Method and system for verifying calculation output stability of optical calculation chip

Abstract

The invention discloses a method and system for verifying the calculation output stability of an optical calculation chip, and the method comprises the steps: repeatedly collecting a plurality of calculation result curves of the same point location of the optical calculation chip under the same input condition; each output result curve is composed of a plurality of steps, and each step is a calculation result value; performing difference calculation on adjacent data points of each calculation output result curve to obtain a plurality of difference results; selecting a data point corresponding to the difference result of which the absolute value is greater than the difference threshold value as fixed point data, and positioning a step in an output result curve; selecting data points from the corresponding calculation output result curve based on the fixed point data to construct steps, and forming a to-be-aligned step curve; step curves to be aligned are placed in a plane coordinate system A, and the starting points of the step curves to be aligned are aligned on an axis representing time, so that an aligned image is formed. According to the method, the calculation output stability of the optical calculation chip can be quickly judged.

Description

Technical Field

[0001] The present invention belongs to the technical field of optical computing chip testing, and in particular relates to a method and system for verifying the stability of computing output of an optical computing chip. Background Art

[0002] As a new type of computing device based on photonic technology, optical computing chips have the advantages of high-speed parallel processing and low energy consumption, and show broad application prospects in large-scale matrix operations, neural network acceleration and other fields.

[0003] The output stability of a photonic computing chip is a key indicator for measuring the consistency and reliability of the chip's output results when performing computing tasks. It is mainly reflected in four aspects: consistency of the output results, stability of signal characteristics, anti-interference capability, and long-term operation stability.

[0004] Among them, the computing output stability of optical computing chips is affected by factors such as optical device characteristics and signal transmission noise. For example, due to long running time or device failure, problems such as reduced and inconsistent computing stability are prone to occur, and a reliable stability verification method is urgently needed.

[0005] Currently, stability testing of integrated circuits typically uses methods such as electrical signal timing analysis and error-checking codes. However, the computational output signals of optical computing chips are unique, and verification methods for traditional electronic chips are difficult to apply directly. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for verifying the stability of the calculation output of an optical computing chip, which partially solves or alleviates the above-mentioned shortcomings in the prior art and can quickly judge the stability of the calculation output of the optical computing chip.

[0007] In order to solve the above-mentioned technical problems, the first aspect of the present invention is to provide a method for verifying the output stability of an optical computing chip, comprising: Under the same input conditions, repeatedly collect several output result curves of the same point of the optical computing chip; each output result curve consists of several steps, and each step is an output result value; Perform differential calculations on adjacent data points of each output result curve to obtain several differential results; Select the data points corresponding to the difference results whose absolute values ​​are greater than the difference threshold as fixed-point data to locate the steps in the output result curve; Selecting data points from the corresponding output result curve based on the fixed-point data to construct steps, thereby forming a step curve to be aligned; The step curves to be aligned are placed in a plane coordinate system A, and the starting point of each step curve to be aligned is aligned on the axis representing time, thereby forming an aligned image; The output stability of the optical computing chip is determined based on the aligned images.

[0008] Furthermore, before performing the difference calculation, each output result is downsampled.

[0009] Furthermore, the method of downsampling the output result is to use the formula: ; Determine the initial sampling interval; where, is the sampling interval, N is the number of data points of the output result curve, m is the number of steps contained in the output result curve, and k is the number of data points required to express one step. is the floor rounding symbol; Downsample the output results according to the initial sampling interval.

[0010] Furthermore, an adjustment interval is preset to downsample the output result according to the initial sampling interval; If there is fixed-point data in the output result curve that has not been captured, the output result is downsampled again after adding the preset adjustment interval to the current sampling interval; repeat this step until all fixed-point data in the output result curve are captured; If all fixed-point data in the calculated output result curve are captured but the amount of data after downsampling is greater than or equal to the data amount threshold, the output result curve is downsampled again after subtracting the preset adjustment interval from the current sampling interval; repeat this step until all fixed-point data in the output result curve are captured and the amount of data after downsampling is less than the data amount threshold.

[0011] Furthermore, if the number of fixed-point data in a certain output result curve is less than a fixed-point data number threshold, the output result curve is excluded.

[0012] Furthermore, the method of selecting fixed-point data includes: All differential results of a certain output result curve are placed in the plane coordinate system B, and the data points corresponding to the differential results greater than the differential threshold are selected as fixed-point data.

[0013] Furthermore, the two data points forming each differential result are used as fixed-point data, and n data points before the previous fixed-point data, m data points after the next fixed-point data, and all data points between the two fixed-point data in the output result curve are selected to construct a step.

[0014] Furthermore, the method for determining the stability of the optical computing chip's computing output based on the aligned image includes: The horizontal axis of the plane coordinate system A is time, and the vertical axis is the value of the data point; If the height of each step in the aligned image does not exceed the height threshold, the output stability of the optical computing chip is determined to be qualified.

[0015] The present invention also provides a system for verifying the stability of computing output of an optical computing chip, which is characterized by comprising: The acquisition module is used to repeatedly acquire several calculation output result curves of the same point of the optical computing chip under the same input conditions; each output result curve consists of several steps, and each step is an output result value; The differential calculation module is used to perform differential calculation on adjacent data points of each output result curve to obtain a number of differential results; A fixed-point data selection module is used to select data points corresponding to the differential results whose absolute values ​​are greater than the differential threshold as fixed-point data, and locate the steps in the output result curve; a step curve construction module, configured to select data points from the corresponding output result curve based on the fixed-point data to construct steps, thereby forming a step curve to be aligned; an alignment module, configured to place the step curves to be aligned in a plane coordinate system A, and align the starting point of each step curve to be aligned on an axis representing time, thereby forming an aligned image; The judgment module is used to judge the stability of the optical computing chip's calculation output according to the aligned image.

[0016] As an improvement, it also includes: The downsampling module is used to downsample each output result before performing the difference calculation.

[0017] Beneficial technical effects: This method determines the initial sampling interval based on a formula and dynamically adjusts it based on the fixed-point data capture and data volume threshold, significantly reducing the data volume while preserving the key characteristics of the step signal. This method avoids the high computational complexity associated with direct processing of the raw data, significantly improving the efficiency of subsequent steps such as differential calculations and curve alignment. The downsampling interval can be dynamically adjusted based on actual conditions. If fixed-point data is not captured, the sampling interval is increased; if the data volume is too large, the sampling interval is decreased. This method can adapt to calculation result curves of different scales and characteristics, enhancing the universality of the solution for various optical computing chip testing scenarios.

[0018] By setting a peak number threshold, invalid curves with unclear step characteristics are automatically excluded to prevent invalid data from interfering with subsequent analysis processes, further streamline data processing volume, and ensure that analysis focuses on valid data.

[0019] Using differential calculation and threshold screening to determine fixed-point data, and combining data before, during, and after the fixed-point data to construct steps, this method accurately captures step boundaries and shapes, providing a reliable data foundation for stability analysis. Compared to traditional methods, this method more accurately identifies step signals and avoids misjudgments of stability due to step positioning errors.

[0020] After aligning the curves on the time axis, stability can be determined by setting step height thresholds. This is efficient and quick, making it suitable for rapid testing of optical computing chips. Furthermore, parameters such as differential thresholds and height thresholds can be flexibly set based on chip type and application scenario to meet diverse testing needs and ensure accurate chip stability assessment under different test conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the embodiments or the description of the prior art. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the various elements or parts are not necessarily drawn according to the actual scale. Obviously, the drawings described below are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without inventive work.

[0022] Figure 1 This is a flow chart of Embodiment 1 of the present invention; Figure 2 A schematic diagram of a single output result curve.

[0023] Figure 3 A schematic diagram of multiple output result curves.

[0024] Figure 4 Schematic diagram of peak screening.

[0025] Figure 5 This is a schematic diagram of multiple step curves aligned.

[0026] Figure 6 This is a structural diagram of embodiment 2 of the present invention. DETAILED DESCRIPTION

[0027] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0028] Herein, suffixes such as "module," "component," or "unit" used to represent elements are only used to facilitate description of the present invention and have no specific meaning. Therefore, "module," "component," or "unit" may be used interchangeably.

[0029] As used herein, terms such as "upper," "lower," "inner," "outer," "front," "back," "one end," and "the other end" indicate positions or locations based on those shown in the accompanying drawings. These terms are intended solely to facilitate and simplify the description of the present invention and are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limitations on the present invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0030] As used herein, unless otherwise expressly specified or limited, the terms "installed," "provided with," and "connected" should be understood broadly. For example, "connected" may refer to a fixed connection, a detachable connection, or an integral connection; it may refer to a mechanical connection, a direct connection, an indirect connection via an intermediate medium, or internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention on a case-by-case basis.

[0031] As used herein, "and / or" includes any and all combinations of one or more of the associated listed items.

[0032] Herein, "plurality" means two or more than two, ie, it includes two, three, four, five, etc.

[0033] Example 1: The purpose of the optical computing chip's compute output consistency test is to verify the stability and step discrimination of a single channel in the optical chip, proving its ability to support the required bit signal. Specifically, under fixed temperature and optical power conditions, a single channel of the optical chip (e.g., the output of at least one waveguide bus) is selected. A stepped modulation voltage consisting of several steps (beginning at a reference, increasing in steps, and then decreasing) is input through a function / arbitrary waveform generator. Multiple electrical modulation cycles are then performed on the optical chip. Finally, the repeatability of the step output signal over these multiple sampling cycles is evaluated to assess its stability.

[0034] In order to quickly verify the stability of a certain optical computing chip's repeated calculation output, such as Figure 1 As shown, this embodiment provides a method for verifying the stability of the computing output of an optical computing chip, and the specific steps include: S1 repeatedly collects several calculation output result curves of the same point of the optical computing chip under the same input conditions; each calculation output result curve consists of several steps, and each step is a calculation output result value.

[0035] When an optical computing chip is operating, its internal computational process should theoretically remain consistent for the same input data, resulting in the same output results. By collecting output curves from the same point multiple times under the same input conditions, multiple sets of comparable data can be obtained. For example, in some embodiments, when solving the same computational task, the same input conditions can be considered to be in place.

[0036] The results of a single acquisition are as follows Figure 2 As shown, the steps in the output result curve represent the output result values ​​of the chip.

[0037] like Figure 3 As shown in the figure, among multiple acquisition results, a single colored output curve represents the acquisition of a single computational output, and each "step" in the curve represents a computational result. (Optical computing chips are generally used for matrix calculations, so each output contains multiple computational result values, or multiple steps.) For example, a "step" can typically represent the computational result of a computational region (such as a computational unit).

[0038] If the computing pathway of the optical computing chip is stable, these output curves should have a high degree of consistency, especially the steps representing the output results, which should be close to consistency in terms of shape and amplitude.

[0039] S2 downsamples each output result curve data.

[0040] The raw output curves collected by optical computing chips often contain large amounts of data, and direct processing consumes significant computing resources and time. Downsampling selectively reduces the number of data points, minimizing the data size while preserving key information, such as step signal characteristics, and improving the efficiency of subsequent data processing.

[0041] More specifically, in S201, the method for downsampling the output result is to use the formula: ; Determine the initial sampling interval; where, is the sampling interval, N is the number of data points of the output result curve, m is the number of steps contained in the output result curve, and k is the number of data points required to express one step. is the floor rounding symbol; S202 downsamples the output result according to the initial sampling interval.

[0042] For example, for an output curve containing 75,003 data points, assuming that each step change requires at least 5 data points to ensure its clear identification, the total number of sampling points required for 48 steps is approximately 240. Therefore, the initial sampling interval is approximately 312.5 data points, which is rounded up to 312 data points.

[0043] For the other set of output result curves containing 50,003 data points, assuming that each step change requires at least 10 data points to ensure its accurate identification, the number of sampling points required for 32 steps is 320. Therefore, the initial sampling interval is approximately 156.26 data points, rounded to 156 data points.

[0044] An appropriate sampling interval can not only reduce data volume and improve the efficiency of subsequent data processing (such as differential calculation and step location), but also ensure that key step signal characteristics are not lost, ensuring accurate and reliable analysis of the optical computing chip's output stability. If the sampling interval is too large, key changes in the step signal may be missed; if the sampling interval is too small, the purpose of downsampling to reduce data volume is not achieved and the computational burden is increased.

[0045] The applicant noted that downsampling using the initial sampling interval can lead to two undesirable situations: first, critical fixed-point data (data points used to locate steps) is not captured, resulting in inaccurate step location and affecting the analysis of the stability of the optical computing chip's output; second, although all fixed-point data is captured, the amount of data after downsampling is still large, failing to achieve the goal of improving computing efficiency. Therefore, it is necessary to adjust and optimize the sampling interval to balance data integrity and processing efficiency.

[0046] To solve this problem, this embodiment provides a method for dynamically adjusting the sampling interval, which specifically includes: S211 presets an adjustment interval and downsamples the output result curve according to the initial sampling interval.

[0047] Since there is no result reference for the first downsampling, downsampling is still performed at the initial sampling interval determined in step S201. At the same time, an adjustment interval needs to be preset, which can be set according to actual conditions.

[0048] S212: If there is any fixed-point data in the output result curve that has not been captured, then re-downsample the calculation result after adding the preset adjustment interval to the current sampling interval; repeat this step until all fixed-point data in the output result curve are captured.

[0049] If fixed-point data in the output result curve is not captured, it means that the initial sampling interval or the current sampling interval is too large, and some key data points were skipped during the sampling process. In this case, the output result is re-downsampled by adding the preset adjustment step to the current sampling interval. For example, if the preset adjustment interval is 5 and the current sampling interval is 20, if any fixed-point data is not captured, the new sampling interval is adjusted to 20 + 5 = 25, and the downsampling operation is performed again using this new sampling interval. This process is repeated until all fixed-point data are captured.

[0050] If all fixed-point data in the output result curve are captured, but the amount of data after downsampling is greater than or equal to the data amount threshold, the output result is downsampled again after subtracting the preset adjustment interval from the current sampling interval; repeat this step until all fixed-point data in the output result curve are captured and the amount of data after downsampling is less than the data amount threshold.

[0051] If all fixed-point data in the output result curve is captured, but the amount of data after downsampling is greater than or equal to the data volume threshold, the current sampling interval is too small. Although the fixed-point data is fully preserved, the purpose of downsampling to reduce the data volume is not fully achieved. In this case, the output result is re-downsampled after subtracting the preset adjustment step from the current sampling interval. For example, if the preset adjustment step is 5 and the current sampling interval is 10, the amount of data after downsampling is still large and exceeds the data volume threshold. In this case, the new sampling interval is adjusted to 10 - 5 = 5, and downsampling is performed again. This step is repeated until all fixed-point data is captured and the amount of data after downsampling is less than the data volume threshold. This ensures that key information is not lost, effectively reduces the data volume, and improves the efficiency of subsequent data processing.

[0052] S3 performs differential calculation on adjacent data points of each output result curve after downsampling to obtain several differential results.

[0053] Differential calculation is a mathematical method used to analyze data changes. For an output curve, the differential calculation of adjacent data points involves subtracting the value of the previous data point from the value of the next data point. This calculation can highlight changes in data at adjacent locations. In the output curve of an optical computing chip, significant data changes occur at locations such as the rising and falling edges of step signals. Differential calculation can keenly capture these changes, providing important evidence for subsequent step location.

[0054] The output curve shows large jumps in the data at the rising and falling edges of steps. Through differential calculations, these jumps produce differential results with large absolute values. By selecting the data points corresponding to the differential results with absolute values ​​greater than the differential threshold as fixed-point data, we can accurately locate the steps' boundaries and, therefore, determine their location and number. The magnitude of the differential results also reflects the degree of signal variation at adjacent locations. In a stable output, data in the step plateau region exhibits minimal variation, and the differential result approaches zero; whereas, in the step transition region, the absolute value of the differential result is larger.

[0055] To illustrate with a simplified example, suppose an output curve has the following series of data point values: 1.2, 1.2, 1.2, 1.5, 1.5, 1.5, 1.0, 1.0, 1.0. After performing the difference calculation: First difference result: 1.2 - 1.2 = 0; Second difference result: 1.2 - 1.2 = 0; The third difference result: 1.5 - 1.2 = 0.3; Fourth difference result: 1.5 - 1.5 = 0; Fifth difference result: 1.5 - 1.5 = 0; Sixth difference result: 1.0 - 1.5 = -0.5; Seventh difference result: 1.0 - 1.0 = 0; Eighth difference result: 1.0 - 1.0 = 0.

[0056] S4 selects the data points corresponding to the differential results whose absolute values ​​are greater than the differential threshold as fixed-point data, and locates the steps in the output result curve.

[0057] When a step in the output curve changes—that is, when transitioning from one output value to another—large differences will occur between adjacent data points. Difference calculations can highlight these changes. Difference results with large absolute values ​​correspond to locations where the curve changes significantly, representing the boundaries of the steps. By setting a difference threshold, we filter out difference results with absolute values ​​greater than the threshold. The corresponding original data points can then be used as key fixed-point data to determine the location of the steps.

[0058] After obtaining the differential result sequence for each output result curve, the absolute value of each differential result is checked one by one. The original data points corresponding to the differential results whose absolute value is greater than the pre-set differential threshold are selected. These selected data points are the fixed-point data.

[0059] Continuing with the previous example, at the location where the data point changes from 1.2 to 1.5, the difference result is 0.3, a relatively large absolute value; at the location where the data point changes from 1.5 to 1.0, the difference result is -0.5, also a large absolute value. Assuming the difference threshold is set to 0.2, the absolute values ​​of the difference results corresponding to these two locations are greater than the threshold, and the corresponding data points can be used as fixed-point data to locate the step boundary. At other locations, the difference results are close to zero, indicating that the data change is small and is located in the step plateau area.

[0060] In this embodiment, Figure 4 As shown, all differential results of a particular output curve are placed in plane coordinate system B. Data points corresponding to 15 differential results (peaks) greater than a differential threshold, such as 0.1, are selected as fixed-point data. This screens out 16 consecutive, gradually ascending steps (one differential result corresponds to the previous and next steps, and adjacent differential results share a step, so 15 differential results correspond to 16 steps) as subsequent judgment material.

[0061] The computational output of an optical computing chip is presented as a step-like pattern, with local peaks in the curve closely related to the step characteristics. Generally, a stable and valid output differential curve should have a certain number of distinct local peaks to reflect clear step transitions. If the number of local peaks in a curve that meets the differential threshold condition falls below a preset threshold (e.g., 15), the step transitions in the curve are unclear and incomplete, failing to accurately represent the chip's output characteristics during normal computation. These unclear curves are often caused by external factors rather than the optical computing chip's computational consistency. Including such curves in subsequent processing can interfere with accurate assessment of chip output stability. For example, during curve alignment, invalid curves may complicate the alignment process due to their irregular step characteristics, leading to biased alignment results and failing to truly reflect the consistency of valid data. When calculating the repeatability of multiple curves to assess stability, invalid curves can lower the overall consistency index, misjudging chip stability, misinterpreting instability as normal fluctuations, or masking potential stability issues in other valid data.

[0062] Since a differential result is determined by two adjacent data points, if the number of fixed-point data in a certain output result curve is less than the fixed-point data number threshold, the output result curve is excluded.

[0063] S5 selects data points from the corresponding output result curve based on the fixed-point data to construct steps, thereby forming a step curve to be aligned.

[0064] Fixed-point data are determined through differential calculation and threshold screening. They mark the key locations where the step changes in the output curve occur. These fixed-point data are used as a benchmark for constructing the step because the data between the fixed-point data points and within a certain range before and after them together form a complete step signal. The two data points that form the differential result are selected as fixed-point data because the data change between these two points reflects the rise or fall of the step. The selection of n data points before the previous fixed-point data point and m data points after the next fixed-point data point ensures that the constructed step accurately reflects the changes in the chip's output.

[0065] For each output curve, first identify all eligible fixed-point data pairs. For example, suppose fixed-point data A and fixed-point data B form a pair. Then, select all data points between A and B. Simultaneously, select n data points before A and m data points after B from the output curve. These selected data points together form a step data set, thus constructing a complete step. This operation is repeated for each pair of fixed-point data on the curve, and all constructed steps are combined to form a step curve to be aligned. Note that m and n can be the same or different.

[0066] S6 places the step curves to be aligned in a plane coordinate system A, and aligns the starting point of each step curve to be aligned on the axis representing time, thereby forming an aligned image.

[0067] The output curves of optical computing chips are presented as steps, and the changes in these curves over time reflect the chip's computational process. Placing the step curves to be aligned in a plane coordinate system A and aligning them along the axis representing time allows for comparison of different curves based on the same time base. Because multiple output curves collected under the same input conditions should have consistent starting points and time progression, alignment on the time axis eliminates curve misalignment caused by time differences, allowing for intuitive comparison of the step changes of different curves at the same time location, leading to more accurate analysis of the chip's output stability.

[0068] First, determine that the horizontal axis of the plane coordinate system A is the time axis, and the vertical axis is the value of the data point (i.e., the output result value). For each step curve to be aligned, place the curve at the corresponding position in the coordinate system according to the time information corresponding to its data point. Then, through operations such as translation, make the starting point of each curve (or other specific reference point) coincide on the time axis to ensure that all curves are aligned in time. For example, if the time corresponding to the first data point of a curve is t1, and the time corresponding to the first data point of the other curve is t2, adjust the two curves through translation so that their first data points are in the same position on the time axis, and so on, to achieve alignment of all curves on the time axis. The aligned image is as follows Figure 5 shown.

[0069] That is to say, the present invention provides a data alignment method for optical computing chips.

[0070] Furthermore, in some embodiments, the method includes: S7 judges the output stability of the optical computing chip based on the aligned image.

[0071] Specifically, the judgment method in this embodiment is: The horizontal axis of the plane coordinate system A is time, and the vertical axis is the value of the data point, such as voltage; if the height of each step in the aligned image does not exceed the height threshold, the output stability of the optical computing chip is determined to be qualified.

[0072] When an optical computing chip performs stable calculations, its output should be highly consistent for the same input conditions. When the output curve is presented as steps, each step represents an output value, and the height of the step reflects the magnitude of that output. If the chip's calculations are stable, then at the same time point (horizontal axis), the step heights (vertical axis) corresponding to each curve should be relatively close, with minimal fluctuation. A height threshold is set to quantify the acceptable range of this fluctuation. When the height of each step does not exceed the threshold, the chip's output remains consistent within an acceptable range, thus confirming that the chip's output stability meets the requirements.

[0073] In plane coordinate system A, the horizontal axis represents time, and the vertical axis represents the data point value. First, observe the height of each step in the aligned image. For each step, compare the height values ​​of the different curves at that step location. If the difference between the height values ​​of all curves at that step location (i.e., the maximum difference between the height value of one curve and the height values ​​of the other curves) is less than the set height threshold, the step is considered to meet the height requirements. This evaluation is repeated for each step in turn. When the heights of all steps meet the threshold, the optical computing chip's output stability is considered acceptable. For example, if the height threshold is set to 0.5, and the height values ​​of the curves on a particular step are 2.0, 2.1, and 2.2, respectively, and the maximum difference between them is 0.2, which is less than 0.5, then the step's height meets the requirements.

[0074] The above method can quickly verify the output stability of the optical computing chip, which is conducive to efficient testing of the optical computing chip.

[0075] Of course, to more comprehensively and accurately determine chip output stability, multiple step characteristics can be considered. Besides step height, other factors can be analyzed, such as step position deviation, step width consistency, and the overall correlation coefficient between curves. Furthermore, a more scientific approach can be used to determine the height threshold, such as by statistically analyzing test data from a large number of stable chips to derive a reasonable threshold range.

[0076] In summary, this invention innovatively utilizes the step features exhibited by multiple calculation results in the form of curves to align the results from different calculation processes. This alignment of step features can mitigate or prevent the effects of curve alignment caused by offsets on the time axis (horizontal axis).

[0077] In other words, due to the time sequence of the calculation results in different calculation processes or due to some fixed device offsets, it is inevitable that there will be some differences on the time axis. In this regard, the present invention can alleviate the problem of difficulty in intuitively determining whether there are differences between curves, especially between steps.

[0078] Example 2: like Figure 6 As shown, this embodiment provides a system for verifying the output stability of an optical computing chip, which is characterized by including: The acquisition module is used to repeatedly acquire several output result curves of the same point of the optical computing chip under the same input conditions; each output result curve consists of several steps, and each step is an output result value; A downsampling module is used to downsample the data in each output result curve before performing the difference calculation; The differential calculation module is used to perform differential calculation on adjacent data points of each output result curve to obtain a number of differential results; A fixed-point data selection module is used to select data points corresponding to the differential results whose absolute values ​​are greater than the differential threshold as fixed-point data, and locate the steps in the output result curve; a step curve construction module, configured to select data points from the corresponding output result curve based on the fixed-point data to construct steps, thereby forming a step curve to be aligned; an alignment module, configured to place the step curves to be aligned in a plane coordinate system A, and align the starting point of each step curve to be aligned on an axis representing time, thereby forming an aligned image; Furthermore, in some embodiments, the system further comprises: The judgment module is used to judge the output stability of the optical computing chip according to the aligned image.

[0079] It should be understood that under the same input conditions, multiple calculations should produce highly similar output results. For example, when an optical computing chip processes a specific matrix operation, the deviation of the matrix element values ​​in each output is extremely small. In other words, when sampling the output result curve at the same point under the same input conditions, if the curve has good repeatability, the closer the corresponding output results are, indicating good stability of the calculation output; conversely, if the curves vary greatly, the stability is poor.

[0080] It is understood that the system of the present invention can implement any of the above methods or steps, which will not be described in detail here. It should be noted that, in this article, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "including a..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0081] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better embodiment. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), including a number of instructions for enabling a computer terminal (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0082] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are protected by the present invention.

Claims

1. A method for verifying the stability of the computing output of an optical computing chip, characterized in that include: Under the same input conditions, repeatedly collect several calculation output result curves at the same point of the optical computing chip; Each calculation output result curve consists of several steps, and each step is a calculation output result value; Perform differential calculations on adjacent data points of each calculation output result curve to obtain several differential results; Select the data points corresponding to the difference results whose absolute values ​​are greater than the difference threshold as fixed-point data to locate the steps in the output result curve; Selecting data points from the corresponding output result curve based on the fixed-point data to construct steps, thereby forming a step curve to be aligned; The step curves to be aligned are placed in a plane coordinate system A, and the starting point of each step curve to be aligned is aligned on the axis representing time, thereby forming an aligned image; The optical computing chip calculates output stability based on the aligned image.

2. The method for verifying the stability of computing output of an optical computing chip according to claim 1, characterized in that: Before performing the differential calculation, the data in each calculation output result curve is downsampled.

3. A method for verifying the stability of computing output of an optical computing chip according to claim 2, characterized in that The method of downsampling the output is to use the formula: ; Determine the initial sampling interval; where, is the sampling interval, N is the number of data points of the output result curve, m is the number of steps contained in the output result curve, and k is the number of data points required to express one step. is the floor rounding symbol; Downsample the calculation output results according to the initial sampling interval.

4. The method for verifying the stability of computing output of an optical computing chip according to claim 3, characterized in that: Preset the adjustment interval and downsample the output results according to the initial sampling interval; If there is fixed-point data in the output result curve that is not captured, the output result curve is downsampled again after adding the preset adjustment interval to the current sampling interval; Repeat this step until all the fixed-point data in the output result curve are captured; If all fixed-point data in the output result curve are captured but the amount of data after downsampling is greater than or equal to the data amount threshold, the output result is downsampled again after subtracting the preset adjustment interval from the current sampling interval; Repeat this step until all fixed-point data in the output result curve are captured and the amount of downsampled data is less than the data amount threshold.

5. The method for verifying the stability of computing output of an optical computing chip according to claim 1, characterized in that: If the number of fixed-point data in a calculation output result curve is less than the fixed-point data number threshold, the output result curve is excluded.

6. The method for verifying the stability of computing output of an optical computing chip according to claim 1, characterized in that Methods for selecting fixed-point data include: Place all differential results of a certain output result curve under plane coordinate system B, and select the value greater than the differential The data points corresponding to the difference results of the thresholds are used as fixed-point data.

7. The method for verifying the stability of computing output of an optical computing chip according to claim 1, characterized in that: The two data points that form each differential result are used as fixed-point data, and n data points before the previous fixed-point data, m data points after the next fixed-point data, and all data points between the two fixed-point data in the output result curve are selected to construct a step.

8. The method for verifying the stability of the computing output of an optical computing chip according to claim 1, characterized in that Methods for determining the output stability of a light computing chip based on aligned images include: The horizontal axis of the plane coordinate system A is time, and the vertical axis is the value of the data point; If the height of each step in the aligned image does not exceed the height threshold, the output stability of the optical computing chip is determined to be qualified.

9. A system for verifying the stability of the computing output of an optical computing chip, characterized in that include: The acquisition module is used to repeatedly acquire several calculation output result curves of the same point of the optical computing chip under the same input conditions; Each output result curve consists of several steps, each step is a calculation output result value; The differential calculation module is used to perform differential calculation on adjacent data points of each output result curve to obtain a number of differential results; A fixed-point data selection module is used to select data points corresponding to the difference results whose absolute values ​​are greater than the difference threshold as fixed-point data, and locate the steps in the calculation output result curve; a step curve construction module, configured to select data points from a corresponding calculation output result curve based on the fixed-point data to construct steps, thereby forming a step curve to be aligned; an alignment module, configured to place the step curves to be aligned in a plane coordinate system A, and align the starting point of each step curve to be aligned on an axis representing time, thereby forming an aligned image; The judgment module is used to judge the output stability of the optical computing chip according to the aligned image.

10. A system for verifying the stability of computing output of an optical computing chip according to claim 9, characterized in that Also includes: The downsampling module is used to downsample each output result data before performing the difference calculation.

Images