Monocrystalline silicon growth process control method, device and equipment and storage medium
By combining PID controller, BP neural network and particle swarm algorithm, the single crystal silicon growth process control method is solved, and the crystal pulling quality and control stability are improved.
Patent Information
- Application Number
- CN202411006471.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-25
- Publication Date
- 2024-11-15
AI Technical Summary
Traditional PID control is difficult to adapt to the complex single crystal pulling growth environment, resulting in poor control accuracy and seriously affecting the crystal pulling quality.
The single crystal silicon growth process control method combining PID controller, preset neural network and preset population optimization algorithm is adopted. The parameters of the PID controller are optimized through the BP neural network, and the particle swarm algorithm is adaptively adjusted to adapt to the single crystal silicon growth process. System changes.
It improves the stability and crystal pulling quality of the PID controller, achieves higher control accuracy and adaptability, reduces local optimal traps, and improves the parameter optimization speed.
Smart Images

Figure CN118957738A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of single crystal silicon technology, and in particular to a single crystal silicon growth process control method, device, equipment and storage medium. Background Art
[0002] Single crystal silicon is one of the most basic materials in the electronic information industry. According to the different growth methods of single crystal silicon in the single crystal furnace, its growth process is divided into CZ method (CZ), FZ method (FZ) and epitaxial method. With the rapid increase in the market demand for single crystal silicon wafers, the market demand for single crystal silicon is also showing a rapid growth trend.
[0003] Among them, the crystal pulling process control of single crystal silicon by the Czochralski method is generally achieved through PID control. However, since the environment in the furnace is constantly changing during the single crystal silicon pulling process and there is a large hysteresis, the stability of traditional PID control will be seriously affected and it is difficult to adapt to the complex crystal pulling growth environment, resulting in poor control accuracy and seriously affecting the crystal pulling quality. Summary of the invention
[0004] The main purpose of this application is to provide a method, device, equipment and storage medium for controlling the single crystal silicon growth process, aiming to solve the technical problem that traditional PID control is difficult to adapt to the complex single crystal silicon pulling growth environment, resulting in poor control accuracy and seriously affecting the pulling quality.
[0005] To achieve the above-mentioned purpose, the present application proposes a method for controlling a single crystal silicon growth process, which is applied to a single crystal silicon growth process control system. The system includes a PID controller, a preset neural network, and a preset population optimization algorithm. The method includes:
[0006] In the closed-loop control of the growth process of single crystal silicon, the system input, system output and system deviation of the single crystal silicon growth process control system are used as the neuron input of the preset neural network, the system input is the reference crystal diameter of the single crystal silicon, the preset neural network is used to optimize its own neuron output, and the preset population optimization algorithm is used to optimize the preset neural network;
[0007] Using the neuron output of the preset neural network as the PID input parameter of the PID controller;
[0008] The growth diameter of the silicon single crystal is controlled by the PID controller according to the PID input parameters, the growth diameter is used as the system output, and the difference between the growth diameter and the reference crystal diameter is used as the system deviation.
[0009] In one embodiment, the preset neural network is a BP neural network, the preset population optimization algorithm is a particle swarm algorithm, and the step of using the system input, system output and system deviation of the single crystal silicon growth process control system as neuron inputs of the preset neural network includes:
[0010] The system input, system output and system deviation of the single crystal silicon growth process control system are used as the neuron input of the BP neural network. The BP neural network optimizes its own neuron output based on the principle of minimizing the error performance function. The particle swarm algorithm optimizes the weight coefficient of the BP neural network.
[0011] In one embodiment, the topological structure of the BP neural network is three layers, including an input layer, a hidden layer and an output layer; the BP neural network optimizes its own neuron output based on the principle of minimizing the error performance function, including:
[0012] Taking the system input, the system output and the system deviation as inputs of the input layer;
[0013] The input of the input layer is calculated according to the weight coefficient and activation function of the hidden layer to obtain the input of the output layer, where the input of the output layer is the neuron output;
[0014] Taking minimization of the error performance function as the optimization principle, the weight coefficients are adjusted to optimize the neuron output of the BP neural network.
[0015] In one embodiment, the step of adjusting the weight coefficients to optimize the neuron output of the BP neural network based on minimization of the error performance function as the optimization principle includes:
[0016] Taking minimization of the error performance function as the optimization principle, the weight coefficients are adjusted by the gradient descent method to optimize the neuron output of the BP neural network.
[0017] In one embodiment, the particle swarm algorithm optimizes the weight coefficients of the BP neural network, including:
[0018] Initializing the weight coefficients of the BP neural network according to the topological structure to obtain initial weights;
[0019] Using the initial weight as an initialization particle, wherein the initialization particle includes a position and a velocity;
[0020] Obtaining a training error of the BP neural network;
[0021] According to the training error, the initialized particles are iterated by the particle swarm algorithm to obtain an optimal weight.
[0022] In one embodiment, the step of iterating the initialized particles by the particle swarm algorithm according to the training error to obtain the optimal weight includes:
[0023] Using the training error as an initial fitness value;
[0024] According to the initial fitness value, the initialized particles are iterated by the particle swarm algorithm to obtain individual extreme values and global extreme values corresponding to the initialized particles;
[0025] An optimal weight is determined according to the individual extreme value and the global extreme value.
[0026] In addition, to achieve the above-mentioned purpose, the present application also proposes a single crystal silicon growth process control device, which is applied to a single crystal silicon growth process control system, the system includes a PID controller, a preset neural network and a preset population optimization algorithm, and the device includes:
[0027] A parameter optimization module, used for using the system input, system output and system deviation of the single crystal silicon growth process control system as the neuron input of the preset neural network in the closed-loop control of the single crystal silicon growth process, the system input is the reference crystal diameter of the single crystal silicon, the preset neural network is used to optimize its own neuron output, and the preset population optimization algorithm is used to optimize the preset neural network;
[0028] A parameter output module, used for using the neuron output of the preset neural network as a PID input parameter of the PID controller;
[0029] A growth control module is used to control the growth diameter of the single crystal silicon through the PID controller according to the PID input parameters, the growth diameter is used as the system output, and the difference between the growth diameter and the reference crystal diameter is used as the system deviation.
[0030] In addition, to achieve the above-mentioned objectives, the present application also proposes a single crystal silicon growth process control device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the single crystal silicon growth process control method as described above.
[0031] 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 single crystal silicon growth process control method described above are implemented.
[0032] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the single crystal silicon growth process control method as described above.
[0033] One or more technical solutions proposed in this application have at least the following technical effects: the method of this application is applied to the control system of the growth process of single crystal silicon, the system includes a PID controller, a preset neural network and a preset group optimization algorithm, in the closed-loop control of the growth process of single crystal silicon, the system input, system output and system deviation of the control system of the growth process of single crystal silicon are used as the neuron input of the preset neural network, the system input is the reference crystal diameter of the single crystal silicon, the preset neural network is used to optimize its own neuron output, and the preset group optimization algorithm is used to optimize the preset neural network; the neuron output of the preset neural network is used as the PID input parameter of the PID controller; the growth diameter of the single crystal silicon is controlled by the PID controller according to the PID input parameter, the growth diameter is used as the system output, and the difference between the growth diameter and the reference crystal diameter is used as the system deviation. Because the present application combines the preset neural network and the preset group optimization algorithm, the parameters of the PID controller are adaptively adjusted to adapt to the changes in the parameters of the control system of the growth process of single crystal silicon, the stability of the PID controller is improved, and the quality of crystal pulling is finally improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] 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.
[0035] 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 labor.
[0036] Figure 1 It is a schematic diagram of a control system for a conventional Czochralski method single crystal silicon growth process;
[0037] Figure 2 This is a control block diagram of the traditional Czochralski method single crystal silicon growth process;
[0038] Figure 3 A schematic diagram of a process flow provided for the first embodiment of the method for controlling the growth process of single crystal silicon of the present application;
[0039] Figure 4A control block diagram of the BP neural network-particle swarm algorithm-PID controller provided in Example 1 during the growth process of single crystal silicon;
[0040] Figure 5 A schematic diagram of a process flow provided for the second embodiment of the method for controlling the growth process of single crystal silicon of the present application;
[0041] Figure 6 The BP neural network structure diagram provided for Example 2;
[0042] Figure 7 The optimization flow chart combining BP neural network and particle swarm algorithm provided in Example 2;
[0043] Figure 8 This is a schematic diagram of the module structure of the single crystal silicon growth process control device according to an embodiment of the present application;
[0044] Fig. 9 Schematic diagram of the equipment structure of the hardware operating environment involved in the single crystal silicon growth process control method in the embodiment of the present application.
[0045] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0046] 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.
[0047] 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.
[0048] It should be noted that the traditional Czochralski method single crystal silicon growth process control is composed of a control system and a single crystal furnace. Figure 1 , Figure 1 This is a schematic diagram of the control system for the traditional Czochralski method of single crystal silicon growth process. The control system's PID controls power, growth pulling speed, crucible lifting speed, pressure and other parameters to achieve closed-loop control of diameter, liquid surface temperature and liquid level.
[0049] Specifically, in the current Czochralski method, single crystal silicon growth process control is mainly based on CCD measurement of crystal diameter PID control. Figure 2 , Figure 2This is a control block diagram of the traditional CZ method for single crystal silicon growth. The control of the crystal diameter is affected by the heater power and the pulling speed, and its closed-loop control logic is as follows: the deviation between the CCD measured diameter and the diameter reference is input into the diameter PID controller, and the diameter is controlled by adjusting the pulling speed and the average pulling speed; the deviation between the pulling speed measurement and the pulling speed reference is input into the growth PID controller, and the pulling speed is controlled by adjusting the heater temperature and the temperature reference of the heater; the temperature PID controller compensates for thermal disturbances by adjusting the heater power.
[0050] However, there are two main technical problems faced in the control of the Czochralski method single crystal silicon growth process:
[0051] 1. Limitations of traditional PID control. The crystal pulling process is affected by many variables such as pulling speed, liquid surface temperature, crystal rise, crystal rotation, crucible rise, and crucible rotation. Traditional PID control is difficult to adapt to the complex crystal growth environment, resulting in poor control accuracy. At the same time, process parameters and control parameters rely on manual experience settings, and the parameter adjustment process is relatively cumbersome. The efficiency and accuracy of parameter adjustment will seriously affect the quality of crystal pulling, and may even cause major accidents such as wire breakage and rod drop. Manual experience adjustments are made based on the curve changes of the crystal pulling process parameters. If the parameters are inaccurate, repeated oscillations are likely to occur, which seriously affects the quality of crystal pulling.
[0052] 2. Applicability of PID control improvement strategy. At present, there have been technical breakthroughs in PID control of crystal pulling growth. On the basis of PID control, an improvement strategy is added to realize intelligent parameter adjustment. However, the accuracy and precision of the improvement strategy are not enough. In the parameter optimization iteration of the adaptive parameter adjustment process, there will be defects such as large amount of solution calculation, slow speed, poor real-time parameter adjustment, and easy to fall into local optimum, which will lead to major hidden dangers in the crystal pulling process.
[0053] Therefore, it is necessary to develop an intelligent PID control method, optimize the control strategy based on traditional PID control, and realize adaptive parameter adjustment of PID during the growth process of single crystal silicon.
[0054] Therefore, the present application proposes a method for controlling the growth process of single crystal silicon, and improves the PID control in the crystal pulling process on the basis of the prior art. In view of the complex crystal growth environment, a BP neural network with strong nonlinear mapping ability is added to enable the automatic parameter adjustment process to have strong self-learning ability and adaptability. Considering that when the BP neural network uses errors for back propagation, the gradient descent method solution process is prone to fall into the local optimum, resulting in low learning efficiency, unable to achieve the global optimal solution, and affecting the optimization accuracy, a particle swarm optimization algorithm is added on the basis of the BP neural network to make up for the defects of the BP neural network, so that the PID parameter optimization speed is faster, and the adaptive process is easier to jump out of the local optimum and achieve the global optimum. Combining the BP neural network and the particle swarm optimization algorithm, by following certain optimal indicators, the PID controller parameters are adjusted online intelligently and adaptively to adapt to the changes in the system parameters and structure of the single crystal furnace crystal pulling process and the changes in the input reference signal, and finally the quality of crystal pulling is improved.
[0055] It should be noted that the execution subject of this embodiment can be a computer device with closed-loop control, parameter optimization and growth control functions, such as a personal computer, a server, etc., or an electronic device capable of realizing the above functions, a single crystal silicon growth process control device (hereinafter referred to as a control device) that executes the single crystal silicon growth process control method of this application, etc., and this embodiment does not limit this. The following takes the control device as an example to illustrate this embodiment and the following embodiments.
[0056] Based on this, the present application embodiment provides a method for controlling a single crystal silicon growth process, referring to Figure 3 , Figure 3 A schematic diagram of the process flow of the single crystal silicon growth process control method provided in Example 1 of the present application.
[0057] In this embodiment, the method is applied to a control system for a single crystal silicon growth process. The system includes a PID controller, a preset neural network, and a preset population optimization algorithm. The single crystal silicon growth process control method includes steps S10 to S30:
[0058] Step S10: In the closed-loop control of the growth process of single crystal silicon, the system input, system output and system deviation of the single crystal silicon growth process control system are used as neuron inputs of the preset neural network, the system input is the reference crystal diameter of the single crystal silicon, the preset neural network is used to optimize its own neuron output, and the preset population optimization algorithm is used to optimize the preset neural network.
[0059] Step S20: using the neuron output of the preset neural network as the PID input parameter of the PID controller.
[0060] It should be noted that the PID controller is a device that controls the controlled object by forming a control deviation based on a given value and an actual output value, and by linearly combining the proportion, integral and differential of the deviation to form a control quantity. Among them, P stands for Proportion, I stands for Integral, and D stands for Derivative.
[0061] It should be noted that the preset neural network is a neural network with strong nonlinear mapping capabilities added for complex crystal growth environments, which can have strong self-learning ability and adaptability in the automatic parameter adjustment process. The preset neural network may include an input layer, a hidden layer (may have multiple layers) and an output layer. By learning the system input, system output and system deviation of the single crystal silicon growth process control system, the connection weights between neurons in the network are adjusted to achieve the mapping of input information to output information.
[0062] Exemplarily, the preset neural network may include a BP neural network, a convolutional neural network, a recurrent neural network, etc., which is not limited in this embodiment.
[0063] It should be noted that the preset swarm optimization algorithm is a preset algorithm for optimizing the preset neural network, so as to avoid the preset neural network falling into the local optimum, resulting in low learning efficiency, failure to reach the global optimal solution, and affecting the optimization accuracy.
[0064] Exemplarily, the preset swarm optimization algorithm may include a particle swarm optimization algorithm, a genetic algorithm, an ant colony optimization algorithm, a butterfly optimization algorithm, a whale optimization algorithm, etc., which is not limited in this embodiment.
[0065] It should be noted that the reference crystal diameter is used as a benchmark or standard crystal diameter size when producing single crystal silicon. During the growth process of single crystal silicon, the growth of the crystal needs to be controlled so that it reaches a specific diameter specification. This reference crystal diameter is the set target diameter.
[0066] It should be noted that the system input is various parameters or signals in the control system of the single crystal silicon growth process, including reference crystal diameter, temperature setting value, pulling speed setting value, crucible speed setting value, etc. These input information will affect the growth process of single crystal silicon.
[0067] It should be noted that the system output is the result produced by a series of processing and control. In the growth of single crystal silicon, it mainly refers to the diameter of the single crystal silicon actually grown.
[0068] It should be noted that the system deviation is the difference between the actual output of the system and the expected output. For example, the single crystal silicon is expected to reach the reference crystal diameter, but the actual grown diameter is different from the expected diameter. This difference is the system deviation.
[0069] Step S30: controlling the growth diameter of the silicon single crystal through the PID controller according to the PID input parameters, the growth diameter being used as the system output, and the difference between the growth diameter and the reference crystal diameter being used as the system deviation.
[0070] It should be noted that the PID input parameters are the parameters input by the PID controller, which are used to control the growth diameter of the single crystal silicon growth process. By adaptively adjusting the PID controller parameters, they can adapt to the changes in the system parameters of the single crystal furnace pulling process, ultimately achieving improved crystal pulling quality.
[0071] For ease of understanding, in a feasible implementation manner, reference is made to Figure 4 , Figure 4 This is a control block diagram of the BP neural network-particle swarm algorithm-PID controller in the single crystal silicon growth process provided in Example 1. This embodiment and the following embodiments are described by taking the preset neural network as the BP neural network and the preset swarm optimization algorithm as the particle swarm algorithm as an example, but this solution is not limited.
[0072] Step S10 of this embodiment may include the steps of: using the system input, system output and system deviation of the single crystal silicon growth process control system as the neuron input of the BP neural network, the BP neural network optimizes its own neuron output based on the principle of minimizing the error performance function, and the particle swarm algorithm optimizes the weight coefficient of the BP neural network.
[0073] It should be noted that the BP (Back Propagation) neural network is a multi-layer feedforward network trained by error back propagation, including input layer, hidden layer (can have multiple layers) and output layer. By learning the system input, system output and system deviation, the connection weights between neurons in the network are adjusted to achieve the mapping of system input to system output. The single crystal silicon growth process can automatically adjust parameters and has strong self-learning ability and adaptability.
[0074] It should be noted that Particle Swarm Optimization (PSO) is a random search algorithm based on group collaboration that simulates the foraging behavior of bird flocks. In the PSO, each particle represents a potential solution to the problem. They move in the solution space and continuously adjust their position and speed by tracking individual extreme values and global extreme values to search for a better solution.
[0075] Adding a particle swarm optimization algorithm on the basis of the BP neural network can make up for the defect of the BP neural network falling into the local optimum, making the parameter optimization speed of the PID control of the single crystal silicon growth process faster and achieving the global optimum.
[0076] It should be noted that the error performance function is a function used in BP neural networks to measure the degree of difference between the network output and the expected output. BP neural networks are trained and optimized with the goal of minimizing this error performance function. By continuously adjusting parameters such as the connection weights between neurons, the error generated by the network when processing new data can be minimized, thereby improving the solution ability of the BP neural network.
[0077] It should be noted that the weight coefficient is a coefficient that reflects the strength of the connection between neurons in the BP neural network. In the BP neural network, the output of each neuron is passed to the neurons in the next layer through the connection, and the weight coefficient determines the degree of influence of each input signal on the output. By adjusting the weight coefficient, the neural network can learn the relationship between different features and their contribution to the final output.
[0078] In this embodiment, if Figure 4 As shown in the figure, in the closed-loop control of the diameter of the single crystal silicon pulling process, the system input (such as the reference crystal diameter), system output and system deviation (the system deviation is obtained through CCD diameter feedback) of the single crystal furnace growth process control system are used as the neuron input of the BP neural network, and the PID parameter K p , K i , K d As the neuron output, it is input to the diameter PID controller; the mapping relationship between neuron input and output is realized through the weight coefficient in the forward propagation of the neural network, and then the gradient descent method of the neural network back propagation is used to minimize the error performance function and continuously optimize the weights between neurons. The particle swarm algorithm uses the expected output and actual output of the neuron as the fitness value, optimizes the initial weights and the direction of weight adjustment, and optimally realizes the intelligent parameter adjustment of PID control to reduce the error between the system input and the system output.
[0079] The present embodiment provides a method for controlling the growth process of single crystal silicon, which is applied to the control system of the growth process of single crystal silicon. The system includes a PID controller, a preset neural network and a preset group optimization algorithm. In the closed-loop control of the growth process of single crystal silicon, the system input, system output and system deviation of the control system of the growth process of single crystal silicon are used as the neuron input of the preset neural network, the system input is the reference crystal diameter of the single crystal silicon, the preset neural network is used to optimize its own neuron output, and the preset group optimization algorithm is used to optimize the preset neural network; the neuron output of the preset neural network is used as the PID input parameter of the PID controller; the growth diameter of the single crystal silicon is controlled by the PID controller according to the PID input parameter, the growth diameter is used as the system output, and the difference between the growth diameter and the reference crystal diameter is used as the system deviation. Since the present embodiment combines the BP neural network and the particle swarm algorithm, the parameters of the PID controller are adaptively adjusted to adapt to the changes in the parameters of the control system of the growth process of single crystal silicon, the stability of the PID controller is improved, and the quality of crystal pulling is finally improved.
[0080] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Figure 3 and Figure 5 , Figure 5 A schematic diagram of the flow chart provided for the second embodiment of the method for controlling the growth process of single crystal silicon of the present application.
[0081] In this example, step S10 also includes steps S11 to S13:
[0082] Step S11: taking the system input, the system output and the system deviation as inputs of the input layer.
[0083] Step S12: Calculate the input of the input layer according to the weight coefficient and activation function of the hidden layer to obtain the input of the output layer, where the input of the output layer is the neuron output.
[0084] Step S13: Taking minimization of the error performance function as the optimization principle, the weight coefficients are adjusted to optimize the neuron output of the BP neural network.
[0085] It should be noted that the main structure of the network can be set to 3 layers, considering the overall effect of diameter control and the training time of the BP neural network. Figure 4 and Figure 6 , Figure 6 This is a diagram of the BP neural network structure provided in Example 2. The topological structure of the BP neural network is three layers, including an input layer, a hidden layer and an output layer.
[0086] The PID controller optimized by the BP neural network can adaptively adjust the above three control parameters of the PID controller (i.e., K p , K i , K d ).
[0087] Specifically, the input of the network input layer is:
[0088]
[0089] There are 4 input variables, x 1 Input rin(k) to the system, x 2 is the system output yout(k), x 3 is the system deviation value e(k), x 4 is the bias term 1.
[0090] The input of the network hidden layer is net i (2) (k), output O j (2) (k) are:
[0091]
[0092] Among them, ω ij (2) is the weight coefficient of the hidden layer, and f(x) is the activation function of the hidden layer.
[0093] The input of the network output layer is net l (3) (k), output O l (3) (k) are:
[0094]
[0095] Among them, ω ij (3) is the weight coefficient of the output layer, and g(x) is the activation function of the output layer.
[0096] For the choice of activation function, the output layer is K p , K i , K d , these three parameters cannot be negative, and positive and negative symmetric Sigmoid functions and non-negative Sigmoid functions are selected as the activation functions between the connection layers (i.e., the input layer and the output layer), respectively. The formulas are as follows:
[0097]
[0098] By choosing the Sigmoid function as the activation function, the function derivative is simple and similar to the actual response of biological neurons.
[0099] Finally, taking the minimization of the error performance function as the optimization principle, the above weight coefficients are adjusted to optimize the neuron output of the BP neural network.
[0100] In a feasible implementation manner, step S13 of this embodiment may include the steps of: taking minimization of the error performance function as the optimization principle, adjusting the weight coefficients by a gradient descent method to optimize the neuron output of the BP neural network.
[0101] It should be noted that the gradient descent method is an iterative algorithm for optimizing the error performance function. The basic idea of the gradient descent method is to calculate the gradient of the error performance function with respect to each parameter (i.e., the direction of the rate of change of the function at that point), and then gradually adjust the value of the parameter in the opposite direction of the gradient so that the function value gradually decreases. By continuously iterating this process, the minimum point of the function is gradually approached.
[0102] Specifically, in this implementation, the error performance function is:
[0103]
[0104] The weight coefficients of the neural network are constantly adjusted, and the update method is the gradient descent method. The error performance function E(k) is searched along the negative direction of the gradient of the weight coefficient. The minimum inertia term that converges quickly is:
[0105]
[0106] Among them, η is the learning rate; α is the inertia coefficient, and Respectively represent the changes in weight coefficients at time k and time k-1, Represents the partial derivative of the error performance function with respect to the change in weight coefficients; represents the partial derivative of the error performance function with respect to the system output; represents the partial derivative of the controller output increment to the output of the neural network output layer; Represents the partial derivative of the output layer output to the output layer input; Represents the partial derivative of the output layer input with respect to the output layer weight.
[0107] The incremental PID control is used in this control system, and its specific control formula is:
[0108] Δu(k)=K p (e(k)-e(k-1))+K i e(k)+K d(e(k)-2e(k-1)+e(k-2))
[0109] u(k)=u(k-1)+Δu(k);
[0110] Wherein, Δu(k) is the increment of controller output at time k; e(k), e(k-1), and e(k-2) represent the system deviations at time k, k-1, and k-2, respectively; and u(k) and u(k-1) represent the controller outputs at time k and k-1, respectively.
[0111] Therefore, in the weighted derivative formula of the error performance function:
[0112]
[0113] The learning formula for the output layer weight coefficient is:
[0114]
[0115] The learning formula for the hidden layer weight coefficient is:
[0116]
[0117] In another possible implementation, reference Figure 7 , Figure 7 The optimization flow chart combining BP neural network and particle swarm algorithm provided for Example 2. The particle swarm algorithm described in this embodiment optimizes the weight coefficients of the BP neural network, including: initializing the weight coefficients of the BP neural network according to the topological structure to obtain initial weights; using the initial weights as initialization particles, the initialization particles including position and speed; obtaining the training error of the BP neural network; according to the training error, iterating the initialization particles through the particle swarm algorithm to obtain the optimal weights.
[0118] It should be noted that the initial weights are random weights of the BP neural network. When the initial weights are inappropriate, it will affect the adjustment direction of the weights and thus affect the optimization accuracy. Therefore, the particle swarm algorithm can be used to optimize the initial weights.
[0119] Considering that the BP neural network operation response time is too long, the learning efficiency is low, and it is easy to produce local optimal solutions, when applied to PID parameter adjustment, its control optimization effect is not ideal. Moreover, when the initial weights are inappropriate, it will affect the adjustment direction of the weights and thus affect the optimization accuracy. Therefore, the particle swarm algorithm can be used to optimize the initial weights. The particle swarm algorithm can directly find the optimal solution. Compared with other intelligent optimization algorithms, its optimization speed is faster and more suitable for the control of the single crystal silicon growth process.
[0120] In this embodiment, the control device can first clarify the topological structure of the BP neural network, initialize the weight coefficients of the BP neural network according to the topological structure, and obtain the initial weights; use the initial weights as the initialization particles, and the initialization particles include position and speed; obtain the training error of the BP neural network; according to the training error, iterate the initialization particles through the particle swarm algorithm to obtain the optimal weights. Then, the BP neural network is trained according to the optimal weights, the training error is calculated again, and it is determined whether the training error meets the termination condition (for example, the error rate of the training error is less than one percent). If not, the weights are updated and the training is repeated; if satisfied, the training is terminated.
[0121] In another feasible implementation, the step of iterating the initialized particles through the particle swarm algorithm according to the training error to obtain the optimal weight described in this embodiment includes: taking the training error as the initial fitness value; iterating the initialized particles through the particle swarm algorithm according to the initial fitness value to obtain individual extreme values and global extreme values corresponding to the initialized particles; and determining the optimal weight according to the individual extreme values and the global extreme values.
[0122] It should be noted that the individual extreme value is the optimal position (i.e. the corresponding function value is optimal) that each particle has reached during the iteration process in the particle swarm algorithm. Each particle has its own individual extreme value record.
[0123] The global extreme value refers to the optimal position (corresponding to the best function value) found by the entire particle group among all the positions experienced by the particles. It represents the optimal solution currently found by the entire group.
[0124] Through continuous iteration, particles adjust their speed and position according to their own individual extreme values and global extreme values, thus gradually approaching the optimal solution to the problem.
[0125] In this embodiment, after initializing the speed and position of the particle, the error obtained by BP neural network training can be used as the fitness value; the speed and position of the particle are updated according to the fitness value, and the fitness value of the updated particle is calculated again to update the individual extreme value and the global extreme value. It is judged whether the termination condition is met. If not, the speed and position of the particle are updated again; if so, the optimal weight is determined according to the individual extreme value and the global extreme value.
[0126] For ease of understanding, specifically, each particle updates its position and speed in the search space according to the principle of fitness optimization, and finds individual extreme values and global extreme values through iteration. In an m-dimensional search space (reflected as the number of weight coefficients in the BP neural network), there are n particles, and when the number of iterations is k, the speed and position update formula of the i-th particle in the d-th dimension is as follows:
[0127]
[0128] Among them, c 1 and c 2 is the learning factor; r 1 and r 2 is a random number distributed in [0,1]; P id is the individual optimal solution of the particle; P gd It is the global optimal solution.
[0129] Among them, the fitness selection principle of the particle swarm algorithm is: BP neural network back propagation, using the gradient descent method to obtain the minimum system deviation, the obtained K p , K i , K d is the expected value at this time; K obtained by the forward propagation of the BP neural network p , K i , K d The actual value at this time, the fitness function is the deviation between the actual value and the expected value, and the initial weights and weight adjustment direction of the neural network are optimized according to the principle of minimizing the fitness function to avoid problems such as slow algorithm convergence and local optimality. The fitness function formula is as follows:
[0130]
[0131] Among them, E i is the fitness function of the ith particle; the expected output and actual output of neuron j corresponding to particle i are O ij and ij .
[0132] Based on the traditional PID control technology, this embodiment improves the PID control in the crystal pulling process, adds BP neural network and particle swarm optimization algorithm to optimize the PID controller, complies with the optimization index of minimizing system deviation, and adjusts the PID controller parameters online and intelligently to adapt to the changes in the single crystal furnace crystal pulling control parameters and structure and the changes in the input reference signal. Through the BP neural network, it has a strong nonlinear optimization ability and strong adaptability in a complex crystal growth environment; at the same time, adding a particle swarm algorithm on the basis of the BP neural network, the optimization speed is faster, and it is easier to jump out of the local optimum and achieve the global optimum. Therefore, it can be better applied in a complex single crystal silicon growth environment, and the PID parameters can be fully automatically adjusted in real time online according to the single crystal silicon pulling state.
[0133] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the single crystal silicon growth process control method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.
[0134] The present application also provides a single crystal silicon growth process control device, please refer to Figure 8 , Figure 8 This is a schematic diagram of the module structure of the single crystal silicon growth process control device of the embodiment of the present application; the device is applied to the single crystal silicon growth process control system, the system includes a PID controller, a preset neural network and a preset population optimization algorithm, and the single crystal silicon growth process control device includes:
[0135] A parameter optimization module 801 is used to use the system input, system output and system deviation of the single crystal silicon growth process control system as the neuron input of the preset neural network in the closed-loop control of the single crystal silicon growth process, the system input is the reference crystal diameter of the single crystal silicon, the preset neural network is used to optimize its own neuron output, and the preset population optimization algorithm is used to optimize the preset neural network;
[0136] A parameter output module 802, used to use the neuron output of the preset neural network as a PID input parameter of the PID controller;
[0137] The growth control module 803 is used to control the growth diameter of the single crystal silicon through the PID controller according to the PID input parameters, the growth diameter is used as the system output, and the difference between the growth diameter and the reference crystal diameter is used as the system deviation.
[0138] The single crystal silicon growth process control device provided by the present application adopts the single crystal silicon growth process control method in the above embodiment, which can solve the technical problem that the traditional PID control is difficult to adapt to the complex single crystal silicon pulling growth environment, resulting in poor control accuracy and serious impact on the quality of the pulling crystal. Compared with the prior art, the beneficial effects of the single crystal silicon growth process control device provided by the present application are the same as the beneficial effects of the single crystal silicon growth process control method provided by the above embodiment, and the other technical features of the single crystal silicon growth process control device are the same as the features disclosed in the above embodiment method, which will not be repeated here.
[0139] The present application provides a single crystal silicon growth process control device, which includes: at least one processor; and a memory that is communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed 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 single crystal silicon growth process control method in the above-mentioned embodiment one.
[0140] Reference below Fig. 9 , Fig. 9The device structure diagram of the hardware operating environment involved in the single crystal silicon growth process control method in the embodiment of the present application shows a schematic diagram of the structure of the single crystal silicon growth process control device suitable for implementing the embodiment of the present application. The single crystal silicon growth process control 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. Fig. 9 The single crystal silicon growth process control device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.
[0141] like Fig. 9 As shown, the single crystal silicon growth process control device may include a processing device 1001 (such as a central processing unit, a graphics processing unit, 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. Various programs and data required for the operation of the single crystal silicon growth process control device are also stored in the RAM 1004. The processing device 1001, the ROM 1002, and the RAM 1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 can allow the single crystal silicon growth process control device to communicate with other devices wirelessly or by wire to exchange data. Although the single crystal silicon growth process control device with various systems is shown in the figure, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have instead.
[0142] 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.
[0143] The single crystal silicon growth process control device provided by the present application adopts the single crystal silicon growth process control method in the above embodiment, which can solve the technical problem that the traditional PID control is difficult to adapt to the complex single crystal silicon pulling growth environment, resulting in poor control accuracy and serious impact on the quality of the pulling crystal. Compared with the prior art, the beneficial effects of the single crystal silicon growth process control device provided by the present application are the same as the beneficial effects of the single crystal silicon growth process control method provided by the above embodiment, and the other technical features of the single crystal silicon growth process control device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.
[0144] 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.
[0145] 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.
[0146] 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 single crystal silicon growth process control method in the above-mentioned embodiment.
[0147] 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.
[0148] The computer-readable storage medium may be included in the single crystal silicon growth process control device; or may exist independently without being assembled into the single crystal silicon growth process control device.
[0149] The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by the single crystal silicon growth process control device, the single crystal silicon growth process control device: in the closed-loop control of the single crystal silicon growth process, the system input, system output and system deviation of the single crystal silicon growth process control system are used as the neuron input of the preset neural network, the system input is the reference crystal diameter of the single crystal silicon, the preset neural network is used to optimize its own neuron output, and the preset population optimization algorithm is used to optimize the preset neural network; the neuron output of the preset neural network is used as the PID input parameter of the PID controller; the growth diameter of the single crystal silicon is controlled by the PID controller according to the PID input parameter, the growth diameter is used as the system output, and the difference between the growth diameter and the reference crystal diameter is used as the system deviation.
[0150] 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).
[0151] 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.
[0152] 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.
[0153] 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 single crystal silicon growth process control method, and can solve the technical problem that the traditional PID control is difficult to adapt to the complex single crystal silicon pulling growth environment, resulting in poor control accuracy and serious impact on the quality of the pulling crystal. 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 single crystal silicon growth process control method provided in the above-mentioned embodiment, and will not be repeated here.
[0154] The present application also provides a computer program product, including a computer program, which implements the steps of the single crystal silicon growth process control method as described above when executed by a processor.
[0155] The computer program product provided by the present application can solve the technical problem that the traditional PID control is difficult to adapt to the complex single crystal silicon pulling growth environment, resulting in poor control accuracy and serious impact on the quality of the crystal pulling. 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 single crystal silicon growth process control method provided by the above embodiment, and will not be repeated here.
[0156] 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 controlling a single crystal silicon growth process, characterized in that: The method is applied to a control system for a single crystal silicon growth process, the system comprising a PID controller, a preset neural network and a preset population optimization algorithm, and the method comprises: In the closed-loop control of the growth process of single crystal silicon, the system input, system output and system deviation of the single crystal silicon growth process control system are used as the neuron input of the preset neural network, the system input is the reference crystal diameter of the single crystal silicon, the preset neural network is used to optimize its own neuron output, and the preset population optimization algorithm is used to optimize the preset neural network; Using the neuron output of the preset neural network as the PID input parameter of the PID controller; The growth diameter of the silicon single crystal is controlled by the PID controller according to the PID input parameters, the growth diameter is used as the system output, and the difference between the growth diameter and the reference crystal diameter is used as the system deviation.
2. The method according to claim 1, characterized in that The preset neural network is a BP neural network, the preset population optimization algorithm is a particle swarm algorithm, and the step of using the system input, system output and system deviation of the single crystal silicon growth process control system as neuron inputs of the preset neural network includes: The system input, system output and system deviation of the single crystal silicon growth process control system are used as the neuron input of the BP neural network. The BP neural network optimizes its own neuron output based on the principle of minimizing the error performance function. The particle swarm algorithm optimizes the weight coefficient of the BP neural network.
3. The method according to claim 2, characterized in that The topological structure of the BP neural network is three layers, including an input layer, a hidden layer and an output layer; the BP neural network optimizes its own neuron output based on the principle of minimizing the error performance function, including: Taking the system input, the system output and the system deviation as inputs of the input layer; The input of the input layer is calculated according to the weight coefficient and activation function of the hidden layer to obtain the input of the output layer, where the input of the output layer is the neuron output; Taking minimization of the error performance function as the optimization principle, the weight coefficients are adjusted to optimize the neuron output of the BP neural network.
4. The method according to claim 3, characterized in that The step of adjusting the weight coefficients based on minimization of the error performance function as the optimization principle to optimize the neuron output of the BP neural network includes: Taking minimization of the error performance function as the optimization principle, the weight coefficients are adjusted by the gradient descent method to optimize the neuron output of the BP neural network.
5. The method according to claim 3, characterized in that The particle swarm algorithm optimizes the weight coefficients of the BP neural network, including: Initializing the weight coefficients of the BP neural network according to the topological structure to obtain initial weights; Using the initial weight as an initialization particle, wherein the initialization particle includes a position and a velocity; Obtaining a training error of the BP neural network; According to the training error, the initialized particles are iterated by the particle swarm algorithm to obtain an optimal weight.
6. The method according to claim 5, characterized in that The step of iterating the initialized particles through the particle swarm algorithm according to the training error to obtain the optimal weight includes: Using the training error as an initial fitness value; According to the initial fitness value, the initialized particles are iterated by the particle swarm algorithm to obtain individual extreme values and global extreme values corresponding to the initialized particles; An optimal weight is determined according to the individual extreme value and the global extreme value.
7. A single crystal silicon growth process control device, characterized in that: The device is applied to a control system for a single crystal silicon growth process, the system comprises a PID controller, a preset neural network and a preset population optimization algorithm, and the device comprises: A parameter optimization module, used for using the system input, system output and system deviation of the single crystal silicon growth process control system as the neuron input of the preset neural network in the closed-loop control of the single crystal silicon growth process, the system input is the reference crystal diameter of the single crystal silicon, the preset neural network is used to optimize its own neuron output, and the preset population optimization algorithm is used to optimize the preset neural network; A parameter output module, used for using the neuron output of the preset neural network as a PID input parameter of the PID controller; A growth control module is used to control the growth diameter of the single crystal silicon through the PID controller according to the PID input parameters, the growth diameter is used as the system output, and the difference between the growth diameter and the reference crystal diameter is used as the system deviation.
8. A single crystal silicon growth process control 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 single crystal silicon growth process control 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 single crystal silicon growth process control 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 method for controlling a single crystal silicon growth process according to any one of claims 1 to 6 are implemented.
