Trichlorosilane reduction control method
The neural network model is used to adjust the temperature and feed parameters in the polysilicon production process in real time, which solves the problem of uneven deposition efficiency in traditional methods, and improves production stability and raw material utilization.
Patent Information
- Application Number
- CN202411212687.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-30
- Publication Date
- 2024-12-06
AI Technical Summary
In the traditional polysilicon production process, the regulation of current and feed depends on manual experience or preset curves, and cannot be adjusted in real time according to the state of the reduction furnace, resulting in uneven deposition efficiency and affecting product quality and equipment life.
A trichlorosilicon reduction control method based on neural network model is constructed. By collecting and processing time series data, a feature extractor is used to extract features, establish a prediction model, adjust the temperature and feed parameters in real time, and optimize the deposition efficiency.
The optimal deposition efficiency under different production conditions is achieved, production fluctuations are reduced, and production stability and raw material utilization are improved.
Smart Images

Figure CN119080005A_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the field of polysilicon manufacturing, and in particular relates to a trichlorosilane reduction control method. Background Art
[0002] Polysilicon is an indispensable basic material in the modern semiconductor and photovoltaic industries. The most commonly used method in its production is the Siemens process, the core reaction of which involves trichlorosilane. Chemical reduction with hydrogen at high temperature produces high-purity polysilicon. The reduction furnace is one of the main equipment for this reaction, and the adjustment of the current has a crucial impact on the reaction efficiency, product quality and energy consumption.
[0003] During the chemical reduction reaction, as silicon is deposited on the surface of the silicon rod, the resistance of the silicon rod will change. In order to maintain the stability of the temperature in the reduction furnace, the current of the silicon rod must also be adjusted accordingly. If the current is too high, the temperature in the reduction furnace will be too high, which will accelerate the aging of the equipment, increase the risk of failure, and may also cause silicon to deposit too quickly, affecting the deposition quality. If the current is too low, the reaction rate of trichlorosilane and hydrogen is low, which will cause the silicon deposition rate to decrease and the crystallization to be uneven. In addition, the feed of hydrogen and trichlorosilane is also one of the key factors affecting the system conversion efficiency.
[0004] Traditional control methods mostly rely on manual experience, manually adjusting the current and feed by observing the production of silicon rods through an observation window, or using a control program to control according to a pre-set current curve or feed curve. The former requires manual real-time monitoring of the growth of silicon rods, which has the disadvantage of low efficiency, while the latter can only be adjusted according to a preset curve and cannot be adjusted in real time according to the reduction furnace status. Summary of the invention
[0005] In order to overcome the defects of the above-mentioned prior art, the present application constructs a neural network model, takes deposition efficiency as an evaluation index, and adjusts temperature, hydrogen, and trichlorosilane feed online to maximize the utilization rate of raw materials.
[0006] The purpose of this application is to provide a trichlorosilane reduction control method, comprising:
[0007] Collect time series data of various parameters in the process of producing polysilicon;
[0008] Use feature extractor to extract features from time series data;
[0009] Establish and train prediction models based on the extracted features;
[0010] The trained prediction model is used to control the temperature, hydrogen feed, and trichlorosilane feed of the polysilicon production process and adjust the deposition efficiency.
[0011] Furthermore, the time series data of each parameter includes time series data of temperature, hydrogen feed and trichlorosilane feed.
[0012] Furthermore, the feature extractor includes a convolutional layer, a recursive layer, a fully connected layer, a deposition efficiency prediction layer and an output layer connected in sequence. The input expression of the feature extractor is:
[0013]
[0014] Where T(t) is the time series data of temperature, is the time series data of hydrogen feed, This is the time series data of trichlorosilane feed.
[0015] Furthermore, the expression of the convolutional layer is:
[0016] X cnn (t) = ReLU(Conv1D(X(t)·W cnn +b cnn ))
[0017] Among them, W cnn is the convolution kernel, b cnn For bias.
[0018] Furthermore, the expression of the recursive layer is:
[0019] H LSTM (t) = LSTM(X cnn (t))
[0020] Furthermore, the expression of the fully connected layer is:
[0021] H fc (t) = ReLU(W fc ·H LSTM (t)+b fc )
[0022] Among them, W fc is the weight of the fully connected layer, b fc is the bias of the fully connected layer.
[0023] Furthermore, the expression of the deposition efficiency prediction layer is:
[0024] η(t)=W out ·H fc (t)+b out
[0025] Among them, W out is the weight of the output layer, b out is the bias of the output layer.
[0026] Furthermore, a prediction model is established and trained based on the extracted features, including:
[0027] The deposition efficiency and time series data obtained by feature extraction are segmented to obtain a training set and a validation set;
[0028] Normalize the data of the training set;
[0029] The normalized training set data is used to train the neural network model, and the validation set data is used to adjust the hyperparameters.
[0030] Furthermore, establishing and training a prediction model based on the extracted features also includes:
[0031] Monitor the loss function and evaluation metrics during training.
[0032] Furthermore, after the prediction model is established and trained based on the extracted features, the prediction results of the prediction model are evaluated for errors, and the mean square error between the predicted result of the deposition efficiency and the actual value is calculated.
[0033] Technical effects and advantages of this application:
[0034] The present application studies the factors that affect the efficiency of converting trichlorosilane to polysilicon, determines the main parameters that affect the polysilicon production process, uses the deposition efficiency of polysilicon as the evaluation standard for the trichlorosilane reduction reaction, constructs and trains a neural network model for predicting the deposition efficiency based on time series data of temperature, hydrogen and trichlorosilane feed, realizes real-time adjustment of temperature, hydrogen feed parameters and trichlorosilane feed parameters, quickly responds to changes in the production process, reduces production fluctuations caused by external factors, improves production stability, and ensures that the best deposition efficiency is always achieved under different production conditions.
[0035] Other features and advantages of the present application will be described in the following description, and partly become apparent from the description, or be understood by practicing the present application. The purpose and other advantages of the present application can be realized and obtained by the structures indicated in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Figure 1 This is a flow chart of a trichlorosilane reduction control method of the present application. DETAILED DESCRIPTION
[0037] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0038] like Figure 1 As shown, the embodiment of the present application provides a trichlorosilane reduction control method, comprising:
[0039] Collect time series data of various parameters in the process of producing polysilicon;
[0040] Using a feature extractor to extract features from the time series data;
[0041] Establish and train prediction models based on the extracted features;
[0042] The trained prediction model is used to control the temperature, hydrogen feed, and trichlorosilane feed of the polysilicon production process and adjust the deposition efficiency.
[0043] The specific steps are as follows:
[0044] 1. Data collection:
[0045] Collect time series data of various parameters in the polysilicon production process, including: temperature time series data T(t), hydrogen feed time series data Time series data of trichlorosilane feed
[0046] 2. Use a feature extractor to extract features from the time series data, wherein the feature extractor includes a convolutional layer, a recursive layer, a fully connected layer, a deposition efficiency prediction layer and an output layer connected in sequence.
[0047] 2.1. Input vector:
[0048]
[0049] 2.2. The feature extractor extracts features from the input vector:
[0050] Convolutional layer calculation:
[0051] X cnn (t) = ReLU(Conv1D(X(t)·W cnn +b cnn ))
[0052] Among them, W cnn is the convolution kernel, b cnn For bias.
[0053] Recursive layer calculation:
[0054] H LSTM (t) = LSTM(X cnn (t))
[0055] The fully connected layer uses the ReLU activation function for feature integration:
[0056] H fc (t) = ReLU(W fc ·H LSTM (t)+b fc )
[0057] Among them, W fc is the weight of the fully connected layer, b fc is the bias of the fully connected layer.
[0058] Deposition efficiency prediction Layer prediction:
[0059] η(t)=W out ·H fc (t)+b out
[0060] Among them, W out is the weight of the output layer, b out is the bias of the output layer.
[0061] The output layer outputs the prediction results.
[0062] 3. Establish and train a prediction model based on the extracted features:
[0063] 3.1. The collected time series data of deposition efficiency, temperature, hydrogen feed, and trichlorosilane feed are divided into a training set and a validation set.
[0064] 3.2. Normalize the data of the training set to improve the training effect of the neural network.
[0065] 3.3. Use the normalized training set to train the neural network model and adjust the hyperparameters using the validation set.
[0066] 3.4. Monitor the loss function and evaluation indicators during training.
[0067] 4. Evaluate the error of the prediction results of the prediction model:
[0068] The mean square error (MSE) between the predicted and actual deposition efficiency was calculated.
[0069]
[0070] 5. Use the trained model to optimize the time profiles of temperature, hydrogen feed, and trichlorosilane feed to obtain maximum deposition efficiency.
[0071] In summary, in view of the fact that traditional control methods rely on manual experience, require an observation window to observe the production status of silicon rods to manually adjust the current and feed, or adjust according to a preset curve, and cannot be adjusted in real time according to the state of the reduction furnace, this application studies the factors affecting the efficiency of trichlorosilicon conversion to polysilicon, determines the main parameters affecting the polysilicon production process, uses the deposition efficiency of polysilicon as the evaluation standard of trichlorosilicon reduction reaction, constructs and trains a neural network model that predicts the deposition efficiency based on time series data of temperature, hydrogen and trichlorosilicon feed, realizes real-time adjustment of temperature, hydrogen feed parameters and trichlorosilicon feed parameters, quickly responds to changes in the production process, reduces production fluctuations caused by external factors, improves production stability, and ensures that the best deposition efficiency is always achieved under different production conditions.
[0072] The present application also provides a trichlorosilane reduction control system, wherein the read / write system includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the above-mentioned read / write method embodiment are implemented, for example: Figure 1 Alternatively, the processor implements the functions of each unit in the above-mentioned reading and writing device embodiments when executing the computer program.
[0073] Exemplarily, the computer program can be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to complete the present application. The one or more modules / units can be a series of computer program instruction segments that can perform specific functions, and the instruction segments are used to describe the execution process of the computer program in the read-write system. The read-write system can be a computing device such as a desktop computer, a notebook, a handheld computer, and a cloud server. The read-write system may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that it may include other more or fewer components, or combine certain components, or different components. For example, the read-write system may also include input and output devices, network access devices, buses, etc.
[0074] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the read-write system, and uses various interfaces and lines to connect various parts of the entire read-write system.
[0075] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the read-write system by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.
[0076] If the module / unit integrated in the read-write system is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.
[0077] Finally, it should be noted that the above is only a preferred embodiment of the present application and is not intended to limit the present application. Although the present application has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.
Claims
1. A trichlorosilane reduction control method, characterized in that: include: Collect time series data of various parameters in the process of producing polysilicon; Using a feature extractor to extract features from the time series data; Establish and train prediction models based on the extracted features; The trained prediction model is used to control the temperature, hydrogen feed, and trichlorosilane feed of the polysilicon production process and adjust the deposition efficiency.
2. A trichlorosilane reduction control method according to claim 1, characterized in that: The time series data of each parameter includes time series data of temperature, hydrogen feed and trichlorosilane feed.
3. A trichlorosilane reduction control method according to claim 1 or 2, characterized in that: The feature extractor includes a convolutional layer, a recursive layer, a fully connected layer, a deposition efficiency prediction layer and an output layer connected in sequence. The input expression of the feature extractor is: Where T(t) is the time series data of temperature, is the time series data of hydrogen feed, This is the time series data of trichlorosilane feed.
4. A trichlorosilane reduction control method according to claim 3, characterized in that: The expression of the convolutional layer is: X cnn (t)=ReLU(Conv1D(X(t)·W cnn +b cnn )) Among them, W cnn is the convolution kernel, b cnn For bias.
5. A trichlorosilane reduction control method according to claim 4, characterized in that: The expression of the recursive layer is: H LSTM (t)=LSTM(X cnn (t))。 6. A trichlorosilane reduction control method according to claim 5, characterized in that: The expression of the fully connected layer is: H fc (t)=ReLU(W fc ·H LSTM (t)+b fc ) Among them, W fc is the weight of the fully connected layer, b fc is the bias of the fully connected layer.
7. A trichlorosilane reduction control method according to claim 6, characterized in that: The expression of the deposition efficiency prediction layer is: η(t)=W out ·H fc (t)+b out Among them, W out is the weight of the output layer, b out is the bias of the output layer.
8. The trichlorosilane reduction control method according to claim 1, characterized in that: The method of establishing and training a prediction model based on the extracted features includes: The deposition efficiency and time series data obtained by feature extraction are segmented to obtain a training set and a validation set; Normalize the data of the training set; The normalized training set data is used to train the neural network model, and the validation set data is used to adjust the hyperparameters.
9. A trichlorosilane reduction control method according to claim 8, characterized in that: The method of establishing and training a prediction model based on the extracted features further includes: Monitor the loss function and evaluation metrics during training.
10. The trichlorosilane reduction control method according to claim 1, characterized in that: After the prediction model is established and trained based on the extracted features, the prediction results of the prediction model are evaluated for errors, and the mean square error between the predicted results and the actual values of the deposition efficiency is calculated.
Citation Information
Cited By
DCS (Distributed Control System)-based reduction furnace control optimization method and related device
CN120972838A
