Crystal diameter prediction method and device, equipment, storage medium and product

Abstract

The invention discloses a crystal diameter prediction method and device, equipment, a storage medium and a product. The disclosed crystal diameter prediction method comprises the steps of obtaining current crystal pulling data; the current crystal pulling data are input into a preset data driving model, and the preset data driving model is obtained by training multiple pieces of actual crystal pulling data in an actual scene and corresponding actual crystal diameters and used for representing the corresponding relation between the crystal pulling data and the crystal diameters; and receiving a crystal diameter predicted value fed back by the data driving model based on the current crystal pulling data. The preset data driving model is obtained by training the actual crystal pulling data and the actual crystal diameter, so that the preset data driving model can learn the rule and mode of monocrystalline silicon growth in an actual scene, and the accuracy of the crystal diameter prediction value obtained based on the preset data driving model is relatively high; and an effective strategy can be provided for design of a control algorithm.

Description

Technical Field

[0001] The present application relates to the field of single crystal silicon technology, and in particular to a crystal diameter prediction method, device, equipment, storage medium and product. Background Art

[0002] Modeling of the single crystal silicon growth process by the Czochralski method is the primary issue in the development of crystal pulling control technology. By establishing an accurate model that describes the crystal growth process, the crystal diameter can be predicted, providing a strategy for the design of the control algorithm.

[0003] At present, the existing method is to establish a mathematical mechanism model based on the growth mechanism of single crystal silicon. However, there are many ideal assumptions in the establishment process of the mathematical mechanism model, which makes the mathematical mechanism model have a large difference from the law and pattern of single crystal silicon growth in actual scenarios, resulting in poor accuracy of crystal diameter predicted by the model, and unable to provide an effective strategy for the design of the control algorithm.

[0004] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of this application is to provide a crystal diameter prediction method, device, equipment, storage medium and product, aiming to solve the technical problem that there are large differences between the laws and patterns of single crystal silicon growth in the existing technology and actual scenarios, resulting in poor accuracy of the crystal diameter predicted by the model and inability to provide effective strategies for the design of control algorithms.

[0006] To achieve the above object, the present application provides a method for predicting crystal diameter, the method comprising:

[0007] Get the current crystal pulling data;

[0008] Inputting the current crystal pulling data into a preset data driving model, wherein the preset data driving model is obtained by training a plurality of actual crystal pulling data and corresponding actual crystal diameters in actual scenes, and is used to characterize the corresponding relationship between the crystal pulling data and the crystal diameter;

[0009] A crystal diameter prediction value fed back by the preset data-driven model based on the current crystal pulling data is received.

[0010] In one embodiment, before the step of inputting the current crystal pulling data into a preset data driven model, the method further includes:

[0011] Obtain multiple actual crystal pulling data and corresponding actual crystal diameters in actual scenes;

[0012] Constructing training data according to each of the actual crystal pulling data and each of the crystal diameters;

[0013] The initial BP neural network model is trained using the training data to obtain a preset data-driven model.

[0014] In one embodiment, the step of training the initial BP neural network model using the training data to obtain a preset data-driven model includes:

[0015] Inputting the training data into the initial BP neural network model;

[0016] Receiving a response value fed back by the initial BP neural network model based on the training data;

[0017] Determine whether the error between the response value and the expected value in the training data reaches a preset error calibration value;

[0018] If yes, then the weights and thresholds between the neurons in the initial BP neural network model are adjusted based on the error, and the step of inputting the training data into the initial BP neural network model is returned until the error is lower than the preset error calibration value;

[0019] The adjusted BP neural network model is used as the preset data-driven model.

[0020] In one embodiment, if yes, then adjusting the weights and thresholds between neurons in the initial BP neural network model based on the error, and returning to the step of inputting the training data into the initial BP neural network model until the error is lower than the preset error calibration value, comprises:

[0021] If yes, then determine the weights and thresholds between neurons in the initial BP neural network model;

[0022] Determine the gradient information of each weight and each threshold value based on the error;

[0023] The weights and the thresholds are adjusted according to the gradient information, and the step of inputting the training data into the initial BP neural network model is returned until the error is lower than the preset error calibration value.

[0024] In one embodiment, before the step of training the initial BP neural network model using the training data to obtain a preset data-driven model, the step further includes:

[0025] Determine the initial weights of the initial BP neural network model;

[0026] Generate an initial particle swarm according to each of the initial weights and each of the initial thresholds;

[0027] Iterating each of the initial weights through the initial particle swarm to obtain an optimal weight;

[0028] Each of the initial weights is replaced by the optimal weight.

[0029] In one embodiment, the step of iterating each of the initial weights through the initial particle swarm to obtain the optimal weight includes:

[0030] Substituting the initial particle swarm into the preset data-driven model to obtain a fitness value;

[0031] Determining whether the fitness value meets a preset termination condition;

[0032] If not, updating the moving speed and position of the initial particle swarm in each of the initial weights based on the fitness value, and returning to the step of substituting the initial particle swarm into the preset data-driven model to obtain the fitness value, until the updated particle swarm meets the preset termination condition;

[0033] A target particle that meets the preset termination condition is obtained, and a weight corresponding to the target particle is used as an optimal weight.

[0034] In addition, to achieve the above-mentioned purpose, the present application also proposes a crystal diameter prediction device, the device comprising:

[0035] A data acquisition module is used to obtain current crystal pulling data;

[0036] A model prediction module, used for inputting the current crystal pulling data into a preset data-driven model, wherein the preset data-driven model is obtained by training a plurality of actual crystal pulling data and corresponding actual crystal diameters in actual scenes, and is used for characterizing the corresponding relationship between the crystal pulling data and the crystal diameter;

[0037] The data receiving module is used to receive the crystal diameter prediction value fed back by the data-driven model based on the current crystal pulling data.

[0038] In addition, to achieve the above-mentioned purpose, the present application also proposes a crystal diameter prediction device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the crystal diameter prediction method as described above.

[0039] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the crystal diameter prediction method described above are implemented.

[0040] In addition, to achieve the above-mentioned purpose, the present application also proposes a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of the crystal diameter prediction method described above are implemented.

[0041] One or more technical solutions proposed in this application have at least the following technical effects:

[0042] The present application obtains the current crystal pulling data; inputs the current crystal pulling data into a preset data-driven model, the preset data-driven model is obtained by training multiple actual crystal pulling data and corresponding actual crystal diameters in actual scenes, and is used to characterize the correspondence between the crystal pulling data and the crystal diameter; and receives the crystal diameter prediction value fed back by the data-driven model based on the current crystal pulling data. Since the preset data-driven model of the present application is obtained by training the actual crystal pulling data and the actual crystal diameter in actual scenes, compared with the prior art, the preset data-driven model of the present application can learn the laws and patterns of single crystal silicon growth in actual scenes, and can more accurately reflect the laws and patterns of single crystal silicon growth in actual scenes, and then the crystal diameter prediction value obtained based on the preset data-driven model is more accurate, which can provide an effective strategy for the design of the control algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0044] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 This is a schematic diagram of the flow chart of the first embodiment of the crystal diameter prediction method of the present application;

[0046] Figure 2 This is a schematic diagram of the flow chart of the second embodiment of the crystal diameter prediction method of the present application;

[0047] Figure 3 This is a schematic diagram of modeling of a preset data-driven model in the second embodiment of the crystal diameter prediction method of the present application;

[0048] Figure 4 This is a schematic diagram of the flow chart of the third embodiment of the crystal diameter prediction method of the present application;

[0049] Figure 5This is a schematic diagram of a neural network training process combined with a weight optimization algorithm in the third embodiment of the crystal diameter prediction method of the present application;

[0050] Figure 6 This is a schematic diagram of the module structure of the crystal diameter prediction device of the present application;

[0051] Figure 7 Schematic diagram of the device structure of the hardware operating environment involved in the crystal diameter prediction method in the embodiment of the present application.

[0052] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0053] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0054] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0055] The main solution of the embodiment of the present application is: the crystal diameter prediction device obtains the current crystal pulling data; the current crystal pulling data is input into a preset data-driven model, and the preset data-driven model is obtained by training multiple actual crystal pulling data and corresponding actual crystal diameters in actual scenes, and is used to characterize the correspondence between the crystal pulling data and the crystal diameter; and the crystal diameter prediction value fed back by the data-driven model based on the current crystal pulling data is received.

[0056] In the embodiments of the present application, for ease of description, the following description is made using a crystal diameter prediction device as the execution state.

[0057] At present, the existing method is to establish a mathematical mechanism model based on the growth mechanism of single crystal silicon. Specifically, according to the energy transfer and energy conservation at the solid-liquid interface, the relationship between the temperature gradient and the crystal growth rate and crystal diameter can be determined, so as to establish a model of thermal field temperature and crystal diameter; the crystal pulling speed determines the axial growth rate of the crystal, and the pulling speed also has a great influence on the crystal diameter. Establishing a model between the pulling speed and the crystal diameter is also a method.

[0058] However, the above mathematical mechanism model has many ideal assumptions in the process of establishing it, which makes the mathematical mechanism model very different from the laws and patterns of single crystal silicon growth in actual scenarios, resulting in poor accuracy of crystal diameter predicted by the model, and unable to provide effective strategies for the design of control algorithms. In addition, the above method is also difficult to model complex coupled process industries.

[0059] The present application provides a solution, which obtains a preset data-driven model by training the actual crystal pulling data and the actual crystal diameter in the actual scene, so that the preset driving model can learn the laws and patterns of single crystal silicon growth in the actual scene, and can more accurately reflect the laws and patterns of single crystal silicon growth in the actual scene. Then, the crystal diameter prediction value obtained based on the preset data-driven model has high accuracy, which can provide an effective strategy for the design of the control algorithm, and for complex coupled industries, by inputting corresponding data to train and optimize the preset data model, the optimized preset data model can be adapted to the complex coupled process industry.

[0060] It should be noted that the execution subject of this embodiment can be a computing service device with crystal diameter prediction, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a crystal diameter prediction device, etc. The following takes the crystal diameter prediction device as an example to illustrate this embodiment and the following embodiments.

[0061] Based on this, the present application embodiment provides a method for predicting crystal diameter, referring to Figure 1 , Figure 1 This is a schematic flow chart of the first embodiment of the crystal diameter prediction method of the present application.

[0062] In this embodiment, the crystal diameter prediction method includes steps S10 to S30:

[0063] Step S10, obtaining current crystal pulling data.

[0064] It should be noted that the current crystal pulling data may be the crystal pulling data for predicting the crystal diameter currently required. The crystal pulling data may be the data generated during the growth of single crystal silicon by using the Czochralski method to prepare single crystal silicon, and the crystal pulling data may at least include heating power, crystal pulling speed, and crucible lifting speed.

[0065] The heating power may be the power for heating the silicon melt. The crystal pulling speed may be the speed at which the seed crystal is lifted out of the silicon melt during the growth process. The crucible lifting speed may be the speed at which the crucible moves relative to the seed crystal during the crystal growth process. The heating power is used to maintain a constant temperature during the crystal growth process to ensure uniform growth of the crystal. The crystal pulling speed affects the diameter of the crystal growth and the quality of the crystal. The crucible lifting speed may be matched with the crystal pulling speed to maintain the stability of the solid-liquid interface and control the uniformity of the crystal diameter.

[0066] Step S20, inputting the current crystal pulling data into a preset data driven model, wherein the preset data driven model is obtained by training a plurality of actual crystal pulling data and corresponding actual crystal diameters in actual scenes, and is used to characterize the correspondence between the crystal pulling data and the crystal diameter.

[0067] Step S30, receiving the crystal diameter prediction value fed back by the preset data-driven model based on the current crystal pulling data.

[0068] It should be noted that the actual crystal pulling data may be the crystal pulling data generated during the actual growth of single crystal silicon. The actual crystal diameter may be the crystal diameter obtained under each actual crystal pulling data. The crystal diameter is the diameter of the single crystal silicon rod grown during the growth of single crystal silicon.

[0069] In a specific implementation, a large amount of actual crystal pulling data and the corresponding actual crystal diameter in the actual scene can be collected to construct training data for model training, and a preset data-driven model that characterizes the corresponding relationship between the crystal pulling data and the crystal diameter can be obtained. Since the preset data-driven model is trained with the actual crystal pulling data and the actual crystal diameter, the preset data-driven model can accurately reflect the law and pattern of single crystal silicon growth in the actual scene. After obtaining the preset data-driven model, the above-mentioned crystal diameter prediction device can input the acquired current crystal pulling data into the preset data-driven model, and the preset data-driven model makes a prediction based on the current crystal pulling data, and feeds back the crystal diameter prediction value corresponding to the current crystal pulling data, thereby completing the prediction of the crystal diameter.

[0070] The present embodiment provides a method for predicting crystal diameter, which includes obtaining current crystal pulling data; inputting the current crystal pulling data into a preset data-driven model, which is obtained by training multiple actual crystal pulling data and corresponding actual crystal diameters in actual scenes, and is used to characterize the correspondence between crystal pulling data and crystal diameter; and receiving a crystal diameter prediction value fed back by the data-driven model based on the current crystal pulling data. Since the preset data-driven model of the present embodiment is obtained by training the actual crystal pulling data and actual crystal diameter in actual scenes, compared with the prior art, the preset data-driven model of the present embodiment can learn the law and pattern of single crystal silicon growth in actual scenes, and can more accurately reflect the law and pattern of single crystal silicon growth in actual scenes, and then the crystal diameter prediction value obtained based on the preset data-driven model is more accurate, which can provide an effective strategy for the design of the control algorithm.

[0071] Based on the first embodiment of the present application, the second embodiment of the present application is proposed. In the second embodiment of the present application, the same or similar contents as those of the first embodiment can be referred to the above introduction, and will not be repeated later. Figure 2 , Figure 2 This is a flow chart of the second embodiment of the crystal diameter prediction method of the present application.

[0072] In this embodiment, before step S20, steps S01 to S03 are also included:

[0073] Step S01, obtaining a plurality of actual crystal pulling data and corresponding actual crystal diameters in an actual scene.

[0074] Step S02, constructing training data according to each of the actual crystal pulling data and each of the crystal diameters.

[0075] Step S03, training the initial BP neural network model using the training data to obtain a preset data-driven model.

[0076] In a specific implementation, a large amount of actual crystal pulling data and the corresponding actual crystal diameter in the actual scene can be collected, and the actual crystal pulling data is used as input data, and the actual crystal diameter is used as output data to construct training data including a training set and a test set, and the training data is used to iteratively train the initial backpropagation (BP) neural network model. The main structure of the initial BP neural network model is divided into an input layer, a hidden layer, and an output layer. The input layer receives the training data, and propagates it to the output layer through each hidden layer, extracts and learns the system dynamic behavior characteristics of the diameter control during the crystal pulling process, and obtains a preset data-driven model that characterizes the corresponding relationship between the crystal pulling data and the crystal diameter.

[0077] In a feasible implementation, step S03 may include steps S031 to S035:

[0078] Step S031, inputting the training data into the initial BP neural network model.

[0079] Step S032, receiving the response value fed back by the initial BP neural network model based on the training data.

[0080] In a specific implementation, the neurons in each hidden layer in the above initial BP neural network model can perform weighted summation on the training data of the input layer, and then perform nonlinear transformation through an activation function, and feedback a response value through the output layer, which can be the crystal diameter obtained based on the training data.

[0081] Step S033, determining whether the error between the response value and the expected value in the training data reaches a preset error calibration value.

[0082] It should be noted that the preset error calibration value may be a threshold for determining whether the error between the response value and the expected value is large. The expected value may be the actual crystal diameter in the training data.

[0083] In a specific implementation, the crystal diameter prediction device can calculate the error between the response value output by the initial BP neural network model and the expected value, and can quantify the error by using an average error or other loss function. Then, the error can be compared with a preset error calibration value to determine whether the error reaches the preset error calibration value.

[0084] In addition, the above process of determining whether the error between the response value and the expected value in the training data reaches a preset error calibration value may be performed in the output layer of the initial BP neural network, which is not limited in this embodiment.

[0085] Step S034, if yes, then adjust the weights and thresholds between the neurons in the initial BP neural network model based on the error, and return to the step of inputting the training data into the initial BP neural network model until the error is lower than the preset error calibration value.

[0086] Step S035, using the adjusted BP neural network model as a preset data driven model.

[0087] In a specific implementation, the crystal diameter prediction device can determine that the error is large when it detects that the error between the response value and the expected value reaches a preset error calibration value. At this time, the back propagation stage can be entered to back propagate the error from the output layer of the BP neural network model back to the input layer. In this stage, the weights and thresholds between the neurons in the initial BP neural network model can be continuously adjusted based on the error. After one adjustment, the step of inputting the training data into the initial BP neural network model is repeated until the error is lower than the preset error calibration value after adjustment, and then the adjusted BP neural network model is used as the preset data-driven model for the final training.

[0088] The weights between the neurons may be weights that characterize the connection strength between neurons, and are used to determine the degree of influence of the output of the previous neuron on the input of the current neuron. The thresholds between the neurons may be bias values ​​of the neurons, and are used to adjust the activation thresholds of the neurons. The weights and thresholds jointly determine the output of the neurons.

[0089] Reference Figure 3 , Figure 3 This is a schematic diagram of modeling of a preset data-driven model in the second embodiment of the crystal diameter prediction method of the present application. Figure 3In the method, a large amount of actual crystal pulling data and the corresponding actual crystal diameter are obtained, and the obtained data are classified to obtain training data and verification data, and then the classified data are analyzed and processed, such as preprocessing, normalization, etc., and then the input and output of the neural network (i.e., the initial BP neural network model) are determined from the processed data, and the structure of the neural network is determined. The training data is input into the initial BP neural network model for neural network training, and the output of the trained BP neural network model is verified using the verification data, and then it is determined whether the requirements are met (i.e., whether the error is lower than the preset error calibration value). If not, the training process is repeated until the requirements are met and the training is completed, and the trained neural network is used as the preset data-driven model, and the input information (such as the current crystal pulling data) is input into the trained neural network to obtain the network output (i.e., the crystal diameter prediction value is obtained).

[0090] In a feasible implementation, step S034 may include steps S0341 to S0343:

[0091] Step S0341: If yes, determine the weights and thresholds between neurons in the initial BP neural network model.

[0092] Step S0342: Determine the gradient information of each weight and each threshold based on the error.

[0093] In a specific implementation, the crystal diameter prediction device can traverse the weights and thresholds between neurons in the initial BP neural network model, and then calculate the partial derivatives of the error with respect to each weight and threshold during the error back propagation process, and each partial derivative constitutes gradient information.

[0094] Step S0343, adjusting each of the weights and each of the thresholds using the gradient information, and returning to the step of inputting the training data into the initial BP neural network model until the error is lower than the preset error calibration value.

[0095] In a specific implementation, the crystal diameter prediction device can use a gradient descent algorithm to adjust each weight and each threshold in the opposite direction of the gradient (i.e., the opposite direction of the error reduction) according to the calculated gradient information, thereby reducing the error. After adjusting once, the step of inputting the training data into the initial BP neural network model is repeated until the error is lower than the preset error calibration value after adjustment, and then the adjusted BP neural network model is used as the preset data-driven model that has been finally trained. The error between the response value and the expected value can be effectively reduced by the above-mentioned gradient descent method, and the accuracy of the preset data-driven model is effectively improved.

[0096] Based on the second embodiment of the present application, the third embodiment of the present application is proposed. In the third embodiment of the present application, the same or similar contents as those of the first embodiment and the second embodiment can be referred to the above introduction, and will not be repeated in the following. Figure 4 , Figure 4 This is a schematic flow chart of the third embodiment of the crystal diameter prediction method of the present application.

[0097] In this embodiment, before step S03, steps S31 to S34 are also included:

[0098] Step S31, determining initial weights of the initial BP neural network model.

[0099] It should be noted that the above initial weights may be weights in the initial BP neural network model that have not been adjusted or initialized.

[0100] It is understandable that the above-mentioned preset data-driven model is obtained based on neural network training, and neural networks have shortcomings in practical applications such as long operation response time, low learning efficiency, and easy to produce local optimal solutions. When applied to crystal diameter prediction, its prediction effect is not ideal and the deviation is large. Moreover, when the initial weight setting is unreasonable, it will affect the adjustment direction of the weight, thereby affecting the optimization accuracy. Therefore, this embodiment is proposed to solve the above problems.

[0101] In a specific implementation, the crystal diameter prediction device can determine the structure of the preset data-driven model, including the number of layers, the number of neurons in each layer, etc. The weight of each neuron is used as the initial weight to be optimized.

[0102] Step S32, generating an initial particle swarm according to each of the initial weights and each of the initial thresholds.

[0103] Step S33, iterating each of the initial weights through the initial particle swarm to obtain an optimal weight.

[0104] In a specific implementation, the crystal diameter prediction device can randomly generate a group of particles, each particle representing any possible initial weight. The error between the response value of the preset data-driven model and the expected value can be used as the fitness value, and then each particle in the initial particle group is substituted into the preset data-driven model to obtain the corresponding fitness value, and each particle in the initial particle group is traversed to determine the global optimal fitness value, obtain the target particle corresponding to the global optimal fitness value, and substitute the weight corresponding to the target particle into the preset data-driven model as the optimal weight to complete the optimization of the initial weight of the preset data-driven model.

[0105] In a feasible implementation, step S33 may include steps S331 to S334:

[0106] Step S331, substituting the initial particle swarm into the preset data-driven model to obtain a fitness value.

[0107] In a specific implementation, the above-mentioned crystal diameter prediction device can substitute each particle in the initial particle group into the preset data-driven model in turn to obtain the fitness value of each particle. Compare the current fitness value of each particle with the previously recorded individual optimal fitness value. If the current fitness value is better, update the individual optimal fitness value, compare the individual optimal fitness values ​​corresponding to each particle, and use the best value among the individual optimal fitness values ​​as the global optimal fitness value, and all particles share the global optimal fitness value. Among them, when any two fitness values ​​are compared, if the previous fitness value is greater than the latter fitness value, it is determined that the previous fitness value is better than the latter fitness value.

[0108] Step S332, determining whether the fitness value satisfies a preset termination condition.

[0109] It should be noted that the above-mentioned preset termination condition may be a condition for stopping adjusting the initial weights, such as reaching a maximum number of iterations or the fitness value no longer significantly improving.

[0110] Step S333, if not, then update the moving speed and position of the initial particle swarm in each of the initial weights based on the fitness value, and return to the step of substituting the initial particle swarm into the preset data-driven model to obtain the fitness value until the updated particle swarm meets the preset termination condition.

[0111] It should be noted that the moving speed of the above-mentioned particle can be the moving direction and step length in the search space, and the speed can guide the particle in which direction (better fitness value) to move and how far to move. The search space can be a space composed of initial weights, and each position in the search space represents each initial weight. The position of the above-mentioned particle is the position in the search space.

[0112] In a specific implementation, the crystal diameter prediction device can determine whether the obtained global optimal fitness value meets the preset termination condition. If the preset termination condition is not met, the moving speed and position of the particle can be adjusted in the search space, the particle can be moved to a position corresponding to another weight value, and then the above process of substituting the initial particle group into the preset data-driven model to obtain the fitness value is repeated. After reaching the global optimal fitness value of the particle at the current position meets the above preset condition, the iteration is stopped.

[0113] Step S334, obtaining a target particle that meets the preset termination condition, and taking the weight corresponding to the target particle as the optimal weight.

[0114] Step S34: replacing each of the initial weights with the optimal weight.

[0115] In a specific implementation, the above-mentioned crystal diameter prediction device can extract the global optimal fitness value that meets the preset termination condition, and use all particles that obtain the global optimal fitness value as the target particles, and then use the weights corresponding to the target particles as the optimal weights to replace the initial weights in the initial BP neural network model to complete the optimization of the initial weights.

[0116] For ease of understanding, refer to Figure 5 This is for illustration only, but does not limit this solution. Figure 5 This is a schematic diagram of the neural network training process combined with the weight optimization algorithm in the third embodiment of the crystal diameter prediction method of the present application. Figure 5 In the process, the topological structure of the BP neural network (i.e., the preset data-driven model) is first clarified, each initial weight is determined, and then the initial weight of the BP data is initialized. In the initialization process, the initial particle group is first randomly generated, and then the speed and position of each particle in the initial particle group are initialized and then inserted into the BP neural network. The error obtained by the BP neural network training is used as the fitness value, and then the speed and position of the particle are updated based on the fitness value. After calculating the fitness value of the updated particle, the individual extreme value (i.e., the individual optimal fitness value) and the global extreme value (i.e., the global optimal fitness value) are updated, and then the global extreme value algorithm is judged to meet the termination condition (i.e., the above-mentioned preset termination condition). If not, the step of updating the speed and position of the particle is returned, and the above process is repeated for iteration. If so, the weight corresponding to the particle that meets the termination condition is taken as the optimal weight, and the optimal weight is substituted into the BP neural network for training, and then the training error is calculated, and the training error algorithm is judged to meet the termination condition (i.e., the above-mentioned preset error calibration value). If not, the weight (i.e., weight) is updated and then the step of BP neural network training is returned. The above process is repeated for iteration until the training error meets the termination condition and the training is completed.

[0117] This embodiment determines the initial weights of the initial BP neural network model, generates an initial particle group according to the initial weights and the initial thresholds, substitutes the initial particle group into the preset data-driven model to obtain a fitness value; determines whether the fitness value meets the preset termination condition; if not, updates the moving speed and position of the initial particle group in each initial weight based on the fitness value, and returns to the step of substituting the initial particle group into the preset data-driven model to obtain the fitness value until the updated particle group meets the preset termination condition; obtains the target particle that meets the preset termination condition, and uses the weight corresponding to the target particle as the optimal weight. This embodiment completes the optimization of the fitness value through the initial particle group, and can determine the optimal weight of the global optimal fitness value, so that the initial weight in the initial BP neural network can be accurately optimized, and the initial weight can be made more reasonable and more suitable for application in the single crystal silicon growth process, thereby effectively improving the prediction effect and reducing the prediction deviation.

[0118] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the crystal diameter prediction method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0119] This application also provides a crystal diameter prediction device, please refer to Figure 6 , Figure 6 This is a schematic diagram of the module structure of the crystal diameter prediction device of the present application.

[0120] like Figure 6 As shown, the crystal diameter prediction device comprises:

[0121] The data acquisition module 10 is used to acquire current crystal pulling data.

[0122] The model prediction module 20 is used to input the current crystal pulling data into a preset data-driven model. The preset data-driven model is obtained by training multiple actual crystal pulling data and corresponding actual crystal diameters in actual scenes, and is used to characterize the correspondence between the crystal pulling data and the crystal diameter.

[0123] The data receiving module 30 is used to receive the crystal diameter prediction value fed back by the data-driven model based on the current crystal pulling data.

[0124] The crystal diameter prediction device provided by the present application adopts the crystal diameter prediction method in the above-mentioned embodiment, which can solve the technical problem that there are large differences between the laws and patterns of single crystal silicon growth in the prior art and in actual scenarios, resulting in poor accuracy of the crystal diameter predicted by the model and failure to provide an effective strategy for the design of the control algorithm. Compared with the prior art, the beneficial effects of the crystal diameter prediction device provided by the present application are the same as the beneficial effects of the crystal diameter prediction method provided by the above-mentioned embodiment, and the other technical features in the crystal diameter prediction device are the same as the features disclosed in the above-mentioned embodiment method, which will not be repeated here.

[0125] The present application provides a crystal diameter prediction device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the crystal diameter prediction method in the above-mentioned embodiment one.

[0126] Reference below Figure 7 , Figure 7 The device structure diagram of the hardware operating environment involved in the crystal diameter prediction method in the embodiment of the present application shows a schematic diagram of the structure of the crystal diameter prediction device suitable for implementing the embodiment of the present application. The crystal diameter prediction device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 7 The crystal diameter prediction device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0127] like Figure 7As shown, the crystal diameter prediction device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the crystal diameter prediction device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the crystal diameter prediction device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows a crystal diameter prediction device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or have alternatively.

[0128] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0129] The crystal diameter prediction device provided by the present application adopts the crystal diameter prediction method in the above-mentioned embodiment, which can solve the technical problem that there are large differences between the laws and patterns of single crystal silicon growth in the prior art and in actual scenarios, resulting in poor accuracy of the crystal diameter predicted by the model and failure to provide an effective strategy for the design of the control algorithm. Compared with the prior art, the beneficial effects of the crystal diameter prediction device provided by the present application are the same as the beneficial effects of the crystal diameter prediction method provided by the above-mentioned embodiment, and the other technical features in the crystal diameter prediction device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0130] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0131] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0132] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the crystal diameter prediction method in the above-mentioned embodiment.

[0133] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0134] The computer-readable storage medium may be included in the crystal diameter prediction device; or may exist independently without being assembled into the crystal diameter prediction device.

[0135] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the crystal diameter prediction device, the crystal diameter prediction device: obtains current crystal pulling data; inputs the current crystal pulling data into a preset data-driven model, wherein the preset data-driven model is obtained by training multiple actual crystal pulling data and corresponding actual crystal diameters in actual scenarios, and is used to characterize the correspondence between crystal pulling data and crystal diameter; and receives the crystal diameter prediction value fed back by the preset data-driven model based on the current crystal pulling data.

[0136] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0137] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0138] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0139] The readable storage medium provided in the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned crystal diameter prediction method, and can solve the technical problem that there are large differences between the laws and patterns of single crystal silicon growth in the prior art and in actual scenarios, resulting in poor accuracy of the crystal diameter predicted by the model, and unable to provide an effective strategy for the design of the control algorithm. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in the present application are the same as the beneficial effects of the crystal diameter prediction method provided in the above-mentioned embodiment, which will not be repeated here.

[0140] The present application also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the steps of the crystal diameter prediction method as described above are implemented.

[0141] The computer program product provided by the present application can solve the technical problem that there are large differences between the laws and patterns of single crystal silicon growth in the prior art and in actual scenarios, resulting in poor accuracy of the crystal diameter predicted by the model and failure to provide an effective strategy for the design of the control algorithm. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the crystal diameter prediction method provided in the above embodiment, which will not be described in detail here.

[0142] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A method for predicting crystal diameter, characterized in that: The method comprises: Get the current crystal pulling data; Inputting the current crystal pulling data into a preset data driving model, wherein the preset data driving model is obtained by training a plurality of actual crystal pulling data and corresponding actual crystal diameters in actual scenes, and is used to characterize the corresponding relationship between the crystal pulling data and the crystal diameter; A crystal diameter prediction value fed back by the preset data-driven model based on the current crystal pulling data is received.

2. The crystal diameter prediction method according to claim 1, characterized in that: Before the step of inputting the current crystal pulling data into the preset data driven model, the method further includes: Obtain multiple actual crystal pulling data and corresponding actual crystal diameters in actual scenes; Constructing training data according to each of the actual crystal pulling data and each of the crystal diameters; The initial BP neural network model is trained using the training data to obtain a preset data-driven model.

3. The crystal diameter prediction method according to claim 2, characterized in that: The step of training the initial BP neural network model by using the training data to obtain a preset data-driven model includes: Inputting the training data into the initial BP neural network model; Receiving a response value fed back by the initial BP neural network model based on the training data; Determine whether the error between the response value and the expected value in the training data reaches a preset error calibration value; If yes, then the weights and thresholds between the neurons in the initial BP neural network model are adjusted based on the error, and the step of inputting the training data into the initial BP neural network model is returned until the error is lower than the preset error calibration value; The adjusted BP neural network model is used as the preset data-driven model.

4. The crystal diameter prediction method according to claim 3, characterized in that: If yes, then based on the error, the weights and thresholds between the neurons in the initial BP neural network model are adjusted, and returning to the step of inputting the training data into the initial BP neural network model until the error is lower than the preset error calibration value, including: If yes, then determine the weights and thresholds between neurons in the initial BP neural network model; Determine the gradient information of each weight and each threshold value based on the error; The weights and the thresholds are adjusted according to the gradient information, and the step of inputting the training data into the initial BP neural network model is returned until the error is lower than the preset error calibration value.

5. The crystal diameter prediction method according to claim 2, characterized in that: Before the step of training the initial BP neural network model by the training data to obtain the preset data driven model, the method further includes: Determine the initial weights of the initial BP neural network model; Generate an initial particle swarm according to each of the initial weights and each of the initial thresholds; Iterating each of the initial weights through the initial particle swarm to obtain an optimal weight; Each of the initial weights is replaced by the optimal weight.

6. The crystal diameter prediction method according to claim 5, characterized in that: The step of iterating the initial weights through the initial particle swarm to obtain the optimal weights comprises: Substituting the initial particle swarm into the preset data-driven model to obtain a fitness value; Determining whether the fitness value meets a preset termination condition; If not, updating the moving speed and position of the initial particle swarm in each of the initial weights based on the fitness value, and returning to the step of substituting the initial particle swarm into the preset data-driven model to obtain the fitness value, until the updated particle swarm meets the preset termination condition; A target particle that meets the preset termination condition is obtained, and a weight corresponding to the target particle is used as an optimal weight.

7. A crystal diameter prediction device, characterized in that: The device comprises: A data acquisition module is used to obtain current crystal pulling data; A model prediction module, used for inputting the current crystal pulling data into a preset data-driven model, wherein the preset data-driven model is obtained by training a plurality of actual crystal pulling data and corresponding actual crystal diameters in actual scenes, and is used for characterizing the corresponding relationship between the crystal pulling data and the crystal diameter; The data receiving module is used to receive the crystal diameter prediction value fed back by the data-driven model based on the current crystal pulling data.

8. A crystal diameter prediction device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the crystal diameter prediction method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the crystal diameter prediction method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the crystal diameter prediction method according to any one of claims 1 to 6 are implemented.

Images