons significantly impacts the performance of the SOLN processor. Due to the significantly higher density of optical neurons compared to other diffractive optical components, achieving sub-micrometer high-precision alignment of optical neurons in spatial cascading operations, while minimizing axial distance error, is challenging using existing optical microscopy techniques and optical alignment marks in the x and y directions. Although structural information of optical neurons can be obtained through methods such as electron microscopy, the small field of view prevents its use in cascading operations.

[0004] On the one hand, existing stacked positioning and measurement systems based on planar markers can achieve relatively accurate measurements in the XY direction. However, for spatial diffractive optical neural networks, the phase modulation capability is significantly reduced as mechanical errors accumulate in the XYZ direction during the cascading assembly of multi-layer processing units. Moreover, the stacked cascading technology based on planar markers cannot obtain the intuitive phase modulation effect of the cascaded processor, making it quite difficult to determine whether the overall cascading effect is optimal.

[0005] On the other hand, it is difficult to directly measure the phase modulation capability of targets with a large number of subwavelength fine structures within the optical frequency band. Furthermore, in existing phase information detection technologies, such as the four-step phase shift method and coaxial holography, the phase information of single-layer thin targets is detected. However, due to the inaccuracy of the phase shift increment, there is a large error in the calculation of the phase difference. When processing samples with strong scattering or complex metamaterial structures, the phase unfolding process may introduce additional errors, which may result in the inability to accurately obtain the phase modulation results after the cascade of multi-layer phase modulation units. Summary of the Invention

[0006] To address at least one of the aforementioned problems, this application proposes a positioning and assembly device and method for the stacking and cascading of computational units in a spatial diffractive optical neural network.

[0007] According to a first aspect of this application, at least one embodiment of this application provides a positioning and assembly device for cascading spatial diffraction optical neural network computing units, comprising: a laser source for providing a measurement beam; a laser beam expander for expanding the measurement beam emitted from the laser source to a desired size; a first beam splitter for reflecting the expanded measurement beam; a spatial light modulator for loading a test phase pattern and transmitting the test phase pattern to the spatial diffraction optical neural network computing unit via the measurement beam; a first photodetector for receiving the measurement beam processed by the spatial diffraction optical neural network computing unit and generating light intensity data; a host computer for acquiring the light intensity data generated by the first photodetector and determining the spatial position of the spatial diffraction optical neural network computing unit based on the light intensity data; and a motion control console for receiving control commands from the host computer and adjusting the positions of the spatial diffraction optical neural network computing unit and the first photodetector according to the control commands to change the measurement beam acquired by the first photodetector.

[0008] For example, in some embodiments of this application, a vacuum suction device is also included for fixing the spatial diffraction optical neural network computing unit.

[0009] For example, in some embodiments of this application, it further includes: a second beam splitter for receiving the measurement beam after processing by the spatial diffraction optical neural network operation unit, and splitting it into a first measurement beam and a second measurement beam; a second photodetector for receiving the second measurement beam after being split by the second beam splitter, and generating light intensity data; wherein the host computer is further used to collect the light intensity data generated by the first photodetector and the second photodetector, and determine the spatial position of the spatial diffraction optical neural network operation unit based on the collected light intensity data.

[0010] For example, in some embodiments of this application, the motion control console is also used to adjust the positions of the spatial diffraction optical neural network operation unit and the second photodetector according to the control command, so as to change the measurement beam acquired by the second photodetector.

[0011] For example, in some embodiments of this application, a microscope and a tube lens are further included, disposed between the spatial diffraction optical neural network operation unit and the first photodetector and the second photodetector, to improve the accuracy of the positioning and assembly of the spatial diffraction optical neural network operation unit.

[0012] According to a second aspect of this application, at least one embodiment of this application provides a positioning and assembly method for the stacked and cascaded computational units of a spatial diffractive optical neural network, executed by a host computer of the positioning and assembly device as described in any one of the first aspects, the positioning and assembly method comprising:

[0013] Step A: With the Nth layer spatial diffraction optical neural network operation unit in place, output a first control command to the motion control console to control the first photodetector to move to a first set position, a second set position, and a third set position, and obtain the light intensity data of the first photodetector at the first set position, the second set position, and the third set position, respectively, to obtain the difference value between the actual complex optical field and the theoretical complex optical field when the test phase pattern reaches a position with a first distance from the first layer spatial diffraction optical neural network operation unit, and determine the spatial coordinate position of the Nth layer spatial diffraction optical neural network operation unit, where N is an integer greater than or equal to 1;

[0014] Step B: Output a second control command to the motion control console to control the movement of the Nth layer spatial diffraction optical neural network operation unit, and repeat Step A until the movement of the Nth layer spatial diffraction optical neural network operation unit is completed according to the set adjustment range;

[0015] Step C: After the Nth layer spatial diffraction optical neural network operation unit has moved according to the set adjustment range, sort all the difference values ​​between the actual complex optical field and the theoretical complex optical field at the first distance of the obtained test phase pattern to the first layer spatial diffraction optical neural network operation unit to generate a difference value sequence.

[0016] Step D: Select the minimum difference value in the difference value sequence, determine the spatial coordinate position of the Nth layer spatial diffraction optical neural network operation unit corresponding to the minimum difference value, and fix the Nth layer spatial diffraction optical neural network operation unit at the determined spatial coordinate position.

[0017] For example, in some embodiments of this application, the step of outputting a second control command to the motion control console to control the movement of the Nth layer spatial diffraction optical neural network operation unit, and repeating step A until the movement of the Nth layer spatial diffraction optical neural network operation unit is completed according to the set adjustment range, includes:

[0018] When the Nth layer spatial diffraction optical neural network operation unit is the first layer spatial diffraction optical neural network operation unit, the Nth layer spatial diffraction optical neural network operation unit is controlled to move in a single-axis direction in the three-axis coordinate system within the first adjustment range with a first adjustment step size, and step A is repeated until the first layer spatial diffraction optical neural network operation unit has completed moving in a single-axis direction within the first adjustment range with a first adjustment step size.

[0019] If the Nth layer spatial diffraction optical neural network operation unit is not the first layer spatial diffraction optical neural network operation unit, the Nth layer spatial diffraction optical neural network operation unit is controlled to move in one of the three-axis coordinates within a set adjustment range with a set adjustment step size, and step A is repeated until the Nth layer spatial diffraction optical neural network operation unit has moved in the three-axis directions within the set adjustment range with a set adjustment step size.

[0020] According to a third aspect of this application, at least one embodiment of this application provides a positioning and assembly method for the stacked and cascaded computational units of a spatial diffractive optical neural network, executed by a host computer of the positioning and assembly device as described in any one of the first aspects, the positioning and assembly method comprising:

[0021] Step A: With the Nth layer spatial diffraction optical neural network operation unit in place, output a third control command to the motion control console to control the second photodetector to move to the first set position and the second set position, and obtain the light intensity data of the first photodetector at the first set position and the second set position respectively, so as to obtain the difference value between the actual complex optical field and the theoretical complex optical field when the test phase pattern reaches the first distance from the first layer spatial diffraction optical neural network operation unit, and determine the spatial coordinate position of the Nth layer spatial diffraction optical neural network operation unit, where N is an integer greater than or equal to 1;

[0022] Step B: Output a second control command to the motion control console to control the movement of the Nth layer spatial diffraction optical neural network operation unit, and repeat Step A until the movement of the Nth layer spatial diffraction optical neural network operation unit is completed according to the set adjustment range;

[0023] Step C: After the Nth layer spatial diffraction optical neural network operation unit has moved according to the set adjustment range, sort all the difference values ​​between the actual complex optical field and the theoretical complex optical field at the first distance of the obtained test phase pattern to the first layer spatial diffraction optical neural network operation unit to generate a difference value sequence.

[0024] Step D: Select the minimum difference value in the difference value sequence, determine the spatial coordinate position of the Nth layer spatial diffraction optical neural network operation unit corresponding to the minimum difference value, and fix the Nth layer spatial diffraction optical neural network operation unit at the determined spatial coordinate position.

[0025] For example, in some embodiments of this application, the step of outputting a second control command to the motion control console to control the movement of the Nth layer spatial diffraction optical neural network operation unit, and repeating step A until the movement of the Nth layer spatial diffraction optical neural network operation unit is completed according to the set adjustment range, includes:

[0026] When the Nth layer spatial diffraction optical neural network operation unit is the first layer spatial diffraction optical neural network operation unit, the Nth layer spatial diffraction optical neural network operation unit is controlled to move in a single-axis direction in the three-axis coordinate system within the first adjustment range with a first adjustment step size, and step A is repeated until the first layer spatial diffraction optical neural network operation unit has completed moving in a single-axis direction within the first adjustment range with a first adjustment step size.

[0027] If the Nth layer spatial diffraction optical neural network operation unit is not the first layer spatial diffraction optical neural network operation unit, the Nth layer spatial diffraction optical neural network operation unit is controlled to move in one of the three-axis coordinates within a set adjustment range with a set adjustment step size, and step A is repeated until the Nth layer spatial diffraction optical neural network operation unit has moved in the three-axis directions within the set adjustment range with a set adjustment step size.

[0028] For example, in some embodiments of this application, the distance between the first photodetector and the first layer of spatial diffraction optical neural network operation unit is equal to the distance between the second photodetector and the first layer of spatial diffraction optical neural network operation unit when the second photodetector is in the first set position.

[0029] Through the above example embodiments, this application provides a positioning and assembly device and method for the stacking and cascading of spatial diffractive optical neural network (SOL) computing units, addressing the technical problems faced in the positioning and assembly of multilayer spatial diffractive optical neural network (SOL) stacks and cascades. This application calculates the phase information of the light field after the cascaded computing units by measuring the light intensity distribution at a distance after the cascaded SOL processing units. The calculated phase information is compared with the ideal phase information of the design to determine whether the cascaded three-axis position (XYZ axes) has reached the optimal position. The computing units to be cascaded are then moved to the ideal position for assembly, ensuring that the position of each SOL computing unit during the cascading process is the current optimal position.

[0030] This application utilizes standardized processes to fabricate optical components and mechanical control equipment, enabling a direct demonstration of the computational performance of cascaded multilayer spatial diffraction optical neural network (SDN) processing units. This overcomes the shortcomings of traditional cascading techniques, which rely solely on judging the overlap of planar markers under an optical microscope, resulting in poor axial positioning accuracy and an inability to showcase cascaded performance. While ensuring accurate positioning during cascading, it minimizes the performance degradation of the SDN caused by accumulated mechanical errors, thus minimizing the negative impact of cascading operations on the SDN's performance. Particularly advantageous for aligning two or more layers of SDN processing units, this application offers flexibility and intuitiveness, allowing for high-precision integration of multiple layers using XYZ coordinates to ensure optimal light field modulation performance.

[0031] It should be understood that the above general description and the following detailed description are merely exemplary and do not limit this application. Attached Figure Description

[0032] The above and other objects, features, and advantages of this application will become more apparent from the detailed description of exemplary embodiments with reference to the accompanying drawings. The drawings described below are merely some embodiments of this application and are not intended to limit the scope of this application.

[0033] Figure 1 A schematic diagram of a positioning assembly apparatus for stacking and cascading spatial diffractive optical neural network computing units is shown in an exemplary embodiment.

[0034] Figure 2 Another embodiment of an exemplary positioning and assembly device for the stacked and cascaded computational units of a spatial diffractive optical neural network is shown.

[0035] Figure 3 Another embodiment of an exemplary positioning and assembly device for the stacked and cascaded computational units of a spatial diffractive optical neural network is shown.

[0036] Figure 4 A flowchart illustrating an exemplary embodiment of a positioning and assembly method for stacking and cascading computational units of a spatial diffractive optical neural network;

[0037] Figure 5 Another embodiment of a flowchart illustrating an exemplary positioning and assembly method for the stacking and cascading of computational units in a spatial diffractive optical neural network is shown.

[0038] Figure 6 This diagram illustrates the structure of an electronic device provided in this application. Detailed Implementation

[0039] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, they are provided so that this application will be thorough and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted.

[0040] The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced without one or more of these specific details, or other methods, components, materials, devices, etc. In these cases, well-known structures, methods, devices, implementations, materials, or operations will not be shown or described in detail.

[0041] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0042] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0043] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of exemplary embodiments, and the modules or processes in the drawings are not necessarily essential for implementing this application, and therefore cannot be used to limit the scope of protection of this application.

[0044] Figure 1 A schematic diagram of a positioning assembly apparatus for cascading spatial diffractive optical neural network computing units is shown, according to an exemplary embodiment.

[0045] like Figure 1 As shown, the positioning and assembly device for the stacked and cascaded computational units of the spatial diffraction optical neural network includes: a laser source 101, a laser beam expander 102, a spatial light modulator 103, a first beam splitter 104, a first photodetector 107, a motion control console 108, and a host computer 109.

[0046] The system includes a laser source 101 for providing a measurement beam for alignment and measurement. A laser beam expander 102 expands the measurement beam emitted from the laser source 101 to the desired size. A first beam splitter 104 reflects the expanded measurement beam. A spatial light modulator 103 loads a test phase pattern and transmits it to a spatial diffraction optical neural network processing unit 105 via the measurement beam. A first photodetector 107 receives the measurement beam processed by the spatial diffraction optical neural network processing unit 105 and generates light intensity data. A host computer 109 collects the light intensity data generated by the first photodetector 107 and determines the spatial position of the spatial diffraction optical neural network processing unit based on the light intensity data. A motion control console 108 receives control commands from the host computer 109 and adjusts the positions of the spatial diffraction optical neural network processing unit 105 and the first photodetector 107 according to the control commands to change the measurement beam acquired by the first photodetector 107.

[0047] According to some embodiments, the positioning and assembly device for the stacking and cascading of spatial diffractive optical neural network computing units also includes a vacuum adsorber 106.

[0048] The vacuum adsorber 106 is used to fix the spatial diffraction optical neural network computing unit 105.

[0049] According to the example embodiment, a single detector can be used to realize the stacked and cascaded positioning assembly of a space diffractive optical neural network, and finally cascaded with a photodetector array to form a complete space diffractive optical neural network processor:

[0050] 1. Move the photodetector 107 to a distance z0 from the exit surface of the spatial diffraction optical neural network operation unit 105 and collect the light intensity distribution. Then, adjust the photodetector 107 to z0-Δd0 and collect the light intensity distribution. Finally, move the photodetector 107 to z0+Δd0 and collect the light intensity distribution.

[0051] 2. Calculate the actual light field information at z0 of the output surface of the spatial diffraction optical neural network operation unit 105 using the light intensity distribution information at three locations. Compare the actual light field information with the light field information theoretically calculated by the host computer 109 and calculate the difference data.

[0052] 3. Move the spatial diffraction optical neural network operation unit 105 to the optical axis direction (taking the Z axis as an example in this application, but this application is not limited to this), and finally determine the spatial coordinates with the smallest difference of the spatial diffraction optical neural network operation unit 105.

[0053] 4. At this point, the relative position of the photodetector 107 and the optical neural network operation unit 105 has been fixed and recorded as (x0, y0, z0) as the reference coordinates for cascading the next layer of spatial diffraction optical neural network operation unit 105'.

[0054] 5. For each additional spatial diffraction optical neural network operation unit 105', it is necessary to collect light intensity data at the photodetector 107 at positions z0-Δd0, z0, and z0+Δd0 to calculate the actual light field data. Simultaneously, the theoretical light field data at z0 after adding the next spatial diffraction optical neural network operation unit 105' is also calculated. The actual light field data calculated after each adjustment of the optical neural network operation unit 105' is then compared with the theoretical light field data to calculate the difference.

[0055] 6. Next, sequentially adjust the position of the spatial diffraction optical neural network operation unit along the Z-axis, X-axis, and Y-axis. Repeat step 5. After adjusting the position of the corresponding axes, select the position with the smallest difference from the theoretical light field. Determine the position with the smallest difference data on the Z-axis, X-axis, and Y-axis for the spatial diffraction optical neural network operation unit, and fix the optical neural network operation unit 105' at this position.

[0056] 7. Subsequently, add the next layer of optical neural network operation units in sequence, and repeat steps 5 and 6. After all operation units have been added, move the photodetector 107 to z0 to complete the cascading with the first layer of spatial diffraction optical neural network operation unit 105.

[0057] Figure 2 Another embodiment of an exemplary positioning assembly device for the cascaded stacking of spatial diffractive optical neural network computing units is shown.

[0058] Figure 2 and Figure 1 The positioning and assembly devices are basically similar in structure, with the difference being: Figure 2 The positioning and assembly device shown also includes: a second beam splitter 104' and a second photodetector 107'.

[0059] The second beam splitter 104' receives the measurement beam processed by the spatial diffraction optical neural network processing unit 105 and splits it into a first measurement beam and a second measurement beam, which are then incident on the first photodetector 107 and the second photodetector 107', respectively. The first photodetector 107 and the second photodetector 107' receive the first and second measurement beams after they have been split by the second beam splitter and generate light intensity data, respectively.

[0060] According to the example embodiment, the host computer 109 is also used to collect light intensity data generated by the first photodetector and the second photodetector, and to determine the spatial position of the spatial diffraction optical neural network operation unit based on the collected light intensity data.

[0061] The motion control console 108' is also used to receive control commands from the host computer 109 and adjust the position of the spatial diffraction optical neural network operation unit 105 and the second photodetector 107' according to the control commands, so as to change the measurement beam acquired by the second photodetector 107'.

[0062] According to the example embodiment, dual detectors can be used to achieve the stacked and cascaded positioning assembly of a spatial diffractive optical neural network, and finally cascaded with a photodetector array to form a complete spatial diffractive optical neural network processor:

[0063] 1. Move photodetectors 107 and 107' to a distance z0 from the exit surface of the spatial diffraction optical neural network operation unit 105 and collect the light intensity distribution. Then adjust photodetector 107' to z0-Δd0 or z0+Δd0 and collect the light intensity distribution.

[0064] 2. Based on the positional difference between photodetector 107 and photodetector 107', the actual light field information at the exit surface z0 of the spatial diffraction optical neural network operation unit 105 is calculated by the algorithm. The actual light field information is compared with the light field information theoretically calculated by the host computer 109, and the difference data is calculated.

[0065] 3. Move the spatial diffraction optical neural network operation unit 105 to the optical axis direction (taking the Z axis as an example in this application, but this application is not limited to this), and finally determine the spatial coordinates with the smallest difference of the spatial diffraction optical neural network operation unit 105.

[0066] 4. At this point, the relative position of the photodetector 107 and the optical neural network operation unit 105 has been fixed and recorded as (x0, y0, z0) as the reference coordinates for cascading the next layer of spatial diffraction optical neural network operation unit 105'.

[0067] 5. For each additional spatial diffraction optical neural network operation unit 105', it is necessary to collect light intensity data at positions z0, z0-Δd0, or z0+Δd0 of the photodetector 107' to calculate the actual light field data. Simultaneously, the theoretical light field data at z0 after adding the next spatial diffraction optical neural network operation unit 105' is also calculated. The actual light field data calculated after each adjustment of the optical neural network operation unit 105' is then compared with the theoretical light field data to calculate the difference.

[0068] 6. Next, adjust the position of the spatial diffraction optical neural network operation unit 105' along the Z-axis, X-axis, and Y-axis in sequence. Repeat step 5. After adjusting the position of the corresponding axis, select the position with the smallest difference from the theoretical light field. Determine the position with the smallest difference data of the spatial diffraction optical neural network operation unit 105' on the Z-axis, X-axis, and Y-axis respectively. Fix the optical neural network operation unit 105' at this position.

[0069] 7. Subsequently, add the next layer of optical neural network operation units in sequence, and repeat steps 5 and 6. After all operation units have been added, move photodetectors 107 and 107' to z0 to complete the cascading with the first layer of spatial diffraction optical neural network operation unit 105.

[0070] Figure 3 Another embodiment of an exemplary positioning assembly device for the cascaded stacking of spatial diffractive optical neural network computing units is shown.

[0071] Figure 3 and Figure 2 The positioning and assembly devices are basically similar in structure, with the difference being: Figure 3 The positioning and assembly device shown also includes: microscope 111 and tube mirrors 110 and 110'.

[0072] The microscope 111 and the tube lenses 110 and 110' are positioned between the spatial diffraction optical neural network operation unit 105 and the first photodetector 107 and the second photodetector 107'. The light field modulated by the spatial diffraction optical neural network operation unit 105 is imaged on the first photodetector 107 and the second photodetector 107' after passing through the microscope objective 111 and the tube lenses 110 and 110'. This is used to improve the accuracy of the positioning and assembly of the spatial diffraction optical neural network operation unit, so as to ensure that the light field modulated by the operation unit can be received by the photodetector with a larger pixel size, and to ensure the accurate correspondence of optical neurons. This magnifies the hundreds of nanometer-level pixels on the operation unit to the order of several micrometers, thereby reducing the alignment error.

[0073] According to the example embodiment, the operation mode of the positioning and assembly device after adding microscope 111 and tube lenses 110, 110' is the same as... Figure 2 Since they are consistent, they will not be elaborated upon here.

[0074] According to some embodiments, microscope 111 and tube mirrors 110, 110' may also be added to... Figure 1 The positioning and assembly device with a single detector shown is used to improve the accuracy of positioning and assembly of the spatial diffraction optical neural network computing unit.

[0075] Figure 4A flowchart illustrating an exemplary embodiment of a positioning and assembly method for stacking and cascading computational units of a spatial diffractive optical neural network is provided.

[0076] This application also provides a positioning and assembly method for the stacking and cascading of computational units in a spatial diffraction optical neural network, such as... Figure 4 As shown, the positioning and assembly method for the stacked and cascaded computational units of a spatial diffractive optical neural network includes steps S401-S404, which are executed by the host computer 109. The positioning and assembly device for the stacked and cascaded computational units of a spatial diffractive optical neural network includes a single photodetector.

[0077] In step S401, when the Nth layer spatial diffraction optical neural network operation unit is placed, a first control command is output to the motion control console to control the first photodetector to move to the first set position, the second set position, and the third set position, and to obtain the light intensity data of the first photodetector at the first set position, the second set position, and the third set position, respectively, so as to obtain the difference value between the actual complex optical field and the theoretical complex optical field when the test phase pattern reaches the first distance from the first layer spatial diffraction optical neural network operation unit, and to determine the spatial coordinate position of the Nth layer spatial diffraction optical neural network operation unit, where N is an integer greater than or equal to 1.

[0078] The host computer 109 calculates the theoretical complex optical field at a distance z0 from the first layer optical neural network processing unit after the test phase pattern has been processed by the placed spatial diffraction optical neural network operation unit.

[0079] According to the example embodiment, Figure 1 For example, the first photodetector is controlled to move to the first set position 107, the second set position 107' and the third set position 107", and the light intensity data of the first photodetector at the first set position 107, the second set position 107' and the third set position 107" are obtained respectively, so as to obtain the actual complex light field of the test phase pattern reaching the first distance from the first layer of spatial diffraction optical neural network operation unit.

[0080] The host computer 109 determines the difference between the actual complex optical field and the theoretical complex optical field, and determines the spatial coordinates of the Nth layer spatial diffraction optical neural network operation unit.

[0081] Specifically, the distance between the first set position 107 and the first layer spatial diffraction optical neural network operation unit 105 is z0, i.e., the first distance. The distance between the second set position 107' and the first layer spatial diffraction optical neural network operation unit 105 is z0-Δd0. The distance between the third set position 107" and the first layer spatial diffraction optical neural network operation unit 105 is z0+Δd0.

[0082] After obtaining the difference value and the spatial coordinates of the Nth layer spatial diffraction optical neural network operation unit, it is necessary to determine whether the Nth layer spatial diffraction optical neural network operation unit has been adjusted within the first range. If the adjustment is completed, proceed to step S403; if the adjustment is not completed, proceed to step S402.

[0083] In step S402, a second control command is output to the motion control console to control the movement of the Nth layer spatial diffraction optical neural network operation unit, and step S401 is repeated.

[0084] According to an example embodiment, the method for controlling the movement of the Nth layer spatial diffraction optical neural network operation unit includes: determining whether the Nth layer spatial diffraction optical neural network operation unit is the first layer spatial diffraction optical neural network operation unit; if yes, proceed to step S4021; if no, proceed to step S4022.

[0085] In step S4021, when the Nth layer spatial diffraction optical neural network operation unit is the first layer spatial diffraction optical neural network operation unit, the first layer spatial diffraction optical neural network operation unit is controlled to move in a single-axis direction in the three-axis coordinate system within the first adjustment range with a first adjustment step size, and step S401 is repeated until the first layer spatial diffraction optical neural network operation unit has moved in a single-axis direction within the first adjustment range with a first adjustment step size.

[0086] For example, taking the first adjustment range as [-Δz, Δz], the first adjustment step size as Δz / 2, and the adjustment direction as the Z-axis direction, the first layer of spatial diffraction optical neural network operation unit needs to be moved to z1, z1+Δz / 2, z1-Δz / 2, z1+Δz, and z1-Δz respectively, and step S401 is repeated to obtain different difference data. The initial spatial coordinate position of the first layer of spatial diffraction optical neural network operation unit is set as (x1, y1, z1).

[0087] According to some embodiments, this application only takes the movement of the first-layer spatial diffraction optical neural network operation unit in the Z-axis direction as an example, without moving the first-layer spatial diffraction optical neural network operation unit in the X-axis or Y-axis direction. However, this application is not limited to this, and it is also possible to move the first-layer spatial diffraction optical neural network operation unit in the X-axis direction without moving it in the Z-axis or Y-axis direction.

[0088] In step S4022, if the Nth layer spatial diffraction optical neural network operation unit is not the first layer spatial diffraction optical neural network operation unit, the Nth layer spatial diffraction optical neural network operation unit is controlled to move in one of the axes of the three-axis coordinate system within the set adjustment range with a set adjustment step size, and step S401 is repeated until the Nth layer spatial diffraction optical neural network operation unit has moved in the three axes respectively within the set adjustment range with a set adjustment step size.

[0089] According to the example embodiment, if the Nth layer spatial diffraction optical neural network operation unit is not the first layer spatial diffraction optical neural network operation unit, the Z-axis, X-axis and Y-axis directions of the Nth layer spatial diffraction optical neural network operation unit need to be adjusted respectively.

[0090] For example, taking an adjustment range of [-Δz, Δz] and an adjustment step of Δz / 2 as an example, the Nth layer spatial diffraction optical neural network operation unit needs to be moved to z2, z2+Δz / 2, z2-Δz / 2, z2+Δz, and z2-Δz respectively, and step S401 is repeated for each, thereby obtaining the corresponding difference data and spatial coordinate position of the Nth layer spatial diffraction optical neural network operation unit in the Z-axis direction. After completing the adjustment in the Z-axis direction, the adjustment in the X-axis direction needs to be completed. Taking an adjustment range of [-Δx, Δx] and an adjustment step of Δx / 2 as an example, the Nth layer spatial diffraction optical neural network operation unit needs to be moved to x2, x2+Δx / 2, x2-Δx / 2, x2+Δx, and x2-Δx respectively, and step S401 is repeated for each, thereby obtaining the corresponding difference data and spatial coordinate position of the Nth layer spatial diffraction optical neural network operation unit in the X-axis direction. After completing the adjustment in the X-axis direction, the adjustment in the Y-axis direction needs to be completed. Taking the adjustment range as [-Δy, Δy] and the adjustment step size as Δy as an example, the Nth layer spatial diffraction optical neural network operation unit needs to be moved to y2, y2+Δy / 2, y2-Δy / 2, y2+Δy, and y2-Δy respectively, and step S401 is repeated for each, thereby obtaining the corresponding difference data and spatial coordinate position of the Nth layer spatial diffraction optical neural network operation unit in the Y-axis direction. The initial spatial coordinate position of the Nth layer spatial diffraction optical neural network operation unit is set as (x2, y2, z2).

[0091] According to some embodiments, this application does not limit the order in which the operation units of the Nth layer spatial diffraction optical neural network are adjusted. The Z-axis can be adjusted first, or the X-axis or Y-axis can be adjusted first. This application only takes adjusting the Z-axis first as an example, but this application is not limited to this.

[0092] In step S403, after the Nth layer spatial diffraction optical neural network operation unit has moved according to the set adjustment range, all the difference values ​​between the actual complex optical field and the theoretical complex optical field at the first distance of the obtained test phase pattern to the first layer spatial diffraction optical neural network operation unit are sorted to generate a difference value sequence.

[0093] According to the example embodiment, after the first-layer spatial diffraction optical neural network operation unit has completed the adjustment of the single-axis direction according to the first adjustment range and obtained the corresponding difference value, all the difference values ​​of the actual complex optical field and the theoretical complex optical field at the first distance of the obtained test phase pattern to the first-layer spatial diffraction optical neural network operation unit are sorted to generate a difference value sequence, and then the process proceeds to step S404.

[0094] According to the example embodiment, after the Nth layer spatial diffraction optical neural network operation unit has completed the adjustment of all three axes according to the set adjustment range and obtained the corresponding difference values, the difference values ​​of the actual complex optical field and the theoretical complex optical field at the first distance of the obtained test phase pattern to the first layer spatial diffraction optical neural network operation unit are sorted to generate a three-axis difference value sequence, namely the difference value sequence corresponding to the X-axis, the difference value sequence corresponding to the Y-axis, and the difference value sequence corresponding to the Z-axis, and then proceed to step S404, where N is an integer greater than or equal to 2.

[0095] In step S404, the minimum difference value is selected from the difference value sequence, and the spatial coordinate position of the Nth layer spatial diffraction optical neural network operation unit corresponding to the minimum difference value is determined. The Nth layer spatial diffraction optical neural network operation unit is fixed at the determined spatial coordinate position.

[0096] According to the example embodiment, the Nth layer spatial diffraction optical neural network operation unit is fixed according to the spatial coordinate position corresponding to the minimum difference value, and it is determined whether it is the last layer spatial diffraction optical neural network operation unit: if yes, the photodetector 107 is moved to the first set position z0 to complete the cascading with the first layer spatial diffraction optical neural network operation unit 105; if no, a new layer of spatial diffraction optical neural network operation unit is placed, and step S401 is repeated.

[0097] Figure 5 Another embodiment of a flowchart illustrating an exemplary positioning and assembly method for stacking and cascading computational units of a spatial diffractive optical neural network is shown.

[0098] like Figure 5As shown, the positioning and assembly method for the stacked and cascaded computational units of a spatial diffractive optical neural network includes steps S501-S504, which are executed by the host computer 109. The positioning and assembly device for the stacked and cascaded computational units of a spatial diffractive optical neural network includes dual photodetectors.

[0099] In step S501, when the Nth layer spatial diffraction optical neural network operation unit is placed, a third control command is output to the motion control console to control the second photodetector to move to the first set position and the second set position, and to obtain the light intensity data of the first photodetector at the first set position and the second set position, respectively, so as to obtain the difference value between the actual complex optical field and the theoretical complex optical field when the test phase pattern reaches the first distance from the first layer spatial diffraction optical neural network operation unit, and to determine the spatial coordinate position of the Nth layer spatial diffraction optical neural network operation unit, where N is an integer greater than or equal to 1.

[0100] The host computer 109 calculates the theoretical complex optical field at a distance z0 from the first layer optical neural network processing unit after the test phase pattern has been processed by the placed spatial diffraction optical neural network operation unit.

[0101] According to the example embodiment, Figure 2 For example, the second photodetector is controlled to move to the first set position 107' and the second set position 107", and the light intensity data of the first photodetector at the first set position 107' and the second set position 107" are obtained respectively, so as to obtain the actual complex light field of the test phase pattern reaching the first layer of spatial diffraction optical neural network operation unit.

[0102] The host computer 109 determines the difference between the actual complex optical field and the theoretical complex optical field, and determines the spatial coordinates of the Nth layer spatial diffraction optical neural network operation unit.

[0103] The distance between the first set position of the second photodetector 107' and the first layer spatial diffraction optical neural network operation unit 105 is z0, and the distance between the second set position 107" and the first layer spatial diffraction optical neural network operation unit 105 is z0-Δd0 or z0+Δd0.

[0104] After obtaining the difference value and the spatial coordinates of the Nth layer spatial diffraction optical neural network operation unit, it is necessary to determine whether the Nth layer spatial diffraction optical neural network operation unit has been adjusted within the first range. If the adjustment is completed, proceed to step S503; if the adjustment is not completed, proceed to step S502.

[0105] According to the example embodiment, the distance between the first photodetector and the first layer of spatial diffraction optical neural network operation unit and the distance between the second photodetector and the first layer of spatial diffraction optical neural network operation unit are equal when the second photodetector is in a first set position.

[0106] In step S502, a second control command is output to the motion control console to control the movement of the Nth layer spatial diffraction optical neural network operation unit, and step A is repeated.

[0107] According to the example embodiment, the method for controlling the movement of the Nth layer spatial diffraction optical neural network operation unit includes: determining whether the Nth layer spatial diffraction optical neural network operation unit is the first layer spatial diffraction optical neural network operation unit; if yes, proceed to step S5021; if no, proceed to step S5022.

[0108] In step S5021, when the Nth layer spatial diffraction optical neural network operation unit is the first layer spatial diffraction optical neural network operation unit, the Nth layer spatial diffraction optical neural network operation unit is controlled to move in a single-axis direction in the three-axis coordinate system within the first adjustment range with a first adjustment step size, and step S501 is repeated until the first layer spatial diffraction optical neural network operation unit has moved in a single-axis direction within the first adjustment range with a first adjustment step size.

[0109] For example, taking the first adjustment range as [-Δz, Δz] and the adjustment direction as the Z-axis direction, and the first adjustment step size as Δz / n, the first layer of spatial diffraction optical neural network operation unit needs to be moved to z1, z1+Δz / n, z1+2Δz / n, ..., z1+(n-1)Δz / n, z1+Δz, z1-Δz / n, z1-2Δz / n, ..., z1-(n-1)Δz / n, z1-Δz, and step S401 is repeated to obtain different difference data. The initial spatial coordinate position of the first layer of spatial diffraction optical neural network operation unit is set as (x1, y1, z1), where n is a positive integer greater than or equal to 2.

[0110] According to some embodiments, this application only takes the movement of the first-layer spatial diffraction optical neural network operation unit in the Z-axis direction as an example, without moving the first-layer spatial diffraction optical neural network operation unit in the X-axis or Y-axis direction. However, this application is not limited to this, and it is also possible to move the first-layer spatial diffraction optical neural network operation unit in the X-axis direction without moving it in the Z-axis or Y-axis direction.

[0111] In step S5022, if the Nth layer spatial diffraction optical neural network operation unit is not the first layer spatial diffraction optical neural network operation unit, the Nth layer spatial diffraction optical neural network operation unit is controlled to move in one of the axes of the three-axis coordinate system within the set adjustment range with a set adjustment step size, and step S501 is repeated until the Nth layer spatial diffraction optical neural network operation unit has moved in the three axes respectively within the set adjustment range with a set adjustment step size.

[0112] According to the example embodiment, if the Nth layer spatial diffraction optical neural network operation unit is not the first layer spatial diffraction optical neural network operation unit, the Z-axis, X-axis and Y-axis directions of the Nth layer spatial diffraction optical neural network operation unit need to be adjusted respectively.

[0113] For example, taking an adjustment range of [-Δz, Δz] and an adjustment step size of Δz / n (where n is 3) as an example, the Nth layer spatial diffraction optical neural network operation unit needs to be moved to z2, z2+Δz / 3, z2-Δz / 3, z2+2Δz / 3, z2-2Δz / 3, z2+Δz, and z2-Δz respectively, and step S401 is repeated for each of these steps to obtain the corresponding difference data and spatial coordinate position of the Nth layer spatial diffraction optical neural network operation unit in the Z-axis direction. After completing the adjustment in the Z-axis direction, the adjustment in the X-axis direction needs to be completed. Taking an adjustment range of [-Δx, Δx] and an adjustment step size of Δx / n (where n is 2) as an example, the Nth layer spatial diffraction optical neural network operation unit needs to be moved to x2, x2+Δx / 2, x2-Δx / 2, x2+Δx, and x2-Δx respectively, and step S401 is repeated for each. This obtains the corresponding difference data and spatial coordinate position of the Nth layer spatial diffraction optical neural network operation unit in the X-axis direction. After completing the adjustment in the X-axis direction, the adjustment in the Y-axis direction needs to be completed. Taking an adjustment range of [-Δy, Δy] and an adjustment step size of Δy / n, where n is 4, as an example, the Nth layer spatial diffraction optical neural network operation unit needs to be moved to y2, y2+Δy / 4, y2-Δy / 4, y2+2Δy / 4, y2-2Δy / 4, y2+3Δy / 4, y2-3Δy / 4, y2+Δy, and y2-Δy respectively, and step S401 is repeated for each of these steps. This obtains the corresponding difference data and spatial coordinate position of the Nth layer spatial diffraction optical neural network operation unit in the Y-axis direction. The initial spatial coordinate position of the Nth layer spatial diffraction optical neural network operation unit is set to (x2, y2, z2).

[0114] According to some embodiments, this application does not limit the order in which the operation units of the Nth layer spatial diffraction optical neural network are adjusted. The Z-axis can be adjusted first, or the X-axis or Y-axis can be adjusted first. This application only takes adjusting the Z-axis first as an example, but this application is not limited to this.

[0115] In step S503, after the Nth layer spatial diffraction optical neural network operation unit has moved according to the set adjustment range, all the difference values ​​between the actual complex optical field and the theoretical complex optical field at the first distance of the obtained test phase pattern to the first layer spatial diffraction optical neural network operation unit are sorted to generate a difference value sequence.

[0116] According to the example embodiment, after the first-layer spatial diffraction optical neural network operation unit has completed the adjustment of the single-axis direction according to the first adjustment range and obtained the corresponding difference value, all the difference values ​​of the actual complex optical field and the theoretical complex optical field at the first distance of the obtained test phase pattern to the first-layer spatial diffraction optical neural network operation unit are sorted to generate a difference value sequence, and then the process proceeds to step S504.

[0117] According to the example embodiment, after the Nth layer spatial diffraction optical neural network operation unit has completed the adjustment of all three axes according to the set adjustment range and obtained the corresponding difference values, the difference values ​​of the actual complex optical field and the theoretical complex optical field at the first distance of the obtained test phase pattern to the first layer spatial diffraction optical neural network operation unit are sorted to generate a three-axis difference value sequence, namely the difference value sequence corresponding to the X-axis, the difference value sequence corresponding to the Y-axis, and the difference value sequence corresponding to the Z-axis, and then proceed to step S504, where N is an integer greater than or equal to 2.

[0118] In step S504, the minimum difference value is selected from the difference value sequence, and the spatial coordinate position of the Nth layer spatial diffraction optical neural network operation unit corresponding to the minimum difference value is determined. The Nth layer spatial diffraction optical neural network operation unit is fixed at the determined spatial coordinate position.

[0119] According to the example embodiment, the Nth layer spatial diffraction optical neural network operation unit is fixed according to the spatial coordinate position corresponding to the minimum difference value, and it is determined whether it is the last layer spatial diffraction optical neural network operation unit: if yes, the photodetector 107' is moved to the first set position z0 to complete the cascading with the first layer spatial diffraction optical neural network operation unit 105; if no, a new layer of spatial diffraction optical neural network operation unit is placed, and step S501 is repeated.

[0120] To address the technical challenges of positioning and assembling multilayer spatial diffractive optical neural networks (SOLNs), this application provides a positioning and assembly device and method for cascading SOLN computational units. This application calculates the phase information of the light field after cascading computational units by measuring the light intensity distribution at a distance after each cascaded SOLN processing unit. The calculated phase information is compared with the ideal phase information of the design to determine whether the cascaded three-axis position (XYZ axes) has reached its optimal value. The computational units to be cascaded are then moved to the ideal position for assembly, ensuring that the position of each SOLN computational unit during the cascading process is the current optimal position.

[0121] This application utilizes standardized processes to fabricate optical components and mechanical control equipment, enabling a direct demonstration of the computational performance of cascaded multilayer spatial diffraction optical neural network (SDN) processing units. This overcomes the shortcomings of traditional cascading techniques, which rely solely on judging the overlap of planar markers under an optical microscope, resulting in poor axial positioning accuracy and an inability to showcase cascaded performance. While ensuring accurate positioning during cascading, it minimizes the performance degradation of the SDN caused by accumulated mechanical errors, thus minimizing the negative impact of cascading operations on the SDN's performance. Particularly advantageous for aligning two or more layers of SDN processing units, this application offers flexibility and intuitiveness, allowing for high-precision integration of multiple layers using XYZ coordinates to ensure optimal light field modulation performance.

[0122] Figure 6 This diagram illustrates the structure of an electronic device provided in this application.

[0123] See Figure 6 , Figure 6 An electronic device is provided, including a processor and a memory. The memory stores computer instructions, which, when executed by the processor, cause the processor to perform the computer instructions to achieve the following: Figure 4 or Figure 5 The method and its detailed scheme are shown.

[0124] It should be understood that the above-described device embodiments are merely illustrative, and the device disclosed in this application can also be implemented in other ways. For example, the division of units / modules in the above embodiments is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units, modules, or components may be combined, or integrated into another system, or some features may be ignored or not executed.

[0125] Furthermore, unless otherwise specified, the functional units / modules in the various embodiments of this application can be integrated into one unit / module, or each unit / module can exist physically separately, or two or more units / modules can be integrated together. The integrated units / modules described above can be implemented in hardware or as software program modules.

[0126] When integrated units / modules are implemented in hardware, the hardware can be digital circuits, analog circuits, etc. The physical implementation of the hardware structure includes, but is not limited to, transistors, memristors, etc. Unless otherwise specified, the processor or chip can be any suitable hardware processor, such as a CPU, GPU, FPGA, DSP, and ASIC, etc. Unless otherwise specified, on-chip cache, off-chip memory, and storage can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc.

[0127] If the integrated unit / module is implemented as a software program module and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments disclosed herein. The aforementioned memory includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0128] This application also provides a non-transitory computer storage medium storing a computer program, which, when executed by multiple processors, causes the processors to perform actions such as... Figure 4 or Figure 5 The method and its detailed scheme are shown.

[0129] It should be clearly understood that this application describes how specific examples are formed and used, but this application is not limited to any details of these examples. Rather, based on the teachings of the disclosure of this application, these principles can be applied to many other embodiments.

[0130] Furthermore, it should be noted that the above figures are merely illustrative representations of the processes included in the method according to exemplary embodiments of this application, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0131] Exemplary embodiments of this application have been specifically shown and described above. It should be understood that this application is not limited to the detailed structures, arrangements, or implementation methods described herein; rather, this application is intended to cover various modifications and equivalent arrangements contained within the spirit and scope of the appended claims.

Claims

1. A positioning and assembly device for stacking and cascading spatial diffraction optical neural network computing units, characterized in that: include: A laser light source for providing a measuring beam; A laser beam expansion unit, used to expand the measuring beam emitted by the laser light source to a required size; A first beam splitter prism, used for reflecting the measurement light beam after the size is expanded; A spatial light modulator, used for loading a test phase pattern and sending the test phase pattern to the spatial diffraction optical neural network operation unit through the measurement light beam; A first photoelectric detector is used to receive the measurement light beam after being calculated by the spatial diffraction optical neural network calculation unit and generate light intensity data; A host computer, used for collecting light intensity data generated by the first photodetector, and determining the spatial position of the spatial diffraction optical neural network operation unit according to the light intensity data; A motion control console is used to receive control instructions from the host computer and adjust the positions of the spatial diffraction optical neural network operation unit and the first photodetector according to the control instructions to change the measurement light beam obtained by the first photodetector.

2. The positioning assembly device according to claim 1, characterized in that: Also includes: A vacuum absorber is used to fix the spatial diffraction optical neural network computing unit.

3. The positioning assembly device according to claim 1 or 2, characterized in that: Also includes: A second beam splitter prism is used to receive the measuring beam after being operated by the spatial diffraction optical neural network operation unit, and split it into a first measuring beam and a second measuring beam; a second photodetector, used for receiving the second measuring light beam after being split by the second beam splitting prism, and generating light intensity data; The host computer is also used to collect light intensity data generated by the first photodetector and the second photodetector, and determine the spatial position of the spatial diffraction optical neural network operation unit based on the collected light intensity data.

4. The positioning assembly device according to claim 3, characterized in that: The motion control console is also used to adjust the positions of the spatial diffraction optical neural network operation unit and the second photodetector according to the control instruction to change the measurement light beam acquired by the second photodetector.

5. The positioning assembly device according to claim 3 or 4, characterized in that: Also includes: A microscope and a tube lens are arranged between the spatial diffraction optical neural network operation unit and the first photodetector and the second photodetector, and are used to improve the accuracy of positioning and assembling the spatial diffraction optical neural network operation unit.

6. A positioning and assembly method for cascading spatial diffraction optical neural network computing units, characterized in that: The positioning and assembly method is performed by a host computer of the positioning and assembly device according to any one of claims 1 to 5, and comprises: Step A: When placing the N-th layer of spatial diffraction optical neural network operation unit, output a first control instruction to the motion control console to control the first photodetector to move to a first set position, a second set position and a third set position, and respectively obtain the light intensity data of the first photodetector at the first set position, the second set position and the third set position, so as to obtain the difference value between the actual complex light field and the theoretical complex light field when the test phase pattern arrives at a first distance from the first layer of spatial diffraction optical neural network operation unit, and determine the spatial coordinate position of the N-th layer of spatial diffraction optical neural network operation unit, where N is an integer greater than or equal to 1; Step B: outputting a second control instruction to the motion control console to control the movement of the N-th layer of spatial diffraction optical neural network operation unit, and repeating step A until the movement of the N-th layer of spatial diffraction optical neural network operation unit is completed according to the set adjustment range; Step C: when the N-th spatial diffraction optical neural network operation unit has completed moving according to the set adjustment range, all difference values ​​between the actual complex light field and the theoretical complex light field obtained when the test phase pattern reaches the first distance of the first spatial diffraction optical neural network operation unit are sorted to generate a difference value sequence; Step D: Select the minimum difference value in the difference value sequence, determine the spatial coordinate position of the N-th layer of spatial diffraction optical neural network operation unit corresponding to the minimum difference value, and fix the N-th layer of spatial diffraction optical neural network operation unit at the determined spatial coordinate position.

7. The positioning and assembling method according to claim 6, characterized in that: Outputting the second control instruction to the motion control console to control the movement of the N-th layer of spatial diffraction optical neural network operation unit, and repeating step A until the N-th layer of spatial diffraction optical neural network operation unit moves within the set adjustment range, including: In the case where the N-th layer of spatial diffraction optical neural network operation unit placed is the first layer of spatial diffraction optical neural network operation unit, the N-th layer of spatial diffraction optical neural network operation unit is controlled to move in a single-axis direction in the three-axis coordinate system within a first adjustment range with a first adjustment step length, and step A is repeated until the first layer of spatial diffraction optical neural network operation unit is completely moved in a single-axis direction within the first adjustment range with a first adjustment step length; When the placed N-th layer of spatial diffraction optical neural network operation unit is not the first-layer spatial diffraction optical neural network operation unit, the N-th layer of spatial diffraction optical neural network operation unit is controlled to move in one of the axis directions in the three-axis coordinate system within the set adjustment range with the set adjustment step, and step A is repeated until the N-th layer of spatial diffraction optical neural network operation unit is moved in the three-axis directions within the set adjustment range with the set adjustment step.

8. A positioning and assembly method for stacking and cascading spatial diffraction optical neural network computing units, characterized in that: The positioning and assembly method is performed by a host computer of the positioning and assembly device according to any one of claims 3 to 5, and comprises: Step A: When placing the N-th layer of spatial diffraction optical neural network operation unit, output a third control instruction to the motion control console to control the second photodetector to move to a first set position and a second set position, and respectively obtain the light intensity data of the first photodetector at the first set position and the second set position, so as to obtain the difference value between the actual complex light field and the theoretical complex light field when the test phase pattern arrives at a first distance from the first layer of spatial diffraction optical neural network operation unit, and determine the spatial coordinate position of the N-th layer of spatial diffraction optical neural network operation unit, where N is an integer greater than or equal to 1; Step B: outputting a second control instruction to the motion control console to control the movement of the N-th layer of spatial diffraction optical neural network operation unit, and repeating step A until the movement of the N-th layer of spatial diffraction optical neural network operation unit is completed according to the set adjustment range; Step C: when the N-th spatial diffraction optical neural network operation unit has completed moving according to the set adjustment range, all difference values ​​between the actual complex light field and the theoretical complex light field obtained when the test phase pattern reaches the first distance of the first spatial diffraction optical neural network operation unit are sorted to generate a difference value sequence; Step D: Select the minimum difference value in the difference value sequence, determine the spatial coordinate position of the N-th layer of spatial diffraction optical neural network operation unit corresponding to the minimum difference value, and fix the N-th layer of spatial diffraction optical neural network operation unit at the determined spatial coordinate position.

9. The positioning and assembling method according to claim 8, characterized in that: Outputting the second control instruction to the motion control console to control the movement of the N-th layer of spatial diffraction optical neural network operation unit, and repeating step A until the N-th layer of spatial diffraction optical neural network operation unit moves within the set adjustment range, including: In the case where the N-th layer of spatial diffraction optical neural network operation unit placed is the first layer of spatial diffraction optical neural network operation unit, the N-th layer of spatial diffraction optical neural network operation unit is controlled to move in a single-axis direction in the three-axis coordinate system within a first adjustment range with a first adjustment step length, and step A is repeated until the first layer of spatial diffraction optical neural network operation unit is completely moved in a single-axis direction within the first adjustment range with a first adjustment step length; When the placed N-th layer of spatial diffraction optical neural network operation unit is not the first-layer spatial diffraction optical neural network operation unit, the N-th layer of spatial diffraction optical neural network operation unit is controlled to move in one of the axis directions in the three-axis coordinate system within the set adjustment range with the set adjustment step, and step A is repeated until the N-th layer of spatial diffraction optical neural network operation unit is moved in the three-axis directions within the set adjustment range with the set adjustment step.

10. The positioning and assembling method according to claim 8, characterized in that: The distance between the first photodetector and the first layer of spatial diffraction optical neural network operation unit is equal to the distance between the second photodetector and the first layer of spatial diffraction optical neural network operation unit when the second photodetector is in the first set position.

Images