Model-based photoelectric hybrid computing system temperature calibration method and system, storage medium and computer system

Abstract

The invention discloses a model-based photoelectric hybrid computing system temperature calibration method and system, a storage medium and a computer system. The method comprises the following steps: setting a plurality of standard inputs, and obtaining a plurality of ideal outputs of a photoelectric hybrid computing system to be calibrated under the plurality of standard inputs; acquiring a plurality of actual outputs corresponding to the plurality of standard inputs on the plurality of temperature points of the photoelectric hybrid computing system to be calibrated according to a set temperature step length; constructing a data set, and training a calibration model by using the data set; and inputting the current system temperature and the system output into the calibration model to obtain corresponding ideal output. The photoelectric hybrid computing system to be calibrated is regarded as a'black box ', and only the final output of the photoelectric hybrid computing system is calibrated at a system level, so that the calibration difficulty is greatly reduced.

Description

Technical Field

[0001] This invention belongs to the field of optical chip technology, and particularly relates to a model-based temperature calibration method, system, storage medium, and computer system for optoelectronic hybrid computing systems. Background Technology

[0002] In AI computing, electronic chips play a crucial role in data transmission and computation. The entire process of artificial neural algorithms involves a large number of matrix multiplication operations, which traditional computer architectures like CPUs struggle to handle, resulting in low computational efficiency. Therefore, academia and industry have turned their attention to new hardware architectures specifically designed for artificial neural networks and deep learning, such as GPUs, ASICs, and FPGAs. However, regardless of the architecture, all these technologies utilize traditional microelectronics technology for design and manufacturing. Improving AI chip performance is inseparable from increasing microelectronic integration. However, in the early 21st century, microelectronics processes have struggled to keep pace with Moore's Law predictions, making it increasingly difficult to increase chip integration.

[0003] Compared to electrons, photons possess many unique properties: photons have no rest mass, there is no interaction between photons, and they experience almost no interference; different wavelengths of light can be used for simultaneous multi-channel communication, and photons can still achieve stable modulation and information transmission at modulation frequencies of tens of hertz, while electrical signals face radiation loss problems at high frequencies; furthermore, optical signals are unaffected by electromagnetic fields, offering strong confidentiality. Even more uniquely, using certain optical structures, photons can perform mathematical operations with zero energy consumption. Therefore, photons can be used to achieve ultra-high-speed, low-energy, or even zero-energy computing, thereby breaking through the performance and cost bottlenecks of traditional microelectronic chips.

[0004] As the semiconductor industry gradually enters the post-Moore era, the development of integrated circuits continues to evolve in different directions. On the one hand, new semiconductor materials are being developed, especially carbon nanotubes and two-dimensional semiconductor materials, to continue the essence of Moore's Law and further shrink the size of devices or chips, i.e., "More Moore"; on the other hand, new architectures and heterogeneous integrated chips are being developed for specific application areas, such as neuromorphic chips, optoelectronic chips, and quantum chips, to achieve "More than Moore".

[0005] Photonic computing chips based on silicon-based optoelectronic technology integrate micron- and nano-scale photonic, electronic, and optoelectronic devices onto a single silicon substrate using materials and processes compatible with integrated circuits. This achieves functional integration and complementary advantages between microelectronic and optoelectronic devices, resulting in high-performance optoelectronic chips. This is an effective way to address the performance bottlenecks and information congestion faced by traditional integrated circuits. Benefiting from the mature application of fiber optic communication, photons, as information carriers, have more multiplexing dimensions compared to electrons, such as amplitude, phase, wavelength, and mode, thus possessing greater bandwidth, faster speed, and lower energy consumption. Early silicon-based optoelectronic chips were developed to replace copper interconnect technology and solve the communication bottleneck between the processor core and memory of microelectronic chips. The microprocessor and memory units were implemented by microelectronic devices, while photonic devices mainly performed signal transmission and reception. With the increasing maturity of silicon photonics technology and the significant advantages of optical communication, attention to silicon photonic computing chips has gradually shifted from information transmission to information processing, including cutting-edge application areas such as analog computing, quantum computing, and neuromorphic computing.

[0006] The computation process of a photonic computing chip includes optical signal generation, optical quality conditioning, logical operations on the optical signal, and optical signal detection and decoding. This entire process requires the coordinated operation of tens or even hundreds of thousands of optical devices. Some of these devices are affected by temperature, with their characteristics changing accordingly. For example, the output of a directional coupler will drift under different temperature conditions; the photodetector's photoelectric conversion responsivity will also change under different temperature conditions. In the actual application environment of photonic computing chips, the temperature cannot be completely controlled, causing these variations to affect the accuracy of the final output of the photonic computing chip. Summary of the Invention

[0007] The purpose of this invention is to provide a model-based temperature calibration method, system, storage medium, and computer system for optoelectronic hybrid computing systems, which partially solves or alleviates the above-mentioned deficiencies in the prior art and can calibrate the temperature drift of optoelectronic hybrid computing systems.

[0008] To address the aforementioned technical problems, the present invention specifically adopts the following technical solution: a model-based temperature calibration method for a photoelectric hybrid computing system, comprising: Set several standard inputs and obtain several ideal outputs of the optoelectronic hybrid computing system to be calibrated under the given standard inputs; According to the set temperature step, obtain several actual outputs corresponding to several standard inputs at several temperature points of the optoelectronic hybrid computing system to be calibrated; Construct a dataset, in which samples include the ideal and actual outputs of the system to be calibrated at several temperature points; A calibration model is trained using a dataset, which can predict the ideal output at a given temperature point based on the actual output. During the operation of the optoelectronic hybrid computing system to be calibrated, the system output is checked for drift caused by temperature. If so, the current system temperature and system output are input into the calibration model to obtain the corresponding ideal output.

[0009] As an improvement, the steps for setting the temperature step size include: The initial temperature step size is calculated based on the operating temperature range of the optoelectronic hybrid computing system to be calibrated and the available storage space. The system temperature is adjusted according to the initial temperature step size, and the actual output of the optoelectronic hybrid computing system to be calibrated is obtained at each temperature point under standard input. The system compares the actual output of the optoelectronic hybrid computing system to be calibrated at adjacent temperature points. If the absolute value of the difference is less than the difference threshold, the current step size is added to the preset incremental step size as the step size for the next adjustment. If the absolute value of the difference is within the difference threshold, the step size remains unchanged. If the absolute value of the difference is greater than the difference threshold, the current step size is subtracted from the incremental step size as the step size for the next adjustment.

[0010] As an improvement, the steps for calculating the initial temperature step size based on the operating temperature range of the optoelectronic hybrid computing system to be calibrated and the available storage space include: The number of temperature points is the result of dividing the available storage space by the space occupied by the data of a single temperature point. The result of dividing the operating temperature range by the number of temperature points is used as the initial temperature step size.

[0011] As an improvement, the system temperature can be adjusted upwards from the lower limit of the normal operating temperature range of the optoelectronic hybrid computing system to be calibrated, or downwards from the upper limit of the normal operating temperature range of the optoelectronic hybrid computing system to be calibrated.

[0012] As an improvement, the next increment step size is half of the previous increment step size.

[0013] As an improvement, the calibration model is at least one of the following: regression model, neural network model, support vector machine model, decision tree and random forest model, and Bayesian network.

[0014] As an improvement, the calibration model is a linear regression model; the mathematical expression of the linear regression model is: y=a0c 2 +a1c+a2; Where y is the ideal output, c is the actual output, and a0, a1, and a2 are coefficients; Its loss function is: L= ; Where L is the loss value, y i For the i-th ideal output at this temperature point, c i This represents the i-th actual output at that temperature point, where n is the number of ideal outputs or ideal inputs.

[0015] This invention also provides a model-based temperature calibration system for a hybrid optoelectronic computing system, comprising: An ideal output acquisition module is used to set several standard inputs and acquire several ideal outputs of the optoelectronic hybrid computing system to be calibrated under the several standard inputs; The actual output acquisition module is used to acquire several actual outputs corresponding to several standard inputs at several temperature points of the optoelectronic hybrid computing system to be calibrated, according to a set temperature step. The dataset construction module is used to construct a dataset, in which samples include the ideal output and actual output of the system to be calibrated at several temperature points; The model training module is used to train a calibration model using a dataset. The calibration model can predict the ideal output at a certain temperature point based on the actual output at that temperature point. The calibration module is used to detect whether the system output is drifted due to temperature during the operation of the optoelectronic hybrid computing system to be calibrated; if so, the current system temperature and system output are input into the calibration model to obtain the corresponding ideal output.

[0016] The present invention also provides a storage medium storing a computer program; when the computer program is executed, it can implement the above-described model-based photoelectric hybrid computing system temperature calibration method.

[0017] The present invention also provides a computer system, including a processor and a memory; the memory stores a computer program; when the computer program is executed by the processor, it can implement the above-described model-based optoelectronic hybrid computing system temperature calibration method.

[0018] Beneficial effects: This invention first obtains several ideal outputs under several standard inputs, and then obtains the actual outputs under standard inputs at several temperature points within the system's operating temperature range according to a set temperature step size. A dataset is then constructed based on the actual and ideal outputs, and a calibration model is trained using this dataset. Therefore, after manufacturing, when the system is running, if temperature drift occurs, the current system temperature and the corresponding system output are input into the trained calibration model to obtain the predicted ideal output, thereby compensating for the temperature of the system output.

[0019] In this invention, the optoelectronic hybrid computing system to be calibrated is treated as a "black box," without analyzing which specific optical components inside will drift due to temperature and the amount of drift. Instead, the entire system is treated as a whole, and only its final output is calibrated at the system level, thereby greatly reducing the difficulty of calibration. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. The elements or parts in the drawings are not necessarily drawn to scale. Obviously, the drawings described below are some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0021] Figure 1 This is a flowchart of Embodiment 1 of the present invention; Figure 2 This is a structural diagram of Embodiment 2 of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0023] In this document, suffixes such as "module," "part," or "unit" used to denote elements are used only for the purpose of illustrative purposes and have no specific meaning in themselves. Therefore, "module," "part," or "unit" may be used interchangeably.

[0024] In this document, the terms "upper," "lower," "inner," "outer," "front," "rear," "one end," and "the other end," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the present invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated 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.

[0025] In this document, unless otherwise explicitly specified and limited, the terms "installed," "equipped with," "connected," etc., should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, a direct connection, or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0026] In this document, "and / or" includes any and all combinations of one or more of the listed related items.

[0027] In this article, "multiple" means two or more, that is, it includes two, three, four, five, etc.

[0028] Example 1: Existing technology Chinese patent application CN202411380839.2 discloses a performance testing and calibration method for photonic interconnect chips based on deep machine learning algorithms, including calibrating the output of the light source, calculating the root mean square error of the light source power over time; considering the connection loss and alignment loss between the optical fiber and the chip, calculating the insertion loss and return loss; using a deep learning analysis algorithm for photonic chips to analyze the spectral response and phase response of the chip, and calculating the spectral response width; measuring the output signal of the photonic interconnect chip to generate an eye diagram, calculating the signal-to-noise ratio and bit error rate; placing the photonic interconnect chip in an environment with adjustable temperature, introducing the influence of material thermal expansion to calculate the temperature coefficient, and establishing an optical output power model.

[0029] In the aforementioned prior art, the photonic interconnect chip is placed in an environment with adjustable temperature, temperature data is collected in real time, the influence of material thermal expansion is introduced to calculate the temperature coefficient, and a model of output power is established to analyze the changes in the chip's optical output power under different temperature conditions. It is evident that the aforementioned prior art only analyzes the impact of temperature on the output of the photonic computing chip, but does not calibrate this impact. Furthermore, the impact of ambient temperature on the system differs from the impact of the system's (actual operating) temperature generated by the operation of its components.

[0030] In view of this, such as Figure 1 As shown, this invention provides a model-based temperature calibration method for a hybrid optoelectronic computing system, comprising: S1 sets several standard inputs and obtains several ideal outputs of the optoelectronic hybrid computing system to be calibrated under the several standard inputs.

[0031] For an optoelectronic computing system, everything from the light source module that emits light signals, the conditioner that modulates the light signals, the computing module that performs logic operations, and the photoelectric converter that converts light signals into electrical signals will be affected by temperature and drift. Moreover, the drift ratio of each component is not the same at different temperatures. Therefore, it is extremely difficult to calibrate the temperature drift of each component. Furthermore, even if each component is calibrated, the final output of the system (or the output of a certain node in the system, such as the output of the PD) is not necessarily the ideal output or close to the ideal output.

[0032] In contrast, this invention takes a different approach, treating the optoelectronic hybrid computing system to be calibrated as a "black box." Instead of analyzing which specific optical components within it will experience temperature drift and the extent of that drift, the entire system is considered as a whole, and only its final output is calibrated at the system level. This significantly reduces the difficulty of calibration.

[0033] The purpose of this step is to obtain the expected standard output of the optoelectronic hybrid computing system to be calibrated under different inputs. In other words, under ideal conditions, without considering the influence of temperature, the ideal output value of the system under a certain standard output, and use this ideal output as the basis for calibration. For example, this can be obtained through a simulation system.

[0034] Understandably, in order to improve the prediction accuracy of a model, the number of samples used for training needs to be sufficient and as accurate as possible.

[0035] S2 acquires several actual outputs corresponding to several standard inputs at several temperature points of the optoelectronic hybrid computing system to be calibrated, according to the set temperature step size.

[0036] Since optical devices are affected by temperature in different proportions, for example, the change in temperature by 10° is not necessarily twice that of temperature by 5°.

[0037] Therefore, this step requires exhaustively listing all possible temperatures. It's understandable that a smaller temperature step size results in more output temperature points, leading to higher calibration accuracy. However, more temperature points also increase computational load and system overhead.

[0038] Therefore, it is necessary to set a suitable temperature step size so that the difference between adjacent temperature points is within the system's tolerance range, while minimizing the number of temperature points. The steps for setting the temperature step size in this process include: S21 calculates the initial temperature step size based on the operating temperature range of the optoelectronic hybrid computing system to be calibrated and the available storage space.

[0039] Hybrid computing systems typically have a defined operating temperature range, such as [-5℃, 45℃]. Each hybrid computing system to be calibrated also has a fixed storage space, thus the available space for data storage is limited. Therefore, these two constraints must be fully considered when setting the initialization step size. In this embodiment, the steps for calculating the initialization temperature step size based on the operating temperature range and available storage space of the hybrid computing system to be calibrated include: S211 divides the available storage space by the space occupied by a single temperature point data as the number of temperature points.

[0040] First, we can estimate the available storage space and the space occupied by each temperature point data. For example, the entire system is pre-allocated to 10KB for storing temperature point data in this invention (of course, some spare space will be reserved for storing other data). According to the estimate, each temperature point data occupies about 0.5KB of space. Therefore, we can roughly estimate that the total storage space used to store temperature data points can store about 20 temperature points.

[0041] S212 uses the result of dividing the operating temperature range by the number of temperature points as the initial temperature step size.

[0042] Theoretically, to distribute 20 temperature points within a 50℃ temperature range, the temperature step size should be 2.5℃. Therefore, 2.5℃ can be used as the initial step size, and then the temperature step size can be further adjusted subsequently.

[0043] S22 adjusts the system temperature according to the initial temperature step size, and calibrates the actual output of the optoelectronic hybrid computing system at each temperature point by acquiring standard input.

[0044] In this embodiment, the adjustment can be made from the upper limit of the system operating temperature range downwards or from the lower limit of the system operating temperature range upwards. Continuing the previous example, the specified system operating temperature range is [-5℃, 45℃], and the initial step size is 2.5℃. If the adjustment starts from the lower limit, the output value at the input value at the lower limit of -5℃ is recorded first, and then the output value at -2.5℃ is recorded.

[0045] S23 compares the actual output of the optoelectronic hybrid computing system to be calibrated at adjacent temperature points of the system; if the absolute value of the difference is less than the difference threshold range, the current step size plus the preset incremental step size is used as the step size for the next adjustment; if the absolute value of the difference is within the difference threshold range, the step size remains unchanged; if the absolute value of the difference is greater than the difference threshold range, the current step size minus the incremental step size is used as the step size for the next adjustment.

[0046] In this embodiment, the step size is not fixed but is adaptively adjusted. By comparing the actual outputs at two adjacent temperature points under the same standard input, if the absolute value of the difference is less than the difference threshold, it is considered that the difference between the actual outputs at the two temperature points is outside the system's tolerance range and is too dense. This embodiment pre-sets incremental compensation; in the above situation, the current compensation step size plus the incremental step size is used as the step size for the next step.

[0047] If the absolute value of the difference between the two is within the difference threshold range, then the difference between the actual outputs at the two temperature points is considered to be tolerable by the system. During calibration, even if the -2.5℃ temperature point is not available, the output at -5℃ is still within the system's allowable deviation range. In the above case, the current step size remains unchanged.

[0048] Of course, if the absolute value of the difference between the two exceeds the difference threshold, they are considered to be irreplaceable, and interpolation is required. Therefore, the step size needs to be shortened to obtain a more densely distributed output value. In the above case, the current step size minus the incremental step size is used as the step size for the next step.

[0049] In addition, the increment step size itself is not fixed. In order to reach the difference threshold range as soon as possible, the next increment step size can be set to 1 / 2 of the previous increment step size.

[0050] The same rules apply until the entire operating temperature range of the system is traversed.

[0051] S3 constructs a dataset, in which samples include the ideal and actual outputs of the system to be calibrated at several temperature points.

[0052] After collecting samples, the data needs to be processed, such as data cleaning to remove outliers and missing values. Normalization may also be performed if necessary.

[0053] In this embodiment, a sample includes a temperature point, the actual output at that temperature point, and the corresponding ideal output. The dataset also needs to be divided, for example, 70% as the training set and 30% as the test set.

[0054] S4 uses a dataset to train a calibration model, which can predict the ideal output at a given temperature point based on the actual output.

[0055] This invention does not limit the specific type of calibration model. For example, regression models, neural network models, support vector machine models, decision tree and random forest models, and Bayesian networks can all be used as calibration models in this invention.

[0056] In this embodiment, a linear regression model is used as the calibration model for explanation.

[0057] Specifically, the mathematical expression of the linear regression model is: y=a0c 2 +a1c+a2; Where y is the ideal output, c is the actual output, and a0, a1, and a2 are coefficients; Its loss function is: L= ; Where L is the loss value, y i For the i-th ideal output at this temperature point, c i This represents the i-th actual output at that temperature point, where n is the number of ideal outputs or ideal inputs.

[0058] The regression model is trained using a training set, and parameters are adjusted and the model optimized. The model's performance is evaluated using a test set to minimize the loss function. The model is then adjusted and retrained as needed to achieve the desired predictive performance.

[0059] S5 detects whether the system output is drifted due to temperature during the operation of the optoelectronic hybrid computing system to be calibrated; if so, it inputs the current system temperature and system output into the calibration model to obtain the corresponding ideal output.

[0060] Once the calibration model is trained, it can be applied to the system's temperature drift calibration. However, to conserve system resources, not all system inputs will be calibrated. This embodiment can also include a detection mechanism. If a sudden temperature change is detected, causing the system output to be affected by temperature drift beyond the system's tolerance range, the trained calibration model will be activated to calibrate the system output. During calibration, only the current system temperature and the actual system output need to be input into the calibration model; the model can then predict the corresponding ideal input, thereby achieving temperature drift calibration.

[0061] Furthermore, in this embodiment, the system output being calibrated is the output electrical signal of the PD (Photodiode) module in the optoelectronic hybrid computing system to be calibrated. This is because the PD module is the last optical device in the entire photonic computing chip, and its function is to convert optical signals into electrical signals. Errors in the optical signals caused by temperature throughout the system will eventually accumulate in the PD module, and the PD module itself is also subject to significant temperature-induced drift. Therefore, in this embodiment, the output of the PD module is used as the object of calibration.

[0062] In other embodiments, because storage space and system operating temperature range are considered when setting the step size, and the step size is not a fixed value but rather an adaptively adjusted preset incremental step size; on the other hand, since the calibration is not performed on a single component within the system, but rather the entire system is treated as a "black box" for system-level calibration, any failure in any component within the system or a failure in the links between components may affect the adjustment of the step size. Therefore, to ensure calibration accuracy and reliability, it is also necessary to verify the calibrated temperature values. Specifically, at multiple specified test temperatures T... j (where j is the index of the test temperature) Given N standard inputs P1-P for each temperature. N Thus, the actual output I of the system at each test temperature can be obtained. j1 '— I jN ', and the ideal output I after calibration by the calibration model. j1 ''— I jN Then compare the ideal output I obtained from the calibration. j1 ''— I jN The variance of '' is equal to the ideal output I corresponding to the standard input obtained in step S1. 11 —I 1N If the average relative error between data groups is less than the preset error threshold, it indicates that the overall difference between the calibrated ideal output data group and the ideal output data group in step S1 is not significant, or in other words, the calibrated ideal data group is relatively close to the ideal output data group in step S1. Therefore, the verification is passed, and the temperature step size set in step S2 is not only reasonable but also meets the accuracy requirements. Conversely, if the average relative error is greater than or equal to the preset error threshold, the verification is not passed, and the incremental step size in step S23 needs to be adjusted accordingly. For example, the incremental step size can be reduced.

[0063] Example 2: Figure 2 As shown, the present invention also provides a model-based temperature calibration system for a hybrid optoelectronic computing system, comprising: An ideal output acquisition module is used to set several standard inputs and acquire several ideal outputs of the optoelectronic hybrid computing system to be calibrated under the several standard inputs; The actual output acquisition module is used to acquire several actual outputs corresponding to several standard inputs at several temperature points of the optoelectronic hybrid computing system to be calibrated, according to a set temperature step. The dataset construction module is used to construct a dataset, in which samples include the ideal output and actual output of the system to be calibrated at several temperature points; The model training module is used to train a calibration model using a dataset. The calibration model can predict the ideal output at a certain temperature point based on the actual output at that temperature point. The calibration module is used to detect whether the system output is drifted due to temperature during the operation of the optoelectronic hybrid computing system to be calibrated; if so, the current system temperature and system output are input into the calibration model to obtain the corresponding ideal output.

[0064] Example 3: The present invention also provides a storage medium storing a computer program; when the computer program is executed, it can implement the above-mentioned model-based photoelectric hybrid computing system temperature calibration method.

[0065] Example 4: The present invention also provides a computer system, including a processor and a memory; the memory stores a computer program; when the computer program is executed by the processor, it can implement the above-mentioned model-based optoelectronic hybrid computing system temperature calibration method.

[0066] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0067] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a computer terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0068] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A model-based temperature calibration method for an optoelectronic hybrid computing system, characterized in that include: Setting a number of standard inputs and obtaining a number of ideal outputs of the optoelectronic hybrid computing system to be calibrated under the number of standard inputs; According to the set temperature step, a plurality of actual outputs corresponding to a plurality of standard inputs of the optoelectronic hybrid computing system to be calibrated at a plurality of temperature points are obtained; Constructing a data set, wherein samples in the data set include ideal outputs and actual outputs of the system to be calibrated at several temperature points; Using the data set to train a calibration model, the calibration model can predict an ideal output at a certain temperature point based on the corresponding actual output; When the optoelectronic hybrid computing system to be calibrated is running, detecting whether the system output has drift caused by temperature influence; If yes, the current system temperature and system output are input into the calibration model to obtain the corresponding ideal output.

2. A method for temperature calibration of an optoelectronic hybrid computing system based on a model according to claim 1, characterized in that The steps to set the temperature step size include: Calculate the initialization temperature step length according to the operating temperature range of the optoelectronic hybrid computing system to be calibrated and the available storage space; The system temperature is adjusted according to the initialization temperature step, and the actual output of the optoelectronic hybrid computing system to be calibrated under the standard input is obtained at each temperature point; Compare the actual outputs of the optoelectronic hybrid computing system to be calibrated at adjacent temperature points of the system; if the absolute value of the difference is less than the difference threshold range, add the current step size to the preset incremental step size as the step size for the next adjustment; if the absolute value of the difference is within the difference threshold range, the step size remains unchanged; if the absolute value of the difference is greater than the difference threshold range, subtract the incremental step size from the current step size as the step size for the next adjustment.

3. A method for temperature calibration of an optoelectronic hybrid computing system based on a model according to claim 1, characterized in that The steps of calculating the initialization temperature step length according to the operating temperature range of the optoelectronic hybrid computing system to be calibrated and the available storage space include: The result of dividing the available storage space by the space occupied by the data of a single temperature point is taken as the number of temperature points; The result of dividing the operating temperature range by the number of temperature points is used as the initialization temperature step size.

4. The method for temperature calibration of a model-based optoelectronic hybrid computing system according to claim 2, characterized in that: The system temperature is adjusted upward from the lower limit of the normal operating temperature range of the optoelectronic hybrid computing system to be calibrated, or the system temperature is adjusted downward from the upper limit of the normal operating temperature range of the optoelectronic hybrid computing system to be calibrated.

5. The method for temperature calibration of a model-based optoelectronic hybrid computing system according to claim 2, characterized in that: The next increment step is 1 / 2 of the previous increment step.

6. The method for temperature calibration of a model-based optoelectronic hybrid computing system according to claim 1, characterized in that: The calibration model is at least one of a regression model, a neural network model, a support vector machine model, a decision tree and random forest model, and a Bayesian network.

7. The method for temperature calibration of an optoelectronic hybrid computing system based on a model according to claim 6, characterized in that: The calibration model is a linear regression model; the mathematical expression of the linear regression model is: <h2 style=";text-align:left;direction:ltr">y=a0c<h2 style=";text-align:left;direction:ltr"> 2 <h2 style=";text-align:left;direction:ltr"> +a1c+a2; Among them, y is the ideal output, c is the actual output, a0, a1, a2 are coefficients; Its loss function is: L= ; Where L is the loss value, y i is the i-th ideal output at this temperature point, c i is the ith actual output at that temperature point, and n is the number of ideal outputs or ideal inputs.

8. A model-based optoelectronic hybrid computing system temperature calibration system, characterized in that include: An ideal output acquisition module, used to set a number of standard inputs and obtain a number of ideal outputs of the optoelectronic hybrid computing system to be calibrated under the number of standard inputs; The actual output acquisition module is used to obtain a number of actual outputs corresponding to a number of standard inputs of the optoelectronic hybrid computing system to be calibrated at a number of temperature points according to a set temperature step; A data set construction module, used to construct a data set, wherein the samples in the data set include the ideal output and the actual output of the system to be calibrated at several temperature points; A model training module, used to train a calibration model using a data set, wherein the calibration model can predict an ideal output at a certain temperature point based on the corresponding actual output; A calibration module, used for detecting whether the system output has drift caused by temperature when the optoelectronic hybrid computing system to be calibrated is running; If yes, the current system temperature and system output are input into the calibration model to obtain the corresponding ideal output.

9. A storage medium, characterized in that: The storage medium stores a computer program; when the computer program is executed, it can implement the model-based optoelectronic hybrid computing system temperature calibration method described in any one of claims 1 to 7.

10. A computer system, characterized in that: It comprises a processor and a memory; the memory stores a computer program; when the computer program is executed by the processor, it can implement the model-based optoelectronic hybrid computing system temperature calibration method described in any one of claims 1 to 7.

Images