Boron diffusion formula parameter optimization method, device, equipment and medium

Abstract

The invention relates to a boron diffusion formula parameter optimization method and device, equipment and a medium, and the method comprises the steps: obtaining boron diffusion sample data, sample sheet resistance data corresponding to the sample data, and formula parameter characteristics corresponding to the sample data; generating process parameter characteristics based on the formula parameter characteristics and a boron diffusion process calculation formula; inputting the formula parameter characteristics and the process parameter characteristics into an initial diffusion quality prediction model, and constructing a loss function based on prediction sheet resistance data and sample sheet resistance data output by the initial diffusion quality prediction model; training and optimizing the initial diffusion quality prediction model by adopting a model optimization algorithm based on the loss function until a preset training ending condition is met, and obtaining a trained diffusion quality prediction model; and according to the trained diffusion quality prediction model, carrying out initial optimal formula parameter generation and key process parameter characteristic real-time regulation and control.

Description

Technical Field

[0001] The present application relates to the field of intelligent optimization technology, and in particular to a method, device, computer equipment, storage medium and computer program product for optimizing boron diffusion formula parameters. Background Art

[0002] As the core of photovoltaic technology, solar cell technology has made remarkable progress in recent decades. Photoelectric conversion efficiency has been continuously improved and production costs have been gradually reduced. Traditional P-type solar cells are gradually being replaced by new generation technologies because their photoelectric conversion efficiency is close to the theoretical limit. As an emerging technology, N-type TOPCon solar cells have become the main development direction of future solar cell technology due to their compatibility with P-type PERC solar cell manufacturing process, small investment in upgrading, mature industrial chain and large-scale mass production conditions.

[0003] In the manufacturing process of N-type TOPCon solar cells, the boron diffusion process plays a vital role. Boron diffusion is a key semiconductor manufacturing step, which is used to dope boron atoms on the surface of N-type silicon wafers to form a PN structure, which directly affects the performance and efficiency of the cell. In semiconductor manufacturing, optimizing the formulation parameters of boron diffusion is crucial to improving product performance and quality.

[0004] Traditional boron diffusion formula parameter optimization methods mainly rely on manual experience and experiments. These methods are not only time-consuming and labor-intensive, but also unable to obtain accurate optimization results. Summary of the invention

[0005] Based on this, it is necessary to provide an accurate boron diffusion formula parameter optimization method, device, computer equipment, storage medium and computer program product to address the above technical problems.

[0006] In a first aspect, the present application provides a method for optimizing boron diffusion recipe parameters. The method comprises:

[0007] Acquire sample data of boron diffusion, wherein the sample data includes sample sheet resistance data and corresponding recipe parameter characteristics;

[0008] Generate process parameter characteristics based on the recipe parameter characteristics and the boron diffusion process calculation formula;

[0009] Inputting the recipe parameter characteristics and the process parameter characteristics into an initial diffusion mass prediction model, and constructing a loss function based on the predicted square resistance data output by the initial diffusion mass prediction model and the sample square resistance data;

[0010] Based on the loss function, the initial diffusion mass prediction model is trained and optimized using a model optimization algorithm until a preset training end condition is reached to obtain a trained diffusion mass prediction model;

[0011] Initial optimal recipe parameters are generated and key process parameter characteristics are controlled in real time based on the trained diffusion quality prediction model.

[0012] In one embodiment, obtaining sample data of boron diffusion includes:

[0013] Obtain historical data on boron diffusion;

[0014] Extracting keywords from the historical data, the keywords including equipment number, pipe number and boat number;

[0015] Concatenate the data corresponding to the same keyword to generate sample data.

[0016] In one embodiment, the sample square resistance data corresponding to the sample data includes:

[0017] Extracting the square resistance value of the test silicon wafer from the sample data based on a preset square resistance sampling rule;

[0018] Calculating the square resistance average and uniformity value of the test silicon wafer according to the square resistance value of the test silicon wafer;

[0019] The square resistance mean value and the square resistance uniformity value are aggregated to obtain sample square resistance data.

[0020] In one embodiment, the method further comprises:

[0021] According to the sample data, obtaining the correlation between the temperature zone of the boron diffusion device and the boat;

[0022] Based on the relationship between the medium temperature zone of the boron diffusion equipment and the small boat, the relationship between the medium temperature zone of the boron diffusion and the small boat square resistance is established;

[0023] An attention mechanism is introduced according to the relationship between the temperature zone and the boat square resistance to generate an initial diffusion quality prediction model.

[0024] In one embodiment, the initial diffusion mass prediction model is trained and optimized using a model optimization algorithm based on the loss function until a preset training end condition is reached, and the trained diffusion mass prediction model is obtained, including:

[0025] Based on the loss function, the initial diffusion mass prediction model is trained by using the Adam optimization algorithm until the minimum value of the loss function is obtained, thereby obtaining a trained diffusion mass prediction model.

[0026] In one embodiment, generating initial optimal recipe parameters and real-time control of key process parameter characteristics according to the trained diffusion quality prediction model includes:

[0027] Based on each type of boron diffusion recipe parameters, set the search range and step size;

[0028] According to the search range and step size of each boron diffusion formula parameter, a plurality of boron diffusion formula parameters are randomly generated and encoded into a plurality of individuals to generate an initial population;

[0029] Using the trained diffusion quality prediction model to predict the diffusion quality of the boron diffusion formula parameters corresponding to the variant population, obtaining a diffusion quality prediction result, and transforming and quantifying the result into a fitness value;

[0030] Based on the fitness value of each individual in the population, the population is selected, crossed, and mutated to obtain a new generation of population;

[0031] Repeating the population diffusion quality prediction, fitness calculation, population selection, crossover, and mutation operations, and continuously evolving to reach a preset iteration stop condition;

[0032] The boron diffusion formula parameters corresponding to the individual with the best diffusion quality in the diffusion quality prediction result are queried to generate initial optimal formula parameters, wherein the diffusion quality is characterized based on the square resistance data.

[0033] In one embodiment, the generating of initial optimal recipe parameters and real-time control of key process parameter characteristics according to the trained diffusion quality prediction model further includes:

[0034] Based on the trained diffusion quality prediction model, obtaining the correlation between adjustable process parameter characteristics and diffusion quality, wherein the adjustable process parameter characteristics include temperature, boron source flow rate, and diffusion time;

[0035] According to the correlation between the process parameter characteristics and the diffusion quality, obtaining the adjustment threshold of each process parameter characteristic for the corresponding diffusion square resistance;

[0036] According to the adjustment thresholds of the various process parameter characteristics, the boron diffusion process is controlled in real time in the order of the adjustable process parameter characteristics.

[0037] In a second aspect, the present application also provides a boron diffusion formula parameter optimization device. The device comprises:

[0038] A data acquisition module, used to acquire sample data of boron diffusion, wherein the sample data includes sample sheet resistance data and corresponding recipe parameter characteristics;

[0039] A process parameter feature acquisition module, used to generate process parameter features based on the recipe parameter features and the boron diffusion process calculation formula;

[0040] A model initialization module, used for inputting the recipe parameter characteristics and the process parameter characteristics into an initial diffusion mass prediction model, and constructing a loss function based on the predicted square resistance data output by the initial diffusion mass prediction model and the sample square resistance data;

[0041] A model training module is used to train and optimize the initial diffusion mass prediction model using a model optimization algorithm based on the loss function until a preset training end condition is reached to obtain a trained diffusion mass prediction model; an optimization and control module is used to generate initial optimal formula parameters and perform real-time control of key process parameter characteristics according to the trained diffusion mass prediction model.

[0042] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0043] Acquire sample data of boron diffusion, wherein the sample data includes sample sheet resistance data and corresponding recipe parameter characteristics;

[0044] Generate process parameter characteristics based on the recipe parameter characteristics and the boron diffusion process calculation formula;

[0045] Inputting the recipe parameter characteristics and the process parameter characteristics into an initial diffusion mass prediction model, and constructing a loss function based on the predicted square resistance data output by the initial diffusion mass prediction model and the sample square resistance data;

[0046] Based on the loss function, the initial diffusion mass prediction model is trained and optimized using a model optimization algorithm until a preset training end condition is reached to obtain a trained diffusion mass prediction model;

[0047] The trained diffusion quality prediction model is used to generate the initial optimal formulation parameters for boron diffusion and to control the key process parameter characteristics in real time.

[0048] In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0049] Acquire sample data of boron diffusion, wherein the sample data includes sample sheet resistance data and corresponding recipe parameter characteristics;

[0050] Generate process parameter characteristics based on the recipe parameter characteristics and the boron diffusion process calculation formula;

[0051] Inputting the recipe parameter characteristics and the process parameter characteristics into an initial diffusion mass prediction model, and constructing a loss function based on the predicted square resistance data output by the initial diffusion mass prediction model and the sample square resistance data;

[0052] Based on the loss function, the initial diffusion mass prediction model is trained and optimized using a model optimization algorithm until a preset training end condition is reached to obtain a trained diffusion mass prediction model;

[0053] The trained diffusion quality prediction model is used to generate the initial optimal formulation parameters for boron diffusion and to control the key process parameter characteristics in real time.

[0054] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0055] Acquire sample data of boron diffusion, wherein the sample data includes sample sheet resistance data and corresponding recipe parameter characteristics;

[0056] Generate process parameter characteristics based on the recipe parameter characteristics and the boron diffusion process calculation formula;

[0057] Inputting the recipe parameter characteristics and the process parameter characteristics into an initial diffusion mass prediction model, and constructing a loss function based on the predicted square resistance data output by the initial diffusion mass prediction model and the sample square resistance data;

[0058] Based on the loss function, the initial diffusion mass prediction model is trained and optimized using a model optimization algorithm until a preset training end condition is reached to obtain a trained diffusion mass prediction model;

[0059] The trained diffusion quality prediction model is used to generate the initial optimal formulation parameters for boron diffusion and to control the key process parameter characteristics in real time.

[0060] The above-mentioned boron diffusion formula parameter optimization method, device, computer equipment, storage medium and computer program product obtain sample data of boron diffusion, sample square resistance data corresponding to the sample data, and formula parameter characteristics corresponding to the sample data; generate process parameter characteristics based on the formula parameter characteristics and the boron diffusion process calculation formula; input the formula parameter characteristics and process parameter characteristics into the initial diffusion quality prediction model, and construct a loss function based on the predicted square resistance data and sample square resistance data output by the initial diffusion quality prediction model; train and optimize the initial diffusion quality prediction model based on the loss function using a model optimization algorithm until the preset training end condition is reached to obtain a trained diffusion quality prediction model; generate the initial optimal formula parameters and control the key process parameter characteristics in real time according to the trained diffusion quality prediction model. In the whole process, the sample square resistance data and process parameter characteristics are obtained for the sample data of boron diffusion, and the loss function is constructed around the square resistance data. Finally, the trained diffusion quality prediction model that characterizes the process parameter characteristics and square resistance data (boron diffusion quality) is trained, and the initial optimal formula parameters are generated and the key process parameter characteristics are controlled in real time through the trained diffusion quality prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a schematic diagram of the reaction chamber of the boron diffusion equipment;

[0062] Figure 2 A diagram showing an application environment of a boron diffusion recipe parameter optimization method in one embodiment;

[0063] Figure 3 A schematic diagram of a process for optimizing boron diffusion recipe parameters in one embodiment;

[0064] Figure 4 Schematic diagram of the architecture of the initial diffusion mass prediction model;

[0065] Figure 5 is a schematic diagram of a sub-process of S500 in one embodiment;

[0066] Figure 6 This is a schematic diagram of the recipe parameter optimization process;

[0067] Figure 7 This is a schematic diagram of the comparison of the average square resistance of furnace No. 1 in an application example;

[0068] Figure 8 It is a structural block diagram of a boron diffusion recipe parameter optimization device in one embodiment;

[0069] Fig. 9 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0070] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0071] In order to explain in detail the technical solution and technical effect of the boron diffusion formula parameter optimization method of the present application, the following will first introduce the contents related to boron diffusion.

[0072] 1. Boron Diffusion Principle

[0073] Boron diffusion is a key process for doping boron atoms into silicon wafers to form P-type semiconductor regions. This process uses high temperatures to migrate boron atoms from high-concentration areas to low-concentration areas, ensuring the uniformity and stability of the doping layer, thereby improving the performance of electronic devices such as solar cells and integrated circuits.

[0074] 2. Boron Diffusion Process

[0075] The boron diffusion process is crucial in semiconductor manufacturing and involves multiple steps to ensure the uniformity and stability of the doped layer. The process requires precise control of multiple parameters, including temperature, time, and gas flow. During the diffusion process, a large amount of nitrogen is required to provide an inert atmosphere to ensure the stability and purity of the process and prevent unnecessary chemical reactions. In addition, nitrogen (small nitrogen) and oxygen that carry the boron source into the tube body must be introduced into the quartz tube at the same time.

[0076] The specific steps include pre-deposition and advancement stages. In the pre-deposition stage, a small amount of nitrogen carries the boron source into the high-temperature quartz tube and reacts with oxygen to generate liquid B. 2 O 3 , Advancement Phase B 2 O 3 The boron atoms react with silicon to form boron atoms, which diffuse into the silicon lattice to form a doped layer. The pre-deposition stage and the push-in stage can be carried out simultaneously to some extent. However, the pre-deposition time is shorter than the push-in stage, resulting in the boron diffusion mainly being carried out in the subsequent push-in stage. The pre-deposition step will form a layer of borosilicate glass (BSG) on the surface of the silicon wafer, which is a kind of boron oxide B 2 O 3 、Silicon oxide SiO 2 A mixture of boron B.

[0077] 1) Chemical reaction

[0078] Boron trichloride (BCl 3 ) is used as an example for the diffusion of boron source, and the specific chemical reaction process is as follows:

[0079] 5BCl 3 +3O 2 →2B 2 O3 +6Cl 2

[0080] 2B 2 O 3 +3Si→3SiO 2 +4B

[0081] Si+O 2 →SiO 2

[0082] 2) Influence of gas flow

[0083] a. Nitrogen flow rate: As a protective gas, it provides an inert environment to prevent the silicon wafer from being oxidized or reacting with other impurities in a high temperature environment. Too large a flow rate will affect the temperature uniformity in the furnace, and thus affect the diffusion resistance uniformity; too small a flow rate may not be able to effectively remove the reaction products and impurity gases, affecting the doping quality.

[0084] b. Small nitrogen flow (carrying boron source): directly affects the quality of diffusion. Too large a flow may cause the surface concentration to be too high, or even alloy points to appear; too small a flow may lead to insufficient concentration of impurity sources in the tube and uneven diffusion. The actual boron source flow carried in small nitrogen is usually difficult to calculate. Since a reduction in small nitrogen flow will inevitably increase the sheet resistance, attention should be paid to the change in sheet resistance uniformity for this flow.

[0085] c. Oxygen flow rate: It helps control the diffusion depth and doping concentration during the boron diffusion process, and can also improve the uniformity and efficiency of doping. Too much oxygen flow rate will lead to a thick oxide layer, affecting subsequent processes; too little flow rate will lead to an insufficient oxide layer and increased surface defects.

[0086] 3) Diffusion process

[0087] During the diffusion process, Fick's law can be used to estimate the impurity concentration or mean square resistance after diffusion. Fick's second law describes the diffusion process of substances in a material, and its basic form is:

[0088]

[0089] Where D is the diffusion coefficient, t is the diffusion time, x is the diffusion depth, and C is the concentration of the impurity. During the boron diffusion process, the various parameters are interrelated through the diffusion equation, affecting the final diffusion effect and material properties. The specific introduction of these parameters is as follows:

[0090] a. Diffusion coefficient D: The diffusion coefficient directly affects the diffusion process, and the diffusion temperature T is one of the most important factors affecting the diffusion process. The relationship between the diffusion coefficient D(T) and the diffusion temperature T can be expressed by the Arrhenius formula:

[0091]

[0092] Among them, D 0 is the pre-exponential factor, E a is the diffusion activation energy, k is the Boltzmann constant, and T is the absolute temperature. As can be seen from the formula, the diffusion coefficient is mainly determined by temperature. When the diffusion temperature is high, the diffusion coefficient is large, the diffusion rate is fast, the diffusion concentration is high, the square resistance mean value will become low, and the uniformity will be difficult to control. When the diffusion temperature is lowered, the diffusion coefficient is small, the diffusion rate is slow, the diffusion concentration is low, and the uniformity is poor.

[0093] b. Diffusion depth x: The diffusion distance of boron atoms in the material, which is affected by the diffusion coefficient D and the diffusion time t (the duration of the diffusion of boron atoms in the silicon wafer), determines the thickness of the doping layer:

[0094]

[0095] c. Impurity concentration C: The concentration of boron atoms at a certain position during the diffusion process directly affects the electrical properties and uniformity of the doped layer and is affected by the total doping dose Q (the total number of boron atoms, which can be determined by controlling the boron source flow rate). Through Fick's second law, the concentration at different times and positions can be calculated according to the following formula.

[0096]

[0097] In summary, the key parameters affecting the diffusion process include diffusion temperature, diffusion time and boron source BCl 3 flow The calculation formula for the total doping dose Q is:

[0098] 3. Boron Diffusion Resistance

[0099] 1) Principle

[0100] Square resistance R s , that is, the resistance per unit area of ​​the thin film material, is calculated as:

[0101]

[0102] Where ρ is the resistivity and t is the film thickness. The sheet resistance is related to the entire diffusion process, and optimizing these parameters in the diffusion process can significantly affect the mean sheet resistance and its uniformity.

[0103] 2) Square resistance average

[0104] Schematic diagram of the reaction chamber of the boron diffusion equipment Figure 1As shown. The furnace body is generally divided into 6 temperature zones according to the distribution of temperature control areas, and the serial numbers 1-6 in the figure represent temperature zones 1-6. The silicon wafer is inserted horizontally, and 12 boats are placed on the boat tow, numbered 1-12. As can be seen from the figure, boats 1 and 3 are mainly in temperature zone 2, boat 5 is in temperature zone 3, boat 7 is in temperature zone 4, boats 10 and 12 are in temperature zone 5, and temperature zones 1 and 6 belong to constant temperature zones. The square resistance sampling rule is to take the middle piece position (the 60th piece) of boats 1, 3, 5, 7, 10, and 12, and use the 9-point method to test the value of the silicon wafer, that is, divide the silicon wafer into 9 squares, and then take 1 point each, and the average value of these 9 points is taken as the diffusion square resistance mean value of the silicon wafer.

[0105] 3) Square resistance uniformity

[0106] In addition, according to the maximum and minimum values ​​of the square resistance values ​​of the 9 points obtained by the above 9-point method, the intra-wafer uniformity of the silicon wafer can be calculated according to the following calculation formula:

[0107]

[0108] The boron diffusion formula parameter optimization method provided in the embodiment of the present application can be applied to Figure 2 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. The terminal 102 sends a boron diffusion recipe parameter optimization request to the server 104, and the server 104 obtains the sample data of the boron diffusion, and the sample data includes the sample square resistance data and the corresponding recipe parameter characteristics; based on the recipe parameter characteristics and the boron diffusion process calculation formula, the process parameter characteristics are generated; the recipe parameter characteristics and the process parameter characteristics are input into the initial diffusion quality prediction model, and the loss function is constructed based on the predicted square resistance data and the sample square resistance data output by the initial diffusion quality prediction model; based on the loss function, the model optimization algorithm is used to train and optimize the initial diffusion quality prediction model until the preset training end condition is reached, and the trained diffusion quality prediction model is obtained; the initial optimal recipe parameters are generated and the key process parameter characteristics are controlled in real time according to the trained diffusion quality prediction model. The terminal 102 may be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. The portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers.

[0109] In one embodiment, Figure 3As shown, a method for optimizing boron diffusion recipe parameters is provided, and the method is applied to Figure 2 Taking the server 104 in the example as an example, the following steps are included:

[0110] S100: Acquire sample data of boron diffusion, where the sample data includes sample sheet resistance data and corresponding recipe parameter characteristics.

[0111] The sample data of boron diffusion includes multiple sample data, each of which includes data composed of multiple keywords. The keywords here can specifically include equipment number, tube number, boat number, etc. In addition, the sample data is generated one by one in combination with the detection time. The sample square resistance data corresponding to the sample data refers to the square resistance data corresponding to the sample data state. The square resistance data can specifically include data in two dimensions: square resistance mean and square resistance uniformity. In practical applications, the square resistance data table should contain data from 9 points of square resistance in areas 1, 3, 5, 7, 10, and 12. Based on these 9 points, the square resistance mean and square resistance uniformity can be calculated. The square resistance mean and uniformity data are shown in Table 1 below.

[0112] Table 1

[0113]

[0114] S200: Generate process parameter characteristics based on the recipe parameter characteristics and the boron diffusion process calculation formula.

[0115] Obtain the formula parameter data (features) corresponding to the sample data, that is, the key parameters in each step, including gas flow data (nitrogen, nitrogen, oxygen), diffusion temperature in six temperature zones, diffusion time, boron source BCl 3 Flow rate. The process parameters and temperature are shown in Table 2 and Table 3 below, respectively. All data are spliced ​​according to the sample number as the final sample.

[0116] Table 2

[0117]

[0118] Table 3

[0119]

[0120] Based on the formulas in the diffusion process mentioned above, process parameter features for enriching feature representation are generated based on the recipe parameter features, including the diffusion coefficient and diffusion depth of each temperature zone, the total doping dose, and the impurity concentration.

[0121] S300: inputting recipe parameter characteristics and process parameter characteristics into an initial diffusion quality prediction model, and constructing a loss function based on predicted square resistance data and sample square resistance data output by the initial diffusion quality prediction model.

[0122] The initial diffusion quality prediction model is a pre-built model, which can be an initial model built for boron diffusion application scenarios. Specifically, Figure 1 As shown in the figure, the schematic diagram of the reaction chamber of the boron diffusion equipment shows the relationship between the temperature zones and the boats. For any temperature zone, it is determined that the temperature of the temperature zone can affect: (1) the silicon wafer in the current temperature zone; (2) the silicon wafer in the adjacent temperature zone near the edge of the temperature zone. Based on this, the initial diffusion quality prediction model is constructed as follows Figure 4 As shown. The entire model includes an embedding layer, a hidden layer, a feature fusion layer and an output layer. A loss function is constructed in the model. The specific loss function is constructed based on the predicted square resistance data and the sample square resistance data. Specifically, the mean square error of the predicted square resistance data and the sample square resistance data can be used as the loss function. In the entire model, the embedding layer is used to input all normalized features into the feature embedding layer (fully connected layer or 1D convolution layer) to extract features. The hidden layer is used to further process features using a convolutional neural network (CNN) or a multi-layer perceptron (MLP) as a hidden layer. The feature fusion layer can consider the association between the temperature zone and the sample boat in the boron diffusion equipment, construct the association between the temperature zone features and the sample boat square resistance, and introduce an attention mechanism. Figure 4 The black solid line part shown in the figure has a higher weight, while the black dotted line part has a lower weight.

[0123] S400: Based on the loss function, the initial diffusion mass prediction model is trained and optimized using a model optimization algorithm until a preset training end condition is reached to obtain a trained diffusion mass prediction model.

[0124] For the initial diffusion quality prediction model constructed by S300, a model optimization algorithm is used to perform training optimization around the loss function until the preset training end condition is reached. Specifically, the model weight can be adjusted by the Adam optimization algorithm. The preset training end condition here can be that the minimum loss function is obtained after iterative training to a preset number of times or multiple iterative calculations, that is, the diffusion quality prediction model obtained by training at this time can enter the input formula parameters to accurately predict the corresponding diffusion quality. Further, the diffusion quality here can be specifically characterized by the square resistance mean and square resistance uniformity.

[0125] S500: Generate initial optimal recipe parameters and adjust key process parameter characteristics in real time based on the trained diffusion quality prediction model.

[0126] The trained diffusion quality prediction model obtained based on S400 can accurately predict the boron diffusion quality corresponding to different boron diffusion formula parameters. The initial optimal formula parameters are first generated, and then in the actual boron diffusion process, the boron diffusion formula parameters are continuously optimized to calculate the boron diffusion quality corresponding to these boron diffusion formula parameters, and the boron diffusion formula parameters corresponding to the optimal boron diffusion quality are found. The key process parameter characteristics are accurately adjusted according to the production process to ensure the quality of boron diffusion.

[0127] The above-mentioned boron diffusion formula parameter optimization method obtains the sample data of boron diffusion, the sample square resistance data corresponding to the sample data, and the formula parameter characteristics corresponding to the sample data; generates the process parameter characteristics based on the formula parameter characteristics and the boron diffusion process calculation formula; inputs the formula parameter characteristics and the process parameter characteristics into the initial diffusion quality prediction model, and constructs a loss function based on the predicted square resistance data and the sample square resistance data output by the initial diffusion quality prediction model; uses the model optimization algorithm to train and optimize the initial diffusion quality prediction model based on the loss function until the preset training end condition is reached to obtain the trained diffusion quality prediction model; generates the initial optimal formula parameters and controls the key process parameter characteristics in real time according to the trained diffusion quality prediction model. In the whole process, the sample square resistance data and process parameter characteristics are obtained for the sample data of boron diffusion, and the loss function is constructed around the square resistance data. Finally, the trained diffusion quality prediction model that characterizes the process parameter characteristics and square resistance data (boron diffusion quality) is trained, and the initial optimal formula parameters are accurately generated and the key process parameter characteristics are controlled in real time through the trained diffusion quality prediction model.

[0128] In one embodiment, obtaining sample data of boron diffusion includes:

[0129] Obtain historical data of boron diffusion; extract keywords from the historical data, the keywords include equipment number, tube number and boat number; splice the data corresponding to the same keyword to generate sample data.

[0130] First, a large amount of historical data is collected, in which the equipment number, pipe number, and boat number constitute the key words, and the detection time is used as an auxiliary. The data with the same key words are spliced ​​together as a sample data (the data with a detection time difference of more than 2 hours is regarded as another set of data).

[0131] In one embodiment, the sample square resistance data corresponding to the sample data includes:

[0132] The square resistance value of the test silicon wafer is extracted from the sample data based on the preset square resistance sampling rule; the square resistance mean value and uniformity value of the test silicon wafer are calculated according to the square resistance value of the test silicon wafer; the square resistance mean value and the square resistance uniformity value are aggregated to obtain the sample square resistance data.

[0133] In this example, the square resistance data includes two dimensions of square resistance mean value and square resistance uniformity value, which can more accurately characterize the quality of the entire boron diffusion. The specific calculation process and method of square resistance mean value and square resistance uniformity value have been introduced in detail before and will not be repeated here.

[0134] In one embodiment, the method further comprises:

[0135] According to the sample data, the correlation between the medium temperature zone of the boron diffusion equipment and the small boat is obtained; based on the correlation between the medium temperature zone of the boron diffusion equipment and the small boat, the correlation between the medium temperature zone of the boron diffusion and the square resistance of the small boat is constructed; according to the correlation between the temperature zone and the square resistance of the small boat, the attention mechanism is introduced to generate the initial diffusion quality prediction model.

[0136] In this embodiment, an attention mechanism is introduced to generate an initial diffusion quality prediction model. Specifically, as shown in 4, the association between the temperature zone and the sampling boat in the boron diffusion equipment is considered, the association between the temperature zone characteristics and the sampling boat square resistance is constructed, and the attention mechanism is introduced. Figure 4 The black solid line part shown in the figure has a higher weight, while the black dotted line part has a lower weight.

[0137] In one embodiment, the initial diffusion mass prediction model is trained and optimized using a model optimization algorithm based on a loss function until a preset training end condition is reached, and the trained diffusion mass prediction model is obtained, including:

[0138] Based on the loss function, the initial diffusion mass prediction model is trained using the Adam optimization algorithm until the minimum value of the loss function is obtained, thereby obtaining the trained diffusion mass prediction model.

[0139] Adam (Adaptive Moment Estimation) is a widely used optimization algorithm that combines the ideas of RMSprop and Momentum. The Adam algorithm designs independent adaptive learning rates for different parameters by calculating the first-order moment estimate and the second-order moment estimate of the gradient. This adaptive learning rate adjustment method enables the Adam algorithm to converge faster in most cases and effectively handle non-stationary objectives and non-convex optimization problems. Specifically, the entire training process includes the following: Initialization: First, initialize the parameters of the diffusion mass prediction model (such as the weights and biases of the neural network) and the parameters of the Adam optimizer (such as learning rate, β1, β2, etc.). Forward propagation: Input the training data into the initial model, calculate the predicted value, and calculate the loss function value. Backward propagation: Based on the loss function value, the gradient of the model parameters is calculated by the chain rule. Parameter update: Use the Adam optimization algorithm to update the model parameters according to the gradient. The Adam algorithm adjusts the learning rate of each parameter according to the first-order moment estimate and the second-order moment estimate of the gradient, and performs parameter update. Iteration: Repeat the process of forward propagation, backpropagation, and parameter update until the preset training end condition is reached. This condition can be that the loss function value drops below a certain threshold, the number of training rounds reaches a preset value, or the performance on the validation set no longer improves.

[0140] like Figure 5 As shown, in one embodiment, S500 includes:

[0141] S510: Based on each type of boron diffusion recipe parameters, set the search range and step size.

[0142] S520: According to the search range and step size of each boron diffusion formula parameter, a plurality of boron diffusion formula parameters are randomly generated and encoded into a plurality of individuals to generate an initial population.

[0143] S530: using the trained diffusion quality prediction model to perform diffusion quality prediction on the boron diffusion formula parameters corresponding to the variant population, obtaining a diffusion quality prediction result, and transforming and quantifying the result into a fitness value.

[0144] S540: Based on the fitness value of each individual in the population, the population is selected, crossed, and mutated to obtain a new generation of population.

[0145] S550: Repeat iterative population diffusion quality prediction, fitness calculation, population selection, crossover, and mutation operations, and continuously evolve so that the fitness reaches a preset iteration stop condition; here, the preset iteration stop condition includes that the fitness after evolution obtains a specified value, or the number of iterations reaches a specified number.

[0146] S560: Query the boron diffusion formula parameters corresponding to the individual with the best diffusion quality in the diffusion quality prediction results, and generate initial optimal formula parameters, wherein the diffusion quality is characterized based on the square resistance data.

[0147] After obtaining the diffusion quality prediction model, the optimal process formula parameters are obtained through genetic algorithm (GA) or other parameter optimization algorithms. The entire formula parameter optimization algorithm process is as follows: Figure 6 As shown. Specifically, it includes the following stages: 1) Individual encoding: set a reasonable search range and step size for each input parameter (such as the temperature of temperature zone 1 is between 850-950, and the step size is 1), then encode all parameters into individuals, and generate an initial population containing multiple individuals (such as 100). 2) Design fitness function: Based on the trained diffusion quality prediction model, the parameter value corresponding to the input individual is predicted to predict the mean and uniformity value of its diffusion square resistance, and then the difference between the predicted value and the target value is quantified through transformation into a fitness function (such as the inverse of the error), so that the higher the fitness, the better the parameter corresponding to the individual. 3) Genetic algorithm: Generate a new generation of populations through selection, crossover and mutation operations, and continuously evolve to find the optimal solution. Finally, generate optimized formula parameters. By generating a specific diffusion quality prediction model for the historical data of different production lines, personalized formula parameter recommendations can be provided according to different production lines.

[0148] In one embodiment, generating initial optimal recipe parameters and real-time control of key process parameter characteristics based on the trained diffusion quality prediction model further includes:

[0149] Based on the trained diffusion quality prediction model, the correlation between the adjustable process parameter characteristics and the diffusion quality is obtained, and the adjustable process parameter characteristics include temperature, boron source flow rate and diffusion time; according to the correlation between the process parameter characteristics and the diffusion quality, the adjustment threshold of each process parameter characteristic for the corresponding diffusion square resistance is obtained; according to the adjustment threshold of each process parameter characteristic, the boron diffusion process is controlled in real time according to the order of the adjustable process parameter characteristics.

[0150] After establishing the correlation between process parameters and diffusion quality, the next step is to determine the adjustment thresholds of each process parameter. These thresholds define the range in which the parameters can be safely adjusted to ensure that the diffusion quality remains within an acceptable level. Finally, the boron diffusion process is regulated in real time or periodically based on the determined adjustment thresholds of each process parameter.

[0151] Specifically, in practical applications, although the parameter optimization algorithm provides an effective combination of recipe parameters, real-time environmental changes and raw material differences need to be considered in actual production. Therefore, it is necessary to collect process parameters and data in real time to ensure the stability of silicon wafer quality.

[0152] The collected data is similar to the historical data collected above. The key parameters include the temperature of temperature zones 1-6, the boron source flow rate, and the diffusion time. The diffusion quality is represented by the square resistance mean and uniformity of zones 1, 3, 5, 7, 10, and 12. Unlike Section 2.1, the boat number is not required as a keyword. Instead, the equipment number and tube number are used as keywords, and the detection time is used as an auxiliary. Arrange the data with the same keyword in chronological order, and take 3-5 consecutive rounds of data as a sample. Sample examples are shown in Table 4 below (the target value should be consistent).

[0153] Table 4

[0154]

[0155] Based on the trained diffusion quality prediction model, the correlation between the characteristic parameter value and the diffusion quality can be obtained. By fixing other parameters, changing one of the parameters and then inputting it into the model, the correlation between a single characteristic parameter and the diffusion quality can be obtained, so as to calculate the adjustment threshold thr of a single parameter feature when the diffusion quality index (such as square resistance) needs to be adjusted to the target value. After the threshold is known, the calculation formula for the parameter adjustment amount v is as follows:

[0156]

[0157] Among them, R t Refers to the target value of square resistance, R n Refers to the actual value of the current square resistance. Here, the thresholds that need to be determined by the model include the direct control threshold of the square resistance corresponding to the temperature, the indirect control threshold of the square resistance corresponding to the temperature (silicon wafer at the edge of the adjacent temperature zone), the control threshold of the square resistance corresponding to the boron source flow, and the control threshold of the square resistance corresponding to the diffusion time.

[0158] (2) Handling of null values ​​and outliers

[0159] First, check whether there are null values. If there are null values, set them to 0 and treat them as outliers in the future. Then, set the range of moderate and severe outliers to determine whether there are outliers. The processing rules include: a. If there are multiple severe outliers, it is determined that there is a problem with the current data and further observation is required; b. If there is only one severe outlier, determine whether the parameters need to be adjusted based on the values ​​of the other rounds. If the square resistance of the other rounds is significantly different, further observation is performed, otherwise only the outliers are removed; c. If there are moderate outliers but no severe outliers, and the square resistance of multiple rounds is significantly different with no obvious trend, further observation is performed, otherwise all values ​​are retained.

[0160] (3) Temperature control

[0161] From the corresponding relationship between square resistance and temperature zone, it can be seen that the direct influence relationship between square resistance and temperature zone is: square resistance 1-temperature zone 2, square resistance 3-temperature zone 2, square resistance 5-temperature zone 3, square resistance 7-temperature zone 4, square resistance 10-temperature zone 5, square resistance 12-temperature zone 5, and the indirect influence of square resistance and temperature zone includes: square resistance 1-temperature zone 1, square resistance 3-temperature zone 3, square resistance 7-temperature zone 3, square resistance 10-temperature zone 4, square resistance 12-temperature zone 6.

[0162] a. According to the difference between the average value of multiple rounds of square resistance and the target boat square resistance in different areas, as well as the control thresholds of direct and indirect influence, the control values ​​of temperature zones 1 to 6 can be calculated (it is necessary to determine which square resistance values ​​should be taken as the square resistance values ​​required for the final calculation based on the change of square resistance and the engineer's process experience).

[0163] b. After adjustment, check whether the adjusted temperature is still within the range limit and whether the temperature difference between different temperature zones exceeds the normal range.

[0164] (4) Boron source flow control

[0165] Temperature control is given priority. When the effect of temperature control is not obvious, the boron source flow rate is regulated according to the real-time collected data.

[0166] a. According to the average difference between the square resistances of multiple regions and the target square resistances, and the control threshold of the square resistance corresponding to the boron source flow rate, the boron source flow rate is controlled.

[0167] b. After adjustment, check whether the boron source flow rate is within the range limit.

[0168] (5) Diffusion time

[0169] When the effects of temperature control and boron source flow control are still not obvious, the diffusion time is controlled according to the real-time collected data.

[0170] a. According to the average difference between the square resistances of multiple regions and the target square resistance, and the control threshold of the square resistance corresponding to the diffusion time, the diffusion time is controlled.

[0171] b. After adjustment, check whether the diffusion time is within the range limit.

[0172] Results comparison

[0173] After the real-time precise parameter control system was activated on site, the average value of the diffusion square resistance was significantly improved, and the square resistance uniformity was also improved to a certain extent, taking furnace No. 1 as an example. Figure 7As shown in the figure below, (1) the mean square resistance comparison is shown in the figure below (target value 140, black line is the AI ​​control time node). The mean change of furnace No. 1 before and after adjustment is 140.19→140.01, the proportion of ±5 is 91.56%→96.18%, and the proportion of ±10 is 98.58%→99.55%. The mean uniformity change is 4.77→4.50.

[0174] In general, the boron diffusion formula parameter optimization method of the present application in the above embodiments has the following significant effects:

[0175] 1. Use deep learning and a large amount of historical data to model the complex nonlinear relationship between key parameters of the boron diffusion process (such as gas flow, diffusion time, temperature in the temperature zone, etc.) and diffusion quality indicators (such as square resistance mean and uniformity) to quickly and accurately predict process results.

[0176] 2. Use genetic algorithms to search for the optimal combination of process recipe parameters in multidimensional parameter space to avoid falling into local optimality, thereby finding the global optimal recipe parameters and improving the diffusion quality of silicon wafers.

[0177] 3. Based on the diffusion quality prediction model, analyze the correlation between each parameter and the average square resistance of the diffusion quality indicator, and determine the control threshold. Combined with the process experience of engineers, adjust key parameters in real time during the production process to adapt to changes in the production environment and material properties to ensure stable product quality.

[0178] 4. The automatic control system has the following advantages: (1) strong timeliness; (2) fast calculation speed; (3) automatic calculation and automatic control; (4) combining big data analysis with engineer experience to reduce errors caused by experience differences and reduce the risk of errors and misoperations.

[0179] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0180] Based on the same inventive concept, the embodiment of the present application also provides a boron diffusion formula parameter optimization device for implementing the boron diffusion formula parameter optimization method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more boron diffusion formula parameter optimization device embodiments provided below can refer to the limitations of the boron diffusion formula parameter optimization method above, and will not be repeated here.

[0181] In one embodiment, Figure 8 As shown, a boron diffusion recipe parameter optimization device is provided, comprising:

[0182] The data acquisition module 100 is used to acquire sample data of boron diffusion, wherein the sample data includes sample sheet resistance data and corresponding recipe parameter characteristics;

[0183] A process parameter feature acquisition module 200 is used to generate process parameter features based on recipe parameter features and a boron diffusion process calculation formula;

[0184] A model initialization module 300 is used to input the recipe parameter characteristics and the process parameter characteristics into the initial diffusion mass prediction model, and to construct a loss function based on the predicted square resistance data and the sample square resistance data output by the initial diffusion mass prediction model;

[0185] A model training module 400 is used to train and optimize the initial diffusion mass prediction model using a model optimization algorithm based on a loss function until a preset training end condition is reached to obtain a trained diffusion mass prediction model;

[0186] The optimization and control module 500 is used to generate initial optimal recipe parameters and perform real-time control of key process parameter characteristics based on the trained diffusion quality prediction model.

[0187] In one embodiment, the data acquisition module 100 is also used to acquire historical data of boron diffusion; extract keywords from the historical data, the keywords include equipment number, tube number and boat number; and concatenate data corresponding to the same keywords to generate sample data.

[0188] In one of the embodiments, the data acquisition module 100 is also used to extract the square resistance value of the test silicon wafer from the sample data based on the preset square resistance sampling rules; calculate the square resistance uniformity value and uniformity value of the test silicon wafer according to the square resistance value of the test silicon wafer; and aggregate the square resistance mean value and the square resistance uniformity value to obtain the sample square resistance data.

[0189] In one embodiment, the model initialization module 300 is also used to obtain the correlation between the medium temperature zone and the boat of the boron diffusion equipment based on the sample data; based on the correlation between the medium temperature zone and the boat of the boron diffusion equipment, construct the correlation between the medium temperature zone of the boron diffusion and the square resistance of the boat; introduce the attention mechanism according to the correlation between the temperature zone and the square resistance of the boat to generate an initial diffusion quality prediction model.

[0190] In one embodiment, the model training module 400 is further used to train the initial diffusion mass prediction model based on the loss function and using the Adam optimization algorithm until the minimum value of the loss function is obtained to obtain the trained diffusion mass prediction model.

[0191] In one embodiment, the optimization and control module 500 is also used to set the search range and step size based on each type of boron diffusion formula parameter; randomly generate multiple boron diffusion formula parameters and encode them into multiple individuals according to the search range and step size of each boron diffusion formula parameter to generate an initial population; use the trained diffusion quality prediction model to predict the diffusion quality of the boron diffusion formula parameters corresponding to the variant population to obtain the diffusion quality prediction results, and transform and quantify them into fitness values; based on the fitness value of each individual in the population, select, crossover, and mutate the population to obtain a new generation of population; repeat iterative population diffusion quality prediction, fitness calculation, population selection, crossover, and mutation operations, and continuously evolve to reach a preset iteration stop condition; query the boron diffusion formula parameters corresponding to the individual with the best diffusion quality in the diffusion quality prediction results to generate the initial optimal formula parameters, wherein the diffusion quality is characterized based on square resistance data.

[0192] In one embodiment, the optimization and control module 500 is also used to obtain the correlation between adjustable process parameter characteristics and diffusion quality based on the trained diffusion quality prediction model, and the adjustable process parameter characteristics include temperature, boron source flow rate and diffusion time; according to the correlation between the process parameter characteristics and the diffusion quality, the adjustment threshold of each process parameter characteristic for the corresponding diffusion square resistance is obtained; according to the adjustment threshold of each process parameter characteristic, the boron diffusion process is controlled in real time according to the order of the adjustable process parameter characteristics.

[0193] Each module in the above-mentioned boron diffusion formula parameter optimization device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0194] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig. 9As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store preset data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for optimizing boron diffusion formula parameters is implemented.

[0195] Those skilled in the art will understand that Fig. 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0196] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above-mentioned boron diffusion recipe parameter optimization method when executing the computer program.

[0197] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned boron diffusion recipe parameter optimization method is implemented.

[0198] In one embodiment, a computer program product is provided, comprising a computer program, which implements the above-mentioned boron diffusion recipe parameter optimization method when executed by a processor.

[0199] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited thereto.

[0200] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0201] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for optimizing boron diffusion formula parameters, characterized in that: The method comprises: Acquire sample data of boron diffusion, wherein the sample data includes sample sheet resistance data and corresponding recipe parameter characteristics; Generate process parameter characteristics based on the recipe parameter characteristics and the boron diffusion process calculation formula; Inputting the recipe parameter characteristics and the process parameter characteristics into an initial diffusion mass prediction model, and constructing a loss function based on the predicted square resistance data output by the initial diffusion mass prediction model and the sample square resistance data; Based on the loss function, the initial diffusion mass prediction model is trained and optimized using a model optimization algorithm until a preset training end condition is reached to obtain a trained diffusion mass prediction model; Initial optimal recipe parameters are generated and key process parameter characteristics are controlled in real time based on the trained diffusion quality prediction model.

2. The method according to claim 1, characterized in that Sample data for obtaining boron diffusion include: Obtain historical data on boron diffusion; Extracting keywords from the historical data, the keywords including equipment number, pipe number and boat number; Concatenate the data corresponding to the same keyword to generate sample data.

3. The method according to claim 1, characterized in that The sample square resistance data corresponding to the sample data includes: Extracting the square resistance value of the test silicon wafer from the sample data based on a preset square resistance sampling rule; Calculating the square resistance average and uniformity value of the test silicon wafer according to the square resistance value of the test silicon wafer; The square resistance mean value and the square resistance uniformity value are aggregated to obtain sample square resistance data.

4. The method according to claim 2, characterized in that: The method further comprises: According to the sample data, obtaining the correlation between the temperature zone of the boron diffusion device and the boat; Based on the relationship between the medium temperature zone of the boron diffusion equipment and the boat, a relationship between the medium temperature zone of the boron diffusion and the square resistance of the boat is established; An attention mechanism is introduced according to the relationship between the temperature zone and the boat square resistance to generate an initial diffusion quality prediction model.

5. The method according to claim 1, characterized in that Based on the loss function, the initial diffusion mass prediction model is trained and optimized using a model optimization algorithm until a preset training end condition is reached, and the trained diffusion mass prediction model is obtained, including: Based on the loss function, the initial diffusion mass prediction model is trained by using the Adam optimization algorithm until the minimum value of the loss function is obtained, thereby obtaining a trained diffusion mass prediction model.

6. The method according to claim 1, characterized in that Generating initial optimal recipe parameters and real-time control of key process parameter characteristics according to the trained diffusion quality prediction model include: Based on each type of boron diffusion recipe parameters, set the search range and step size; According to the search range and step size of each boron diffusion formula parameter, a plurality of boron diffusion formula parameters are randomly generated and encoded into a plurality of individuals to generate an initial population; Using the trained diffusion quality prediction model to predict the diffusion quality of the boron diffusion formula parameters corresponding to the variant population, obtaining a diffusion quality prediction result, and transforming and quantifying the result into a fitness value; Based on the fitness value of each individual in the population, the population is selected, crossed, and mutated to obtain a new generation of population; Repeating the population diffusion quality prediction, fitness calculation, population selection, crossover, and mutation operations, and continuously evolving to reach a preset iteration stop condition; The boron diffusion formula parameters corresponding to the individual with the best diffusion quality in the diffusion quality prediction result are queried to generate initial optimal formula parameters, wherein the diffusion quality is characterized based on the square resistance data.

7. The method according to claim 1, characterized in that The generating of initial optimal recipe parameters and real-time control of key process parameter characteristics according to the trained diffusion quality prediction model also includes: Based on the trained diffusion quality prediction model, obtaining the correlation between adjustable process parameter characteristics and diffusion quality, wherein the adjustable process parameter characteristics include temperature, boron source flow rate, and diffusion time; According to the correlation between the process parameter characteristics and the diffusion quality, obtaining the adjustment threshold of each process parameter characteristic; According to the adjustment thresholds of the various process parameter characteristics, the boron diffusion process is controlled in real time in the order of the adjustable process parameter characteristics.

8. A boron diffusion recipe parameter optimization device, characterized in that: The device comprises: A data acquisition module, used to acquire sample data of boron diffusion, wherein the sample data includes sample sheet resistance data and corresponding recipe parameter characteristics; A process parameter feature acquisition module, used to generate process parameter features based on the recipe parameter features and the boron diffusion process calculation formula; A model initialization module, used for inputting the recipe parameter characteristics and the process parameter characteristics into an initial diffusion mass prediction model, and constructing a loss function based on the predicted square resistance data output by the initial diffusion mass prediction model and the sample square resistance data; A model training module, used to train and optimize the initial diffusion mass prediction model using a model optimization algorithm based on the loss function until a preset training end condition is reached to obtain a trained diffusion mass prediction model; The optimization and control module is used to generate initial optimal formula parameters and control key process parameter characteristics in real time according to the trained diffusion quality prediction model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images