Polycrystalline silicon wafer surface coating uniformity detection system based on machine learning
Through the polycrystalline silicon wafer surface coating uniformity detection system based on machine learning, the neural network model is used to predict coating uniformity, the problem of high detection cost in the existing technology is solved, the efficiency of evaluation of the uniformity of the finished silicon wafer is improved, and the production cost is reduced.
Patent Information
- Application Number
- CN202410945516.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-15
- Publication Date
- 2024-11-08
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art has high cost and time cost in the detection of surface coating uniformity of polycrystalline silicon wafers, resulting in low uniformity pass rate of silicon wafer finished products, affecting the production efficiency of solar cells.
Using a polycrystalline silicon wafer surface coating uniformity detection system based on machine learning, we can build a neural network model by obtaining the material, membrane properties and equipment characteristic parameters of silicon wafer samples, predicting the coating uniformity characterization value, and achieving rapid uniformity of silicon wafer finished products Evaluate.
It reduces the detection cost, improves the efficiency of uniformity evaluation of finished silicon wafers, reduces the furnace recovery rate, and reduces the production cost of polycrystalline silicon wafers.
Smart Images

Figure CN118918991A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of polycrystalline silicon wafer performance detection, and in particular to a polycrystalline silicon wafer surface coating uniformity detection system based on machine learning. Background Art
[0002] Polycrystalline silicon wafers are mostly used in electronics and solar power generation, and are an important raw material in the solar industry chain. Polycrystalline silicon cells occupy a large share of the photovoltaic market due to their low cost. Silicon wafer coating is a relatively important process in the production process of solar cells. By coating silicon wafers, the light absorption capacity of solar cells can be increased, the conversion efficiency of cells can be improved, and polycrystalline silicon wafers have good optical and electrical properties. The commonly used coating equipment for coating the surface of polycrystalline silicon wafers is plasma enhanced chemical vapor deposition (PECVD) equipment. Due to the large differences in lattice orientation, velvet characteristics, surface defects, and even reflectivity between different grains on the surface of polycrystalline silicon wafers, the uniformity of the SiNx film prepared on its surface is poor.
[0003] At present, the detection of the uniformity of the coating on the surface of multi-crystalline silicon wafers is mainly carried out by testing the refractive index of the coating film layer, and the statistical parameters of the refractive index, such as variance, standard deviation, etc., are used to characterize the uniformity of the coating. If the variance value of the measured refractive index is larger, it indicates that the coating uniformity is worse. The above coating uniformity detection method is carried out for finished silicon wafers. In the actual production process, since silicon wafers are processed in the form of assembly line operations, when the product uniformity qualification rate does not meet the requirements during the sampling inspection of finished products, it means that the batch of silicon wafers cannot be used for the manufacture of solar cells and can only be used as waste to re-melt ingots. If the uniformity of the products is tested on a large scale to reduce the remelting rate, it will lead to an increase in the detection cost and time cost, which will increase the production cost of multi-crystalline silicon wafers. To this end, we propose a multi-crystalline silicon wafer surface coating uniformity detection system based on machine learning. Summary of the invention
[0004] The main purpose of the present invention is to provide a polysilicon wafer surface coating uniformity detection system based on machine learning, which can effectively solve the problems in the background technology.
[0005] In order to achieve the above object, the technical solution adopted by the present invention is:
[0006] The machine learning-based multi-crystalline silicon wafer surface coating uniformity detection system includes:
[0007] A silicon wafer parameter acquisition module for acquiring material characteristic parameters of silicon wafer samples before coating and film characteristic parameters after coating;
[0008] A coating equipment parameter acquisition module for acquiring equipment characteristic parameters of silicon wafer sample coating processing equipment;
[0009] A processing parameter acquisition module for acquiring environmental characteristic parameters during the coating process of a silicon wafer sample by a coating processing device;
[0010] A data processing module for preprocessing the acquired characteristic parameters, the data processing module is also used to construct a data set D=(d i , d j , d u , d v ), where d i Expressed as the value of the i-th material characteristic parameter, d j Expressed as the jth film characteristic parameter, d u It is represented by the value of the characteristic parameter of the u-th device, d v It is expressed as the vth environmental characteristic parameter, i / j / u / v=1,2,3,...;
[0011] A coating uniformity characterization module for measuring and obtaining the refractive index of a silicon wafer sample after coating processing, wherein the coating uniformity characterization module is further used to calculate the variance mean of the measured refractive index as a coating uniformity characterization value of the silicon wafer sample;
[0012] A correlation analysis module for calculating the correlation coefficient between characteristic parameters and coating uniformity characterization values;
[0013] A neural network model unit module is used to construct a neural network model with characteristic parameter values as input parameters and coating uniformity characterization values as output parameters, and the mapping relationship between the characteristic parameter values and the coating uniformity characterization values is output through the constructed neural network model. The expression of the mapping relationship is:
[0014]
[0015] In the formula, f represents the characteristic parameter value d α The mapping relationship to the coating uniformity characterization value Uc;
[0016] The characteristic parameters of the silicon wafer to be tested are input into the mapping relationship to obtain the coating uniformity characterization value of the silicon wafer to be tested.
[0017] Further, the material characteristic parameters include at least one of the following: the coating area of the silicon wafer, the proportion of the defect area of the shape defect of the silicon wafer surface, the proportion of the defect area of the color defect of the silicon wafer surface, and the proportion of the defect area of the texture defect of the silicon wafer surface;
[0018] The film characteristic parameters include at least one of film thickness, film density, and film compactness;
[0019] The equipment characteristic parameters include at least one of the temperature control parameters, vacuum parameters, flow control parameters, and plasma system parameters of the processing equipment;
[0020] The environmental characteristic parameters include at least one of the temperature value, humidity value, air pressure value, and air cleanliness level of the environment surrounding the silicon wafer during the coating process.
[0021] Further, the processing equipment is a plasma enhanced chemical vapor deposition system;
[0022] The temperature control parameters include at least one of the maximum operating temperature of the deposition system, heating power, and temperature control accuracy;
[0023] The vacuum parameter is the vacuum degree of the vacuum subsystem of the deposition system when it is working;
[0024] The flow control parameter includes at least one of a leakage rate, a resolution, a response time gas characteristic, and a response time electrical characteristic of the deposition system;
[0025] The plasma system parameters include at least one of power output, signal frequency, reflected power, power stability, harmonic components, and supply voltage of a plasma subsystem of the deposition system.
[0026] Furthermore, the weights of the neural network model input parameters are determined according to the correlation coefficient with the coating uniformity characterization value and satisfy the following relationship:
[0027]
[0028] In the formula, ω k is the weight of the kth feature parameter, K is the type of feature parameter, and K≤i+j+u+v; r k is the correlation coefficient between the kth characteristic parameter and the coating uniformity characterization value, r k The value range is (-1,1).
[0029] Furthermore, the calculation formula of the correlation coefficient between the characteristic parameters and the coating uniformity characterization is:
[0030]
[0031] In the formula, h kq is the qth data value of the kth feature parameter; is the mean value of the kth characteristic parameter; Uc q is the qth data value of the coating uniformity characterization value; is the mean of the coating uniformity characterization value; Q is the total amount of data.
[0032] Furthermore, the neural network model is a three-layer structure of input layer-hidden layer-output layer, wherein the number of neurons in the input layer is equal to the number of types of characteristic parameters, the number of neurons in the output layer is equal to the number of types of output parameters, and the number of neurons in the hidden layer is determined according to an empirical formula, which is:
[0033]
[0034] Where s is the number of neurons in the hidden layer; m is the number of neurons in the input layer, n is the number of neurons in the output layer, and a is an integer from 1 to 9. In this model, m = k and n = 1.
[0035] The present invention has the following beneficial effects:
[0036] Compared with the prior art, the technical solution of the present invention pre-processes the silicon wafer sample by obtaining material characteristic parameters before coating, film characteristic parameters after coating, equipment characteristic parameters of the coating equipment of the silicon wafer sample and environmental characteristic parameters of the coating equipment during the coating process, constructs a data set according to the pre-processed data, measures and obtains the refractive index of the silicon wafer sample after coating, calculates the variance mean of the measured refractive index as the coating uniformity characterization value of the silicon wafer sample, calculates the correlation coefficient between the characteristic parameter and the coating uniformity characterization value, constructs a neural network model with the characteristic parameter value as an input parameter and the coating uniformity characterization value as an output parameter, outputs a mapping relationship between the characteristic parameter value and the coating uniformity characterization value through the constructed neural network model, obtains the coating uniformity characterization value of the silicon wafer to be tested according to the mapping relationship, constructs a neural network model by collecting the characteristic parameter values in the production process as input, predicts the coating uniformity characterization value, and can evaluate the uniformity of the silicon wafer before the finished product is off the line, thereby reducing the detection cost. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 It is an overall structural block diagram of the polycrystalline silicon wafer surface coating uniformity detection system based on machine learning of the present invention;
[0038] Figure 2 This is a workflow diagram of the polysilicon wafer surface coating uniformity detection system based on machine learning of the present invention;
[0039] Figure 3 It is a structural diagram of the neural network model;
[0040] Figure 4 Schematic diagram of the refractive index measurement system for silicon wafer samples. DETAILED DESCRIPTION
[0041] The present invention will be further described below in conjunction with specific implementation methods, wherein the accompanying drawings are only used for exemplary descriptions and represent only schematic diagrams rather than actual drawings, and should not be understood as limiting the present invention. In order to better illustrate the specific implementation methods of the present invention, some parts of the accompanying drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product.
[0042] Example 1
[0043] like Figure 1-4 As shown, the specific implementation process of the technical solution of the present invention includes the following steps:
[0044] Step 1: Analyze the factors that affect the uniformity of the coating on the surface of the multi-crystalline silicon wafer, mainly including three types of factors: coating equipment factors, silicon wafer factors and processing environment factors. According to the analysis results, obtain the historical data of the coating processing of the silicon wafer samples, including the material characteristic parameters of the silicon wafer samples before coating, the film characteristic parameters after coating, the equipment characteristic parameters of the coating processing equipment of the silicon wafer samples, and the environmental characteristic parameters of the coating processing equipment during the coating process of the silicon wafer samples, among which:
[0045] The material characteristic parameters include the silicon wafer coating area, the silicon wafer surface shape defect area ratio, the silicon wafer surface color defect area ratio, and the silicon wafer surface texture defect area ratio;
[0046] The film characteristic parameters include film thickness, film density, and film compactness;
[0047] The equipment characteristic parameters include the temperature control parameters, vacuum parameters, flow control parameters, and plasma system parameters of the processing equipment;
[0048] Processing equipment plasma enhanced chemical vapor deposition system, there are
[0049] Temperature control parameters include the maximum temperature of the deposition system, heating power, and temperature control accuracy;
[0050] The vacuum parameter is the vacuum degree when the vacuum subsystem of the deposition system is working;
[0051] The flow control parameters include the deposition system’s leak rate, resolution, response time gas characteristics, and response time electrical characteristics;
[0052] Plasma system parameters include power output, signal frequency, reflected power, power stability, harmonic components, and supply voltage of the plasma subsystem of the deposition system;
[0053] Environmental characteristic parameters include the temperature, humidity, air pressure, and air cleanliness level of the environment surrounding the silicon wafer during the coating process;
[0054] In this embodiment, we take the silicon wafer coating area, the silicon wafer surface shape defect area ratio in the material characteristic parameters, the film thickness in the film characteristic parameters, and the vacuum degree of the deposition system vacuum subsystem in the equipment characteristic parameters as examples to illustrate this solution;
[0055] Step 2: Preprocess the acquired characteristic parameters through the data processing module, and construct a data set D = (d1, d2, d3, d4) based on the preprocessed data, where d1 is the silicon wafer coating area value of the silicon wafer sample, and d2 is the silicon wafer surface shape defect area ratio of the silicon wafer sample. The calculation formula is: s d / s, where s d is the defect area of the silicon wafer sample surface shape defect, s is the coating area of the silicon wafer sample, d3 is the film thickness of the wafer sample, and d4 is the vacuum degree of the vacuum subsystem of the deposition system when it is working;
[0056] Step 3: Use an ellipsometer to test the refractive index of the coating layer of the silicon wafer sample. The measurement principle is as follows: Figure 4 As shown, the refractive index value Uc of the silicon wafer sample after coating is obtained. In the actual test process, multiple measurements can be performed to obtain sufficient data samples, and the variance mean of the measured refractive index is calculated as the coating uniformity characterization value of the silicon wafer sample;
[0057] Step 4: Calculate the correlation coefficient between the characteristic parameter and the coating uniformity characterization value. The calculation formula is:
[0058]
[0059] In the formula, h kq is the qth data value of the kth feature parameter; is the mean value of the kth characteristic parameter; Uc q is the qth data value of the coating uniformity characterization value; is the mean value of the coating uniformity characterization value; Q is the total amount of data. In this embodiment, the number of characteristic parameters k=4;
[0060] Step 5: Determine the influence weights ω1, ω2, ω3, and ω4 of the four characteristic parameters of silicon wafer coating area, silicon wafer surface shape defect area ratio, film thickness, and vacuum degree of the vacuum subsystem of the deposition system on the coating uniformity characterization value based on the calculation results of the relationship number, and the influence weights satisfy the following relationship:
[0061]
[0062] Determine the specific values of the influence weights ω1, ω2, ω3, and ω4 according to the above formula;
[0063] Step 6: Construct a neural network model with characteristic parameter values as input parameters and coating uniformity characterization values as output parameters, such as Figure 3 As shown in the figure, the neural network model is a three-layer structure of input layer-hidden layer-output layer, where the number of neurons in the input layer is equal to the number of types of characteristic parameters, the number of neurons in the output layer is equal to the number of types of output parameters, and the number of neurons in the hidden layer is determined according to an empirical formula, which is:
[0064]
[0065] Wherein, s is the number of neurons in the hidden layer; m is the number of neurons in the input layer, n is the number of neurons in the output layer, and a is an integer of 1-9. In this model, m=k, n=1. In this embodiment, a=4, m=k=4 are taken, then s=6, and the neural network model is a three-layer structure with 4 neurons in the input layer, 6 neurons in the hidden layer, and 1 neuron in the output layer. Taking the BP neural network toolbox of MATLAB as an example, the network is trained with the standard BP neural network algorithm, and part of the data in the data sample is selected as the test sample, and the remaining data is the training sample. The data volume of the general training sample accounts for no less than 70% of the total sample data volume. The target error and the maximum number of cycles are set, and the transfer function of the hidden layer neuron is selected as the S-type tangent function tans ig, the transfer function of the output layer neuron is selected as the S-type logarithmic function logs ig, and the tra ingdx function is selected as the training function. The model is constructed by training the neural network;
[0066] Step 7: The mapping relationship between the characteristic parameter value and the coating uniformity characterization value is output by the constructed neural network model. The expression of the mapping relationship is:
[0067]
[0068] In the formula, f represents the characteristic parameter value d α The mapping relationship between the coating uniformity characterization value Uc and the characteristic parameters of the silicon wafer to be tested is input into the mapping relationship to obtain the coating uniformity characterization value of the silicon wafer to be tested in the processing flow.
[0069] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.
Claims
1. A polysilicon wafer surface coating uniformity detection system based on machine learning, characterized in that: It includes silicon wafer parameter acquisition module, coating equipment parameter acquisition module, processing parameter acquisition module, data processing module, coating uniformity characterization module, correlation analysis module, and neural network model unit module; The silicon wafer parameter acquisition module is used to obtain material characteristic parameters of silicon wafer samples before coating and film characteristic parameters after coating; The coating equipment parameter acquisition module is used to acquire equipment characteristic parameters of silicon wafer sample coating processing equipment; The processing parameter acquisition module is used to obtain environmental characteristic parameters during the coating process of the coating processing equipment on the silicon wafer sample; The data processing module is used to preprocess the acquired characteristic parameters and construct a data set D=(d i , d j , d u , d v ), where d i Expressed as the value of the i-th material characteristic parameter, d j Expressed as the jth film characteristic parameter, d u It is represented by the value of the characteristic parameter of the u-th device, d v It is expressed as the vth environmental characteristic parameter, i / j / u / v=1,2,3,...; The coating uniformity characterization module is used to measure and obtain the refractive index of the silicon wafer sample after coating processing, and calculate the variance mean of the measured refractive index as the coating uniformity characterization value of the silicon wafer sample; The correlation analysis module is used to calculate the correlation coefficient between the characteristic parameter and the coating uniformity characterization value; The neural network model unit module is used to construct a neural network model with characteristic parameter values as input parameters and coating uniformity characterization values as output parameters, and the mapping relationship between the characteristic parameter values and the coating uniformity characterization values output by the constructed neural network model is expressed as follows: In the formula, f represents the characteristic parameter value d α The mapping relationship to the coating uniformity characterization value Uc; The characteristic parameters of the silicon wafer to be tested are input into the mapping relationship to obtain the coating uniformity characterization value of the silicon wafer to be tested.
2. The polysilicon wafer surface coating uniformity detection system based on machine learning according to claim 1 is characterized in that: The material characteristic parameters include at least one of the following: the coating area of the silicon wafer, the proportion of the defect area of the shape defect on the silicon wafer surface, the proportion of the defect area of the color defect on the silicon wafer surface, and the proportion of the defect area of the texture defect on the silicon wafer surface; The film characteristic parameters include at least one of film thickness, film density, and film compactness; The equipment characteristic parameters include at least one of the temperature control parameters, vacuum parameters, flow control parameters, and plasma system parameters of the processing equipment; The environmental characteristic parameters include at least one of the temperature value, humidity value, air pressure value, and air cleanliness level of the environment surrounding the silicon wafer during the coating process.
3. The polysilicon wafer surface coating uniformity detection system based on machine learning according to claim 1 is characterized in that: The weights of the neural network model input parameters are determined according to the correlation coefficient with the coating uniformity characterization value and satisfy the following relationship: In the formula, ω k is the weight of the kth feature parameter, K is the type of feature parameter, and K≤i+j+u+v; r k is the correlation coefficient between the kth characteristic parameter and the coating uniformity characterization value, r k The value range is (-1,1).
4. The polysilicon wafer surface coating uniformity detection system based on machine learning according to claim 1 is characterized in that: The calculation formula of the correlation coefficient between the characteristic parameters and the coating uniformity characterization is: In the formula, h kq is the qth data value of the kth feature parameter; is the mean value of the kth characteristic parameter; Uc q is the qth data value of the coating uniformity characterization value; is the mean of the coating uniformity characterization value; Q is the total amount of data.
5. The polysilicon wafer surface coating uniformity detection system based on machine learning according to claim 1 is characterized in that: The neural network model is a three-layer structure of input layer-hidden layer-output layer, wherein the number of neurons in the input layer is equal to the number of types of characteristic parameters, the number of neurons in the output layer is equal to the number of types of output parameters, and the number of neurons in the hidden layer is determined according to an empirical formula, which is: Where s is the number of neurons in the hidden layer; m is the number of neurons in the input layer, n is the number of neurons in the output layer, and a is an integer from 1 to 9. In this model, m = k and n = 1.
6. The polysilicon wafer surface coating uniformity detection system based on machine learning according to claim 2 is characterized in that: The processing equipment is a plasma enhanced chemical vapor deposition system; The temperature control parameters include at least one of the maximum operating temperature of the deposition system, heating power, and temperature control accuracy; The vacuum parameter is the vacuum degree when the vacuum subsystem of the deposition system is working; The flow control parameter includes at least one of a leakage rate, a resolution, a response time gas characteristic, and a response time electrical characteristic of the deposition system; The plasma system parameters include at least one of power output, signal frequency, reflected power, power stability, harmonic components, and supply voltage of a plasma subsystem of the deposition system.
