A method and system for monitoring the execution of a non-conductive vacuum metallization (NCVM) process

Abstract

The application discloses an NCVM non-conductive vacuum coating process execution monitoring method and system, relates to the related field of vacuum coating, and comprises the following steps: carrying out dynamic grid division on a detection unit for a substrate surface; performing real-time scanning on the adaptive grid substrate surface by using an infrared thermal imager to obtain a substrate dynamic temperature distribution matrix; predicting the substrate dynamic temperature distribution matrix according to a mapping relationship between coating thickness and temperature gradient to obtain a coating uniformity prediction graph; fusing the substrate dynamic temperature distribution matrix and the coating uniformity prediction graph to obtain a cross-validation grid; when any one grid fails to pass the verification, generating a three-dimensional compensation vector; and real-time regulation and control of process parameters of a coating device. The application solves the technical problems that the existing coating process execution monitoring is not accurate enough in predicting coating uniformity and is difficult to realize real-time and accurate regulation and control of process parameters, and achieves the technical effects of accurately predicting coating uniformity and realizing real-time and accurate regulation and control of process parameters.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of vacuum coating, in particular to an NCVM non-conductive vacuum coating process execution monitoring method and system. BACKGROUND

[0002] In the vacuum coating process, the uniformity of the coating is one of the key factors affecting product quality, especially for non-conductive material coating (NCVM), poor uniformity may lead to inconsistent film performance, affecting the appearance and functionality of the product. Currently, the industry mainly monitors the temperature distribution of the substrate surface to indirectly evaluate the uniformity of the coating, and adjusts the process parameters accordingly. However, the existing method only relies on static analysis of temperature distribution, ignoring the dynamic correlation of temperature change rate and film thickness deviation, resulting in inaccurate prediction of coating uniformity, making it difficult to real-time and accurately regulate process parameters, thereby affecting the quality of the coating.

[0003] At present, in the related technology, the NCVM non-conductive vacuum coating process execution monitoring has the technical problem of inaccurate prediction of coating uniformity, which makes it difficult to real-time and accurately regulate process parameters. SUMMARY

[0004] The present application provides an NCVM non-conductive vacuum coating process execution monitoring method and system, which adopts dynamic grid division of the substrate surface in the vacuum coating chamber, obtains an adaptive grid substrate surface, uses an infrared thermal imager to scan the adaptive grid substrate surface in real time, generates a substrate dynamic temperature distribution matrix, analyzes and predicts the substrate dynamic temperature distribution matrix according to the mapping relationship between the coating thickness and the temperature gradient, generates a coating uniformity prediction map, aligns and fuses the substrate dynamic temperature distribution matrix and the coating uniformity prediction map, generates a cross-validation grid, verifies each grid in the cross-validation grid, if the verification fails, generates a three-dimensional compensation vector (including plasma gun angle, sputtering power and residence time), adjusts the process parameters of the vacuum coating equipment in real time according to the three-dimensional compensation vector, optimizes the coating quality, etc. Technical means, solves the technical problem of inaccurate prediction of coating uniformity in the existing NCVM non-conductive vacuum coating process execution monitoring, which makes it difficult to real-time and accurately regulate process parameters, achieves the technical effect of accurately predicting the uniformity of the coating, thereby real-time and accurately regulating the process parameters.

[0005] The application provides an NCVM non-conductive vacuum coating process execution monitoring method, comprising: performing detection unit grid dynamic division on the surface of a substrate in a vacuum coating cavity to obtain an adaptive grid substrate surface; performing real-time scanning on the adaptive grid substrate surface by an infrared thermal imager to obtain a substrate dynamic temperature distribution matrix; analyzing and predicting the substrate dynamic temperature distribution matrix according to a mapping relationship between coating thickness and temperature gradient to obtain a coating uniformity prediction graph; performing alignment fusion processing on the substrate dynamic temperature distribution matrix and the coating uniformity prediction graph to obtain a cross-validation grid; when any grid in the cross-validation grid fails to pass the verification, generating a three-dimensional compensation vector; and real-time regulating and controlling the process parameters of a vacuum coating device according to the three-dimensional compensation vector.

[0006] In a possible implementation, the surface of the substrate in the vacuum coating cavity is divided into detection unit grids dynamically to obtain an adaptive grid substrate surface, and the following processing is performed: presetting a conversion relationship between the surface curvature of the substrate and the grid density; obtaining a plurality of surface curvatures of the current substrate in real time, and calculating the adaptive grid density according to the conversion relationship; and dividing the surface of the substrate in the vacuum coating cavity into detection unit grids dynamically according to the adaptive grid density, setting the geometric center point of each grid as a temperature detection reference point, and obtaining the adaptive grid substrate surface.

[0007] In a possible implementation, the surface of the substrate in the vacuum coating cavity is divided into detection unit grids dynamically to obtain an adaptive grid substrate surface, and the following processing is further performed: extracting a temperature change rate from the substrate dynamic temperature distribution matrix to obtain a substrate dynamic temperature change rate matrix; traversing the substrate dynamic temperature change rate matrix, extracting a unit grid with a temperature change rate greater than a preset temperature change rate threshold, and obtaining a temperature abnormal grid; and updating the adaptive grid substrate surface after encrypting and reconstructing the temperature abnormal grid.

[0008] In a possible implementation, the substrate dynamic temperature distribution matrix is analyzed and predicted according to the mapping relationship between the coating thickness and the temperature gradient to obtain the coating uniformity prediction graph, and the following processing is performed: obtaining historical coating data, including a historical substrate dynamic temperature distribution matrix and a historical film thickness deviation value matrix, wherein the film thickness deviation value refers to the difference between the actual film thickness and the standard film thickness; calculating the temperature gradient of adjacent grids according to the historical substrate dynamic temperature distribution matrix to obtain a historical substrate temperature gradient distribution matrix; inputting the historical substrate temperature gradient distribution matrix and the historical film thickness deviation value matrix into a convolutional neural network model for training and learning to obtain the mapping relationship between the coating thickness and the temperature gradient; and analyzing and predicting the substrate dynamic temperature distribution matrix according to the mapping relationship to obtain the coating uniformity prediction graph.

[0009] In a possible implementation, the historical substrate temperature gradient distribution matrix and the historical film thickness deviation value matrix are input into a convolutional neural network model for training and learning to obtain a mapping relationship between film thickness and temperature gradient, and the following processing is performed: a double-branch convolutional neural network model is constructed, wherein the double-branch convolutional neural network model includes a first branch and a second branch; the first branch receives the historical substrate temperature gradient distribution matrix and extracts spatial features through a convolution kernel; the second branch receives historical substrate temperature gradient distribution matrices at the same position in the previous N production batches, and extracts time features through an LSTM layer, wherein N is greater than or equal to 3; the spatial features and the time features are weighted and spliced to obtain a film thickness deviation value prediction matrix; when the mean square error of the film thickness deviation value prediction matrix and the historical film thickness deviation value matrix is lower than a preset mean square error threshold and a prediction accuracy reaches a preset accuracy threshold, the training is stopped and the mapping relationship between film thickness and temperature gradient is exported.

[0010] In a possible implementation, when any one of the cross-validation grids fails the verification, a three-dimensional compensation vector is generated, and the following processing is performed: the absolute temperature and the temperature rise rate are extracted from the cross-validation grids; if the absolute temperature does not satisfy a preset temperature interval and / or the temperature rise rate does not satisfy a preset temperature rise rate interval, the corresponding grid is marked as a dangerous grid; a film thickness deviation prediction value is extracted from the dangerous grid for cross-validation, and if the film thickness deviation prediction value does not satisfy a preset film thickness deviation interval, the verification fails, and a three-dimensional compensation vector is generated.

[0011] In a possible implementation, the following processing is performed: the verification failure also needs to satisfy a temperature-film thickness collaborative verification condition, and the collaborative verification condition is as follows: a temperature abnormality degree is calculated according to the substrate material and the absolute temperature and the temperature rise rate; a film thickness abnormality degree is calculated according to the substrate material and the film thickness deviation prediction value; the temperature abnormality degree and the film thickness abnormality degree are multiplied to obtain a collaborative factor; if the temperature abnormality degree is greater than a preset temperature abnormality degree threshold, the film thickness abnormality degree is greater than a preset film thickness abnormality degree threshold, and the collaborative factor is greater than a preset collaborative factor threshold, it is determined that the verification fails.

[0012] In a possible implementation, when any one of the cross-validation grids fails the verification, a three-dimensional compensation vector is generated, and the following processing is performed: temperature abnormality information and film thickness abnormality information are extracted from the verification-failed grid; an angle compensation amount of the plasma gun is calculated according to the temperature abnormality information; a sputtering power compensation amount and a residence time compensation amount are calculated according to the film thickness abnormality information; the angle compensation amount, the sputtering power compensation amount, and the residence time compensation amount are integrated to generate the three-dimensional compensation vector.

[0013] In a possible implementation, the following processing is further performed: calculating an edge compensation coefficient according to a radius of curvature of the substrate edge region grid; adjusting the three-dimensional compensation vector according to the edge compensation coefficient to obtain a three-dimensional reinforced compensation vector; after performing compensation on the edge region grid according to the three-dimensional reinforced compensation vector, starting edge secondary verification, and using a laser thickness gauge to retest the edge region grid to obtain a retest result; and performing coating optimization on the edge region grid according to the retest result.

[0014] The application further provides an NCVM non-conductive vacuum coating process execution monitoring system, comprising: a detection unit grid dynamic division module, configured to perform detection unit grid dynamic division on a substrate surface in a vacuum coating cavity to obtain an adaptive grid substrate surface; an infrared thermal imaging scanning module, configured to perform real-time scanning on the adaptive grid substrate surface by using an infrared thermal imager to obtain a substrate dynamic temperature distribution matrix; a coating uniformity prediction module, configured to analyze and predict the substrate dynamic temperature distribution matrix according to a mapping relationship between coating thickness and temperature gradient to obtain a coating uniformity prediction graph; an alignment fusion processing module, configured to perform alignment fusion processing on the substrate dynamic temperature distribution matrix and the coating uniformity prediction graph to obtain a cross-verification grid; a three-dimensional compensation vector generation module, configured to generate a three-dimensional compensation vector when any one grid in the cross-verification grid fails verification; and a process parameter regulation module, configured to regulate process parameters of a vacuum coating device in real time according to the three-dimensional compensation vector.

[0015] The application provides an NCVM non-conductive vacuum coating process execution monitoring method and system. First, a detection unit grid dynamic division is performed on a substrate surface in a vacuum coating cavity to obtain an adaptive grid substrate surface. Then, real-time scanning is performed on the adaptive grid substrate surface by using an infrared thermal imager to obtain a substrate dynamic temperature distribution matrix. Next, analysis and prediction are performed on the substrate dynamic temperature distribution matrix according to a mapping relationship between coating thickness and temperature gradient to obtain a coating uniformity prediction graph. Then, alignment fusion processing is performed on the substrate dynamic temperature distribution matrix and the coating uniformity prediction graph to obtain a cross-verification grid. When any one grid in the cross-verification grid fails verification, a three-dimensional compensation vector is generated. Finally, process parameters of a vacuum coating device are regulated in real time according to the three-dimensional compensation vector. The technical effect of accurately predicting the uniformity of coating and regulating process parameters in real time and accurately is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application will be briefly introduced below, and the flowcharts are used to illustrate the operations performed by the system according to the embodiments of the present application in the present application. It should be understood that the foregoing or the following operations are not necessarily performed in sequence. On the contrary, various steps can be processed in reverse order or simultaneously according to needs. Meanwhile, other operations can be added to these processes, or one or more steps of operations can be removed from these processes.

[0017] Figure 1 A flowchart of an NCVM non-conductive vacuum coating process execution monitoring method provided by the embodiments of the present application.

[0018] Figure 2 A structural diagram of an NCVM non-conductive vacuum coating process execution monitoring system provided by the embodiments of the present application.

[0019] Label explanation: detection unit grid dynamic division module 10, infrared thermal imaging scanning module 20, coating uniformity prediction module 30, alignment fusion processing module 40, three-dimensional compensation vector generation module 50, process parameter regulation and control module 60. DETAILED DESCRIPTION

[0020] The above description is only a summary of the technical solutions of the present application, in order to more clearly understand the technical means of the present application, the content of the specification can be implemented, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described.

[0021] In order to make the purposes, technical solutions and advantages of the present application more clear, the present application will be further described in detail below with reference to the drawings, and the described embodiments should not be regarded as limiting the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0022] In the following description, "some embodiments" are referred to, which describe a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict, and the term "first\second" referred to is only to distinguish similar objects, and does not represent a specific order for the objects. The terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or modules not clearly listed or inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by those skilled in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application.

[0023] The embodiments of the present application provide an NCVM non-conductive vacuum coating process execution monitoring method, as shown in the accompanying drawings, the method comprises: Figure 1 The method comprises:

[0024] Step S100, the surface of the substrate in the vacuum coating cavity is detected by a unit grid dynamic division, and an adaptive grid substrate surface is obtained.

[0025] Specifically, a plurality of high-precision position sensors and optical sensors are installed in the vacuum coating cavity for real-time detection of the shape, size and position information of the substrate surface. Start the sensor array to collect image data of the substrate surface. The collected image data is transmitted to an image processing system, and the image processing system processes the image collected by the sensor through computer vision technology (such as noise reduction, contrast enhancement), and adaptively adjusts the size and distribution of the grid according to the shape and size of the substrate, to ensure that each grid can accurately reflect the local features of the substrate surface.

[0026] For example, assuming that the substrate is an irregularly shaped plastic part, after the sensor array collects the surface image of the part, the image processing system will divide the substrate surface into a plurality of grid units of different sizes according to the contour and surface features of the part. If the substrate surface has complex curved surfaces, the grid will be more densely distributed in the curved surface area to ensure accurate capture of local changes.

[0027] In a possible implementation, the surface of the substrate in the vacuum coating chamber is dynamically divided into a grid by the detection unit to obtain an adaptive grid substrate surface. Step S100 further includes step S110 of presetting a conversion relationship between the curvature of the curved surface of the substrate and the grid density. Specifically, the influence of the grid density on the uniformity of the coating under different curvatures of the curved surface is studied through experiments, and a mathematical model is established according to the experimental data to describe the quantitative relationship between the curvature of the curved surface and the grid density. For example, when the curvature of the curved surface increases within a certain range, the grid density increases linearly; when the curvature of the curved surface exceeds a certain threshold, the grid density increases at a faster speed. This relationship can be described by a piecewise function or a nonlinear function, and the specific form is determined according to the experimental data.

[0028] Step S120, a plurality of curvatures of the curved surface of the current substrate are obtained in real time, and an adaptive grid density is calculated according to the conversion relationship. Specifically, a plurality of high-precision sensors (such as optical sensors or laser scanners) are installed in the vacuum coating chamber to measure the curvature of the curved surface of the substrate in real time. The sensor array is started to measure the curvature of the curved surface of a plurality of points on the substrate surface in real time. The curvature data measured by the sensors is transmitted to a data processing system, and the data processing system calculates the adaptive grid density corresponding to each measurement point according to the preset conversion relationship.

[0029] For example, the surface of the substrate has a plurality of regions, and the curvatures of the curved surface of each region are different. After the sensor array measures the curvatures of these regions, the data processing system will calculate the grid density of each region according to the preset conversion relationship. For example, if the curvature of the curved surface of a certain region is high, the grid density calculated according to the conversion relationship will also be high accordingly; if the curvature of the curved surface of another region is low, the grid density will be low.

[0030] Step S130, the surface of the substrate in the vacuum coating chamber is dynamically divided into a grid by the detection unit according to the adaptive grid density, the geometric center point of each grid is set as a temperature detection reference point, and an adaptive grid substrate surface is obtained. Specifically, the surface of the substrate is dynamically divided according to the calculated adaptive grid density using image processing and grid division algorithms. For each divided grid, the geometric center point is calculated, and the geometric center point of each grid is set as a temperature detection reference point. For example, the surface of the substrate is divided into a plurality of grids, and the size and shape of each grid are determined according to the adaptive grid density. For each grid, the geometric center point is calculated by finding the boundary of the grid and calculating the average position of all points within the boundary. This average position is the geometric center point of the grid, which is used as the reference point for temperature detection.

[0031] The implementation can capture the changes of the substrate surface more finely by dynamically dividing the detection unit grid and adjusting the grid density according to the curvature of the curved surface, especially in the area with large curvature, which helps to improve the uniformity of the coating and reduce the problem of uneven coating thickness caused by irregular surface shape.

[0032] In a possible implementation, the substrate surface in the vacuum coating chamber is detected by dynamic division of the detection unit grid to obtain an adaptive grid substrate surface, and step S100 further includes step S140 of extracting a temperature change rate according to the substrate dynamic temperature distribution matrix to obtain a substrate dynamic temperature change rate matrix. Specifically, the substrate dynamic temperature distribution matrix is read from the database, time series analysis is performed on the temperature data of each grid, and the temperature change rate thereof is calculated. The calculated temperature change rate is stored as a new matrix, i.e., the substrate dynamic temperature change rate matrix.

[0033] For example, the substrate dynamic temperature distribution matrix records the temperature values of each grid at different time points. By calculating the temperature difference between adjacent time points and dividing by the time interval, the temperature change rate of each grid can be obtained. If the temperature of a grid at time point t1 is T1 and the temperature at time point t2 is T2, the temperature change rate of the grid is (T2-T1) / (t2-t1).

[0034] Step S150, traversing the substrate dynamic temperature change rate matrix, extracting the unit grid with a temperature change rate greater than a preset temperature change rate threshold to obtain a temperature abnormal grid. Specifically, the substrate dynamic temperature change rate matrix is read from the database, and each grid in the matrix is traversed to extract the temperature change rate thereof. The temperature change rate of each grid is compared with the preset temperature change rate threshold. If the temperature change rate of a certain grid is greater than the preset temperature change rate threshold, the grid is marked as a temperature abnormal grid, and all temperature abnormal grids are stored in the database. For example, assuming that the preset temperature change rate threshold is 3℃ / min, if the temperature change rate of a certain grid is 5℃ / min, the grid will be marked as a temperature abnormal grid.

[0035] Step S160, after the temperature anomaly grid is encrypted and reconstructed, update the adaptive grid substrate surface. Specifically, each temperature anomaly grid is encrypted to increase the grid density to capture temperature changes more finely. The encrypted grid is re-integrated into the adaptive grid substrate surface to update the adaptive grid substrate surface, ensuring that the geometric center point of all grids remains as the temperature detection reference point. For example, assume that the original grid size of a certain temperature anomaly grid is 1 cm x 1 cm, and after encryption it is divided into smaller grids, such as 0.5 cm x 0.5 cm. In this way, the changes in the temperature anomaly area can be captured more finely. This implementation can capture temperature changes more finely after the temperature anomaly grid is encrypted and reconstructed, thereby providing more accurate data support for subsequent film uniformity prediction and process parameter control.

[0036] Step S200, real-time scanning of the adaptive grid substrate surface by an infrared thermal imager to obtain a substrate dynamic temperature distribution matrix.

[0037] Specifically, during the vacuum coating process, the infrared thermal imager installed in the coating cavity is started to ensure that it can cover the entire substrate surface. The infrared thermal imager scans the substrate surface at a set frequency (e.g., multiple times per second) to capture the temperature information of the substrate surface in real time. The infrared thermal imager outputs the scanned temperature data in the form of images, which contain temperature information at different positions on the substrate surface. The data acquisition system converts the temperature image data into digital signals and transmits the data to the data processing system through a high-speed data transmission system. After receiving the temperature data from the infrared thermal imager, the data processing system pre-processes the data (such as denoising, calibration, etc.). The pre-processed temperature data is organized according to the results of adaptive grid division to generate a substrate dynamic temperature distribution matrix. The temperature value of each grid is recorded in the matrix, and the rows and columns of the matrix correspond to the grid positions on the substrate surface, and the values in the matrix represent the real-time temperature of the grid. For example, assume that the substrate surface is divided into 100 grids, and the substrate dynamic temperature distribution matrix records the real-time temperature value of each grid to form a 10 x 10 matrix.

[0038] Step S300, according to the mapping relationship between the coating thickness and the temperature gradient, analyze and predict the substrate dynamic temperature distribution matrix to obtain a film uniformity prediction map.

[0039] Specifically, a mapping relationship model between coating thickness and temperature gradient is established in advance through historical data analysis. The substrate dynamic temperature distribution matrix is input into the mapping relationship model to predict the coating uniformity and generate a coating uniformity prediction map. For example, suppose the mapping relationship model indicates that the coating thickness in a region with a large temperature gradient will be thinner. If the temperature gradient of a certain region on the substrate surface is large, the mapping relationship model will predict that the coating thickness of the region is thinner, and the prediction will be marked on the coating uniformity prediction map.

[0040] In a possible implementation, according to the mapping relationship between the coating thickness and the temperature gradient, the substrate dynamic temperature distribution matrix is analyzed and predicted to obtain a coating uniformity prediction map, and step S300 further includes step S310 of acquiring historical coating data, including a historical substrate dynamic temperature distribution matrix and a historical film thickness deviation value matrix, wherein the film thickness deviation value refers to the difference between the actual film thickness and the standard film thickness. Specifically, the historical coating data, including the historical substrate dynamic temperature distribution matrix and the historical film thickness deviation value matrix, are read from the database. The collected historical coating data are cleaned and preprocessed to remove noise and outliers, and ensure the integrity and accuracy of the data. For example, the historical coating data include records of multiple coating processes, and each record includes the temperature distribution of different grids on the substrate surface and the corresponding film thickness deviation value. For example, in a certain historical record, the temperature of a certain grid is 30°C, the actual film thickness is 1.2 μm, and the standard film thickness is 1.0 μm, so the film thickness deviation value of the grid is 0.2 μm.

[0041] Step S320, according to the historical substrate dynamic temperature distribution matrix, the temperature gradient of adjacent grids is calculated to obtain a historical substrate temperature gradient distribution matrix. Specifically, for each grid in the historical substrate dynamic temperature distribution matrix, the temperature difference between the grid and the adjacent grid is calculated and divided by the grid spacing to obtain the temperature gradient. The calculated temperature gradient value is stored in a new matrix to form the historical substrate temperature gradient distribution matrix. For example, suppose the temperature of a certain grid is 30°C, the temperature of its adjacent grid is 32°C, and the grid spacing is 1 cm, then the temperature gradient of the grid is (32-30) / 1=2℃ / cm.

[0042] Step S330, input the historical substrate temperature gradient distribution matrix and the historical film thickness deviation value matrix into the convolutional neural network model for training and learning, and obtain the mapping relationship between the film thickness and the temperature gradient. Specifically, the historical substrate temperature gradient distribution matrix and the historical film thickness deviation value matrix are input into the convolutional neural network model for training. The model is trained using the back propagation algorithm, the model parameters are adjusted, and the model performance is optimized. The prediction accuracy of the model is evaluated by cross-validation and other methods to ensure the reliability and accuracy of the model. The trained convolutional neural network model is stored in the database. The convolutional neural network model includes multiple convolutional layers and pooling layers. The features of the temperature gradient distribution are extracted through convolution operation, and the feature dimension is reduced through pooling operation. The output of the model is the predicted film thickness deviation value. Through the back propagation algorithm, the model parameters are adjusted according to the difference between the predicted value and the actual value, and the model performance is optimized.

[0043] Step S340, according to the mapping relationship, analyze and predict the substrate dynamic temperature distribution matrix to obtain the film uniformity prediction map. Specifically, the current substrate dynamic temperature distribution matrix is read from the database, the adjacent grid temperature gradient is calculated to obtain the current substrate temperature gradient distribution matrix, the current substrate temperature gradient distribution matrix is input into the trained convolutional neural network model, and the model outputs the predicted film thickness deviation value matrix. According to the predicted film thickness deviation value matrix, the film uniformity prediction map is generated.

[0044] For example, assuming that the temperature of a certain grid in the current substrate dynamic temperature distribution matrix is 31℃, the temperature of its adjacent grid is 33℃, and the calculated temperature gradient is 2℃ / cm. Input these data into the convolutional neural network model, and the model outputs the predicted film thickness deviation value of this grid as 0.15μm. According to the prediction result, the film thickness deviation of this grid is marked in the film uniformity prediction map. This implementation mode can establish an accurate mapping relationship between the film thickness and the temperature gradient by training and learning the historical data using the convolutional neural network model. Compared with traditional empirical formula or simple model, the convolutional neural network can capture more complex nonlinear relationship, thereby improving the prediction accuracy.

[0045] In a possible implementation, the historical substrate temperature gradient distribution matrix and the historical film thickness deviation value matrix are input into a convolutional neural network model for training and learning to obtain a mapping relationship between the film thickness and the temperature gradient, and step S330 further includes step S331 of constructing a double-branch convolutional neural network model, where the double-branch convolutional neural network model includes a first branch and a second branch. Specifically, a convolutional neural network model including two branches is constructed, which are respectively used to extract spatial features and temporal features. The first branch is used to process static spatial features, and the second branch is used to process dynamic temporal features. The outputs of the two branches are fused through weighted splicing. Specifically, the first branch adopts a convolutional layer and a pooling layer, which are used to extract spatial features of the historical substrate temperature gradient distribution matrix. The second branch adopts a long short-term memory (LSTM) layer, which is used to extract temporal features of the historical substrate temperature gradient distribution matrix at the same position in the previous N production batches. The outputs of the two branches are fused through weighted splicing to form a film thickness deviation value prediction matrix. A loss function and a prediction accuracy are defined as training targets, and preset mean square error threshold and accuracy threshold are set.

[0046] In step S332, the first branch receives the historical substrate temperature gradient distribution matrix and extracts spatial features through a convolution kernel. Specifically, the historical substrate temperature gradient distribution matrix is input into the convolutional layer of the first branch, the convolutional layer performs convolution operation on the input matrix using the convolution kernel to extract local spatial features. The pooling layer performs pooling operation on the output of the convolutional layer to reduce the feature dimension. The convolution and pooling operations are repeated to extract higher-level spatial features. The extracted spatial features are output to a subsequent weighted splicing module.

[0047] For example, the convolution kernel has a size of 3x3 and a step of 1, and the convolution operation slides on the input matrix to calculate the convolution value at each position. For example, for a 5x5 temperature gradient distribution matrix, the convolution operation generates a new feature matrix with a size of 3x3. The pooling layer further reduces the dimension of the feature matrix, for example, by maximum pooling operation to reduce the 3x3 feature matrix to 2x2.

[0048] In step S333, the second branch receives the historical substrate temperature gradient distribution matrix at the same position in the previous N production batches and extracts temporal features through an LSTM layer, where N≥3. Specifically, the historical substrate temperature gradient distribution matrix at the same position in the previous N production batches is read from the database and input into the LSTM layer of the second branch. The LSTM layer processes the input time series data to extract temporal features, and the extracted temporal features are output to a subsequent weighted splicing module.

[0049] For example, each of the first N production batches has a temperature gradient distribution matrix at the same position, which forms a time series. The LSTM layer processes these matrices one by one, capturing long-term dependencies in the time series through the gating mechanism. For example, if the temperature gradient at a certain position gradually increases in the first few batches, the LSTM layer can capture this trend and output it as a time feature.

[0050] At step S334, the spatial features and the temporal features are weighted and spliced to obtain a film thickness deviation value prediction matrix. Specifically, according to the preset weight or the weight learned by the model, the spatial features extracted by the first branch and the temporal features extracted by the second branch are weighted and spliced to form a film thickness deviation value prediction matrix. The film thickness deviation value prediction matrix is compared with the historical film thickness deviation value matrix to calculate the mean square error and the prediction accuracy.

[0051] At step S335, when the mean square error of the film thickness deviation value prediction matrix and the historical film thickness deviation value matrix is lower than the preset mean square error threshold and the prediction accuracy reaches the preset accuracy threshold, the training is stopped and the mapping relationship between the film thickness and the temperature gradient is derived. Specifically, the mean square error (MSE) is used as the loss function to measure the difference between the film thickness deviation value prediction matrix and the historical film thickness deviation value matrix. During the training process, the mean square error between the film thickness deviation value prediction matrix and the historical film thickness deviation value matrix is calculated, and the accuracy of the model in predicting the film thickness deviation is calculated. If the mean square error is lower than the preset mean square error threshold and the prediction accuracy reaches the preset accuracy threshold, the training is stopped. The trained model is derived to obtain the mapping relationship between the film thickness and the temperature gradient. For example, the preset mean square error threshold is 0.01 and the accuracy threshold is 90%. During the training process, the model continuously adjusts the parameters to optimize the mean square error and the prediction accuracy. When the mean square error is less than 0.01 and the classification accuracy is greater than 90%, the training stops and the model is derived. This implementation mode extracts spatial features and temporal features simultaneously through a double-branch convolutional neural network model, which can more comprehensively capture factors affecting film thickness deviation. Compared with a single feature extraction method, this double-branch model can significantly improve the prediction accuracy.

[0052] At step S400, the substrate dynamic temperature distribution matrix and the film coating uniformity prediction map are aligned and fused to obtain a cross-validation grid.

[0053] Specifically, the substrate dynamic temperature distribution matrix records the real-time temperature value of each grid on the substrate surface, and the coating uniformity prediction map marks the film thickness deviation prediction value of each grid. Using an image alignment algorithm, the substrate dynamic temperature distribution matrix and the coating uniformity prediction map are spatially aligned to ensure that the grids in the two matrices correspond one by one in position. For example, if the substrate surface is divided into a 10x10 grid, the aligned matrix will ensure that the temperature value and the film thickness deviation prediction value of each grid can be accurately matched. The data in the aligned substrate dynamic temperature distribution matrix and the coating uniformity prediction map are fused, that is, the temperature information and the film thickness deviation prediction information are integrated into a unified grid to form a cross-validation grid. For example, for each grid, the temperature value and the film thickness deviation prediction value can be combined into a feature vector.

[0054] Step S500, when any one of the cross-validation grids fails the verification, a three-dimensional compensation vector is generated.

[0055] Specifically, each grid in the cross-validation grid is verified to determine whether it meets the preset coating quality standard. If a grid fails the verification, the position and verification result of the grid are recorded. According to the verification result, a three-dimensional compensation vector is generated, including the angle compensation amount of the plasma gun, the sputtering power compensation amount, and the residence time compensation amount. The angle compensation amount of the plasma gun is used to adjust the jet angle of the plasma gun to change the direction and coverage of the coating. The sputtering power compensation amount is used to adjust the power of the sputtering source to change the deposition rate of the coating. The residence time compensation amount is used to adjust the residence time of the plasma gun on each grid to change the thickness of the coating.

[0056] For example, assuming that the coating thickness of a certain grid in the cross-validation grid is lower than the standard value, a three-dimensional compensation vector is generated according to the position and thickness deviation of the grid: adjust the angle of the plasma gun to +5° to make the coating more evenly cover the area; increase the sputtering power by 10% to increase the deposition rate of the coating; increase the residence time by 2 seconds to increase the coating thickness of the area.

[0057] In one possible implementation, when any one of the cross-validation grids fails the verification, a three-dimensional compensation vector is generated, and step S500 further includes step S510 of extracting the absolute temperature and the temperature rise rate from the cross-validation grid. Specifically, the cross-validation grid is read from the database, and for each grid, its absolute temperature value is extracted. The temperature rise rate of each grid is calculated, that is, the temperature difference between adjacent time points divided by the time interval.

[0058] Step S520, if the absolute temperature does not meet the preset temperature interval and / or the temperature rise rate does not meet the preset temperature rise rate interval, mark the corresponding grid as a dangerous grid. Specifically, compare the absolute temperature of each grid with the preset temperature interval to determine whether it is within the allowed range. Compare the temperature rise rate of each grid with the preset temperature rise rate interval to determine whether it is within the allowed range. If the absolute temperature or temperature rise rate of a certain grid does not meet the preset interval, mark the grid as a dangerous grid. For example, assume that the preset temperature interval is [20℃, 40℃] and the temperature rise rate interval is [0.1℃ / min, 1.0℃ / min]. For a certain grid, if its absolute temperature is 45℃ and / or the temperature rise rate is 1.2℃ / min, the grid is marked as a dangerous grid.

[0059] Step S530, extract the film thickness deviation prediction value from the dangerous grid for cross-validation, if the film thickness deviation prediction value does not meet the preset film thickness deviation interval, the validation fails, and a three-dimensional compensation vector is generated. Specifically, for each dangerous grid, extract its film thickness deviation prediction value. Compare the film thickness deviation prediction value with the preset film thickness deviation interval to determine whether it meets the condition. If the film thickness deviation prediction value of a certain dangerous grid does not meet the preset interval, the validation fails. For the grid that fails the validation, generate a three-dimensional compensation vector, including the angle compensation amount of the plasma gun, the sputtering power compensation amount and the residence time compensation amount. For example, assume that the preset film thickness deviation interval is [-0.1μm, 0.1μm]. For a certain dangerous grid, if its film thickness deviation prediction value is 0.2μm, the grid fails the validation.

[0060] This implementation can more comprehensively evaluate the film coating quality of each grid by considering the absolute temperature, temperature rise rate and film thickness deviation prediction value at the same time, effectively avoiding the misjudgment that may be caused by relying on the prediction value alone. For example, even if the film thickness deviation prediction value shows that the film thickness of a certain area may exceed the standard range, if the temperature data (absolute temperature and temperature rise rate) of the area are within the normal range, the area may not need to be adjusted. Conversely, if the film thickness deviation prediction value is within the normal range, but the temperature data is abnormal (such as the absolute temperature is too high and / or the temperature rise rate is too fast), the area may still have potential problems and needs further attention. This combined validation mechanism improves the accuracy of the system in identifying problem areas and avoids misjudgment caused by single data abnormality.

[0061] In a possible implementation, the step S500 further includes a step S540, wherein the verification failure further needs to satisfy a temperature-film thickness coordination verification condition, and the coordination verification condition is as follows: calculating a temperature abnormality degree according to the substrate material and the absolute temperature and the temperature rise rate; calculating a film thickness abnormality degree according to the substrate material and the film thickness deviation prediction value; multiplying the temperature abnormality degree and the film thickness abnormality degree to obtain a coordination factor; and if the temperature abnormality degree is greater than a preset temperature abnormality degree threshold, the film thickness abnormality degree is greater than a preset film thickness abnormality degree threshold, and the coordination factor is greater than a preset coordination factor threshold, determining that the verification fails.

[0062] Specifically, the characteristic parameters of the substrate material and the absolute temperature and the temperature rise rate of each dangerous grid are read from the database. According to the characteristics of the substrate material, the temperature abnormality degree of each dangerous grid is calculated. The temperature abnormality degree reflects the deviation degree of the current temperature and the temperature rise rate from the allowable range of the substrate material. For example, if the substrate material is allowed to have an absolute temperature range of [20℃, 40℃] and a temperature rise rate range of [0.1℃ / min, 1.0℃ / min] under the current process condition, the degree of exceeding these ranges will be quantified as the temperature abnormality degree.

[0063] The characteristic parameters of the substrate material and the film thickness deviation prediction value of each dangerous grid are read from the database. According to the characteristics of the substrate material, the film thickness abnormality degree of each dangerous grid is calculated. The film thickness abnormality degree reflects the deviation degree of the film thickness deviation prediction value from the allowable range of the substrate material. For example, if the substrate material is allowed to have a film thickness deviation range of [-0.1μm, 0.1μm], the degree of exceeding this range will be quantified as the film thickness abnormality degree.

[0064] The temperature abnormality degree and the film thickness abnormality degree of each dangerous grid are multiplied to obtain a coordination factor. The coordination factor comprehensively reflects the abnormality degrees of the two dimensions of temperature and film thickness. For example, if the temperature abnormality degree of a certain grid is 1.2 and the film thickness abnormality degree is 1.5, the coordination factor is 1.2x1.5=1.8.

[0065] The calculated temperature abnormality degree, film thickness abnormality degree and coordination factor are compared with preset threshold values respectively. If the temperature abnormality degree is greater than a preset temperature abnormality degree threshold, the film thickness abnormality degree is greater than a preset film thickness abnormality degree threshold, and the coordination factor is greater than a preset coordination factor threshold, it is determined that the verification of the grid fails.

[0066] For example, assume that the substrate material allows an absolute temperature range of [20°C, 40°C] under current process conditions, and a temperature rise rate range of [0.1°C / min, 1.0°C / min]. Assume that the absolute temperature of a certain grid is 45°C, the absolute temperature anomaly can be equal to (45-40) / (40-20) = 0.25. Assume that the temperature rise rate of this grid is 1.2°C / min. The temperature rise rate anomaly can be equal to (1.2-1) / (1-0.1) ≈ 0.222. The temperature anomaly can be the maximum of the absolute temperature anomaly and the temperature rise rate anomaly, or a weighted average of the two. Assuming the temperature anomaly is the maximum of the two, then the temperature anomaly = 0.25. Assume that the substrate material allows a film thickness deviation range of [-0.1 μm, 0.1 μm]. Assume that the film thickness deviation prediction value of this grid is 0.2 μm. The film thickness anomaly can be equal to (0.2-0.1) / [0.1-(-0.1)] = 0.5. Then the synergy factor = 0.25 x 0.5 = 0.125. Assuming that the preset temperature anomaly threshold is 1.0, the film thickness anomaly threshold is 1.0, and the synergy factor threshold is 1.5. Since the temperature anomaly 0.25 < 1.0, the film thickness anomaly 0.5 < 1.0, and the synergy factor 0.125 < 1.5, it is determined that the grid passes the verification.

[0067] This implementation can more accurately identify the areas that truly have problems by considering both temperature anomaly and film thickness anomaly, and calculating the synergy factor. The synergy factor comprehensively reflects the abnormality degree of both temperature and film thickness, enabling a more comprehensive assessment of the severity of the problem. For example, even if the temperature anomaly and film thickness anomaly are not high individually, their combination may still indicate a potential problem. This synergistic verification mechanism avoids false positives that may result from relying solely on a single anomaly, improving the accuracy and reliability of the verification.

[0068] In one possible implementation, when any one of the cross-verification grids fails the verification, a three-dimensional compensation vector is generated, and step S500 further includes step S550 of extracting temperature anomaly information and film thickness anomaly information from the grid that fails the verification. Specifically, a list of grids that fail the verification is read from the database. For each grid that fails the verification, its temperature anomaly information (absolute temperature, temperature rise rate) and film thickness anomaly information (film thickness deviation prediction value) are extracted. For example, assume that a certain grid has a verification result of failure, its absolute temperature is 45°C, its temperature rise rate is 1.2°C / min, and its film thickness deviation prediction value is 0.2 μm. These information is extracted and stored in a temporary database.

[0069] Step S560, according to the temperature abnormal information, calculate the angle compensation amount of the plasma gun. Specifically, read the temperature abnormal information (absolute temperature and / or temperature rise rate) from the temporary database, use the preset compensation amount calculation model, and calculate the angle compensation amount of the plasma gun according to the temperature abnormal information. For example, it is assumed that the preset model indicates that when the absolute temperature exceeds 40℃, the angle of the plasma gun needs to be adjusted to reduce the heat input. The specific compensation amount can be calculated by the following formula: angle compensation amount = k x (actual absolute temperature - maximum allowed absolute temperature), wherein k is a model parameter.

[0070] Step S570, according to the film thickness abnormal information, calculate the sputtering power compensation amount and the residence time compensation amount. Specifically, similar to step S560, read the film thickness abnormal information (film thickness deviation prediction value) from the temporary database, use the preset compensation amount calculation model, and calculate the sputtering power compensation amount and the residence time compensation amount according to the film thickness abnormal information.

[0071] Step S580, integrate the angle compensation amount, the sputtering power compensation amount and the residence time compensation amount to generate the three-dimensional compensation vector. Specifically, integrate the calculated angle compensation amount, sputtering power compensation amount and residence time compensation amount into a three-dimensional compensation vector in the format (angle compensation amount, sputtering power compensation amount, residence time compensation amount). This implementation can provide an accurate compensation strategy by calculating the angle compensation amount, sputtering power compensation amount and residence time compensation amount of the plasma gun respectively and integrating them into a three-dimensional compensation vector, thereby improving the uniformity and consistency of the film coating and effectively reducing the rate of defective products.

[0072] In a possible implementation, step S500 further includes step S590, according to the radius of curvature of the substrate edge area grid, calculate the edge compensation coefficient. Specifically, use a high-precision sensor or optical measurement equipment to measure the radius of curvature of the substrate edge area grid, and calculate the edge compensation coefficient by substituting the measured radius of curvature into the preset compensation coefficient calculation model. For example, the preset compensation coefficient calculation model indicates that the edge compensation coefficient is inversely proportional to the radius of curvature.

[0073] Step S5100, according to the edge compensation coefficient, adjust the three-dimensional compensation vector to obtain a three-dimensional reinforced compensation vector. Specifically, according to the edge compensation coefficient, adjust each component of the three-dimensional compensation vector. Integrate the adjusted compensation amount into a new three-dimensional compensation vector to obtain a three-dimensional reinforced compensation vector.

[0074] Step S5110, after performing compensation on the edge area grid according to the three-dimensional reinforcement compensation vector, starting edge secondary verification, using a laser thickness gauge to retest the edge area grid, and obtaining a retest result. Specifically, according to the three-dimensional reinforcement compensation vector, adjusting the process parameters, and performing compensation on the edge area grid. Using a laser thickness gauge to retest the compensated edge area grid, and obtaining a retest result.

[0075] Step S5120, according to the retest result, optimizing the film coating of the edge area grid. Specifically, according to the retest result, judging whether the film thickness is within the allowable range. If the film thickness deviation value is less than the allowable minimum deviation value, calculate a new compensation amount, adjust the process parameters, and further optimize the film coating quality of the edge area grid. If the film thickness deviation value is greater than the allowable maximum deviation value, first remove the excess coating, and then re-deposit to ensure that the film thickness is within the allowable range. This implementation can make more accurate compensation for the special shape of the edge area by calculating the edge compensation coefficient and adjusting the three-dimensional compensation vector. This targeted compensation improves the film coating quality of the edge area. Through retesting by a laser thickness gauge, and further optimizing the process parameters according to the retest result, the film thickness uniformity of the edge area is ensured.

[0076] Step S600, according to the three-dimensional compensation vector, real-time control of the process parameters of the vacuum coating equipment.

[0077] Specifically, input the three-dimensional compensation vector into the process parameter control system, and the process parameter control system adjusts the process parameters of the vacuum coating equipment according to the three-dimensional compensation vector: adjusting the angle of the plasma gun, adjusting the power of the sputtering source, adjusting the residence time of the plasma gun on each grid. The adjusted process parameters are fed back to the equipment to ensure that the coating process is carried out according to the new parameters. Real-time monitoring of the coating process ensures that the coating quality meets the requirements. For example, if the three-dimensional compensation vector indicates that the film thickness of a certain area needs to be increased, the process parameter control system adjusts the angle of the plasma gun to +5°, increases the sputtering power by 10%, and increases the residence time by 2 seconds. Through these adjustments, it is ensured that the film thickness of the area reaches the standard value.

[0078] The embodiment of the application adopts dynamic grid division of a detection unit for a substrate surface in a vacuum coating cavity to obtain an adaptive grid substrate surface, uses an infrared thermal imager to perform real-time scanning on the adaptive grid substrate surface to generate a substrate dynamic temperature distribution matrix, analyzes and predicts the substrate dynamic temperature distribution matrix according to a mapping relationship between coating thickness and temperature gradient to generate a coating uniformity prediction map, performs alignment and fusion processing on the substrate dynamic temperature distribution matrix and the coating uniformity prediction map to generate a cross-validation grid, verifies each grid in the cross-validation grid, generates a three-dimensional compensation vector (including a plasma gun angle, a sputtering power and a residence time) if the verification fails, and adjusts process parameters of a vacuum coating device in real time according to the three-dimensional compensation vector to optimize coating quality and other technical means, solves the technical problem that existing NCVM non-conductive vacuum coating process execution monitoring is not accurate enough in predicting coating uniformity, which leads to difficulty in real-time and accurate adjustment of process parameters, and achieves the technical effect that the coating uniformity is accurately predicted to realize real-time and accurate adjustment of process parameters.

[0079] In the foregoing, with reference to Figure 1 A NCVM non-conductive vacuum coating process execution monitoring method according to an embodiment of the application is described in detail. Next, with reference to Figure 2 A NCVM non-conductive vacuum coating process execution monitoring system according to an embodiment of the application is described.

[0080] The NCVM non-conductive vacuum coating process execution monitoring system according to the embodiment of the application is used to solve the technical problem that existing NCVM non-conductive vacuum coating process execution monitoring is not accurate enough in predicting coating uniformity, which leads to difficulty in real-time and accurate adjustment of process parameters, and achieves the technical effect that the coating uniformity is accurately predicted to realize real-time and accurate adjustment of process parameters. The NCVM non-conductive vacuum coating process execution monitoring system includes a detection unit grid dynamic division module 10, an infrared thermal imaging scanning module 20, a coating uniformity prediction module 30, an alignment and fusion processing module 40, a three-dimensional compensation vector generation module 50, and a process parameter adjustment module 60.

[0081] The detection unit grid dynamic division module 10 is configured to perform detection unit grid dynamic division on the substrate surface in the vacuum coating cavity to obtain an adaptive grid substrate surface; the infrared thermal imaging scanning module 20 is configured to perform real-time scanning on the adaptive grid substrate surface by using an infrared thermal imager to obtain a substrate dynamic temperature distribution matrix; the coating uniformity prediction module 30 is configured to analyze and predict the substrate dynamic temperature distribution matrix according to a mapping relationship between coating thickness and temperature gradient to obtain a coating uniformity prediction graph; the alignment fusion processing module 40 is configured to perform alignment fusion processing on the substrate dynamic temperature distribution matrix and the coating uniformity prediction graph to obtain a cross-validation grid; the three-dimensional compensation vector generation module 50 is configured to generate a three-dimensional compensation vector when any one grid in the cross-validation grid fails to pass the verification; and the process parameter regulation module 60 is configured to regulate the process parameters of the vacuum coating equipment in real time according to the three-dimensional compensation vector.

[0082] In the following, the specific configuration of the detection unit grid dynamic division module 10 will be described in detail. As described above, the detection unit grid dynamic division module 10 can further include a conversion relationship preset unit configured to preset a conversion relationship between the surface curvature of the substrate and the grid density; an adaptive grid density calculation unit configured to obtain a plurality of surface curvatures of the current substrate in real time, and calculate the adaptive grid density according to the conversion relationship; and a detection unit grid dynamic division unit configured to perform detection unit grid dynamic division on the substrate surface in the vacuum coating cavity according to the adaptive grid density, set the geometric center point of each grid as a temperature detection reference point, and obtain the adaptive grid substrate surface.

[0083] In the following, the specific configuration of the detection unit grid dynamic division module 10 will be described in detail. As described above, the detection unit grid dynamic division module 10 can further include a conversion relationship preset unit configured to preset a conversion relationship between the surface curvature of the substrate and the grid density; an adaptive grid density calculation unit configured to obtain a plurality of surface curvatures of the current substrate in real time, and calculate the adaptive grid density according to the conversion relationship; and a detection unit grid dynamic division unit configured to perform detection unit grid dynamic division on the substrate surface in the vacuum coating cavity according to the adaptive grid density, set the geometric center point of each grid as a temperature detection reference point, and obtain the adaptive grid substrate surface.

[0084] In the following, the specific configuration of the coating uniformity prediction module 30 will be described in detail. As described above, the coating uniformity prediction module 30 can further include: a historical coating data acquisition unit configured to acquire historical coating data, including a historical substrate dynamic temperature distribution matrix and a historical film thickness deviation value matrix, wherein the film thickness deviation value refers to the difference between the actual film thickness and the standard film thickness; a neighboring grid temperature gradient calculation unit configured to calculate the neighboring grid temperature gradient according to the historical substrate dynamic temperature distribution matrix to obtain a historical substrate temperature gradient distribution matrix; a convolution training and learning unit configured to input the historical substrate temperature gradient distribution matrix and the historical film thickness deviation value matrix into a convolution neural network model for training and learning to obtain a mapping relationship between the coating thickness and the temperature gradient; and a mapping unit configured to analyze and predict the substrate dynamic temperature distribution matrix according to the mapping relationship to obtain a coating uniformity prediction map.

[0085] The historical substrate temperature gradient distribution matrix and the historical film thickness deviation value matrix are input into a convolution neural network model for training and learning to obtain a mapping relationship between the coating thickness and the temperature gradient. The convolution training and learning unit can further include: a dual-branch convolution neural network model construction subunit configured to construct a dual-branch convolution neural network model, wherein the dual-branch convolution neural network model includes a first branch and a second branch; a spatial feature extraction subunit configured to receive the historical substrate temperature gradient distribution matrix by the first branch and extract spatial features through a convolution kernel; a temporal feature extraction subunit configured to receive historical substrate temperature gradient distribution matrices of the same position in the previous N production batches by the second branch and extract temporal features through an LSTM layer, wherein N≥3; a weighted splicing subunit configured to weight and splice the spatial features and the temporal features to obtain a film thickness deviation value prediction matrix; and a mapping relationship derivation subunit configured to stop training and derive the mapping relationship between the coating thickness and the temperature gradient when the mean square error of the film thickness deviation value prediction matrix and the historical film thickness deviation value matrix is lower than a preset mean square error threshold and the prediction accuracy reaches a preset accuracy threshold.

[0086] In the following, the specific configuration of the three-dimensional compensation vector generation module 50 will be described in detail. As described above, when any one of the cross-validation grids fails the validation, a three-dimensional compensation vector is generated, and the three-dimensional compensation vector generation module 50 can further include: a temperature information extraction unit for extracting the absolute temperature and the temperature rise rate from the cross-validation grids; a dangerous grid marking unit for marking the corresponding grid as a dangerous grid if the absolute temperature does not satisfy the preset temperature interval and / or the temperature rise rate does not satisfy the preset temperature rise rate interval; a cross-validation unit for extracting the film thickness deviation prediction value from the dangerous grid for cross-validation, and if the film thickness deviation prediction value does not satisfy the preset film thickness deviation interval, the validation fails, and a three-dimensional compensation vector is generated.

[0087] In the following, the specific configuration of the three-dimensional compensation vector generation module 50 will be described in detail. As described above, when any one of the cross-validation grids fails the validation, a three-dimensional compensation vector is generated, and the three-dimensional compensation vector generation module 50 can further include: a temperature information extraction unit for extracting the absolute temperature and the temperature rise rate from the cross-validation grids; a dangerous grid marking unit for marking the corresponding grid as a dangerous grid if the absolute temperature does not satisfy the preset temperature interval and / or the temperature rise rate does not satisfy the preset temperature rise rate interval; a cross-validation unit for extracting the film thickness deviation prediction value from the dangerous grid for cross-validation, and if the film thickness deviation prediction value does not satisfy the preset film thickness deviation interval, the validation fails, and a three-dimensional compensation vector is generated.

[0088] In the following, the specific configuration of the three-dimensional compensation vector generation module 50 will be described in detail. As described above, when any one of the cross-validation grids fails the validation, a three-dimensional compensation vector is generated, and the three-dimensional compensation vector generation module 50 can further include: a temperature information extraction unit for extracting the absolute temperature and the temperature rise rate from the cross-validation grids; a dangerous grid marking unit for marking the corresponding grid as a dangerous grid if the absolute temperature does not satisfy the preset temperature interval and / or the temperature rise rate does not satisfy the preset temperature rise rate interval; a cross-validation unit for extracting the film thickness deviation prediction value from the dangerous grid for cross-validation, and if the film thickness deviation prediction value does not satisfy the preset film thickness deviation interval, the validation fails, and a three-dimensional compensation vector is generated.

[0089] In the following, the specific configuration of the three-dimensional compensation vector generation module 50 will be described in detail. As described above, when any one of the cross-validation grids fails the validation, a three-dimensional compensation vector is generated, and the three-dimensional compensation vector generation module 50 can further include: a temperature information extraction unit for extracting the absolute temperature and the temperature rise rate from the cross-validation grids; a dangerous grid marking unit for marking the corresponding grid as a dangerous grid if the absolute temperature does not satisfy the preset temperature interval and / or the temperature rise rate does not satisfy the preset temperature rise rate interval; a cross-validation unit for extracting the film thickness deviation prediction value from the dangerous grid for cross-validation, and if the film thickness deviation prediction value does not satisfy the preset film thickness deviation interval, the validation fails, and a three-dimensional compensation vector is generated.

[0090] The NCVM non-conductive vacuum coating process execution monitoring system provided by the embodiments of the present application can execute the NCVM non-conductive vacuum coating process execution monitoring method provided by any of the embodiments of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0091] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or server, and the various units and modules are only divided according to the functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of the functional units are only for the convenience of mutual differentiation, and do not limit the protection scope of the present application.

[0092] The above specific embodiments do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application. In some cases, the actions or steps described in the present application can be executed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

Claims

1. A method for monitoring the execution of an NCVM non-conductive vacuum coating process, characterized in that, The method includes: The substrate surface inside the vacuum coating cavity is dynamically divided into detection unit grids to obtain an adaptive grid substrate surface; The surface of the adaptive grid substrate is scanned in real time using an infrared thermal imager to obtain the dynamic temperature distribution matrix of the substrate; Based on the mapping relationship between coating thickness and temperature gradient, the dynamic temperature distribution matrix of the substrate is analyzed and predicted to obtain a coating uniformity prediction map. The dynamic temperature distribution matrix of the substrate and the predicted uniformity of the coating are aligned and fused to obtain a cross-validation grid. When any one of the cross-validation grids fails validation, a three-dimensional compensation vector is generated. The process parameters of the vacuum coating equipment are adjusted in real time according to the three-dimensional compensation vector. Specifically, based on the mapping relationship between coating thickness and temperature gradient, the dynamic temperature distribution matrix of the substrate is analyzed and predicted to obtain a coating uniformity prediction map, including: Acquire historical coating data, including historical substrate dynamic temperature distribution matrix and historical film thickness deviation matrix, where film thickness deviation refers to the difference between actual film thickness and standard film thickness; The temperature gradient of adjacent grids is calculated based on the historical substrate dynamic temperature distribution matrix to obtain the historical substrate temperature gradient distribution matrix. The historical substrate temperature gradient distribution matrix and the historical film thickness deviation matrix are input into a convolutional neural network model for training and learning to obtain the mapping relationship between film thickness and temperature gradient. Based on the mapping relationship, the dynamic temperature distribution matrix of the substrate is analyzed and predicted to obtain a coating uniformity prediction map. Specifically, the historical substrate temperature gradient distribution matrix and the historical film thickness deviation matrix are input into a convolutional neural network model for training to obtain the mapping relationship between coating thickness and temperature gradient, including: Construct a two-branch convolutional neural network model, wherein the two-branch convolutional neural network model includes a first branch and a second branch; The first branch receives the historical substrate temperature gradient distribution matrix and extracts spatial features through a convolution kernel; The second branch receives the historical substrate temperature gradient distribution matrix at the same location in the previous N production batches, and extracts the time features through the LSTM layer, where N≥3; The spatial features and the temporal features are weighted and concatenated to obtain the film thickness deviation prediction matrix; When the mean square error of the film thickness deviation prediction matrix and the historical film thickness deviation matrix is ​​lower than the preset mean square error threshold and the prediction accuracy reaches the preset accuracy threshold, training is stopped and the mapping relationship between film thickness and temperature gradient is derived. When any one of the cross-validation grids fails validation, a three-dimensional compensation vector is generated, including: Extract the absolute temperature and temperature rise rate from the cross-validation grid; If the absolute temperature does not meet the preset temperature range and / or the temperature rise rate does not meet the preset temperature rise rate range, the corresponding grid will be marked as a dangerous grid. The predicted film thickness deviation is extracted from the dangerous grid and cross-validated. If the predicted film thickness deviation does not meet the preset film thickness deviation range, the validation fails and a three-dimensional compensation vector is generated. In addition to failing the verification, the temperature-film thickness co-verification condition must also be met. The co-verification condition is as follows: The temperature anomaly is calculated based on the substrate material, the absolute temperature, and the rate of temperature rise. The film thickness anomaly is calculated based on the substrate material and the predicted film thickness deviation. Multiply the temperature anomaly and the film thickness anomaly to obtain the synergy factor; If the temperature anomaly is greater than a preset temperature anomaly threshold, the film thickness anomaly is greater than a preset film thickness anomaly threshold, and the synergy factor is greater than a preset synergy factor threshold, then the verification is deemed unsuccessful. When any one of the cross-validation grids fails validation, a three-dimensional compensation vector is generated, including: Extract temperature anomaly and film thickness anomaly information from the grid that failed the verification; Calculate the angle compensation amount of the plasma gun based on the temperature anomaly information; Based on the film thickness anomaly information, calculate the sputtering power compensation and residence time compensation. The angle compensation, sputtering power compensation, and dwell time compensation are integrated to generate the three-dimensional compensation vector.

2. The method for monitoring the execution of an NCVM non-conductive vacuum coating process as described in claim 1, characterized in that, The substrate surface within the vacuum coating cavity is dynamically divided into detection unit grids to obtain an adaptive grid substrate surface, including: The conversion relationship between the surface curvature of the preset substrate and the mesh density; The curvature of multiple surfaces of the current substrate is acquired in real time, and the adaptive mesh density is calculated based on the transformation relationship. The substrate surface inside the vacuum coating cavity is dynamically divided into detection unit grids according to the adaptive grid density, and the geometric center point of each grid is set as the temperature detection reference point to obtain the adaptive grid substrate surface.

3. The method for monitoring the execution of an NCVM non-conductive vacuum coating process as described in claim 2, characterized in that, The substrate surface within the vacuum coating cavity is dynamically divided into detection unit grids to obtain an adaptive grid substrate surface, which also includes: Based on the dynamic temperature distribution matrix of the substrate, the temperature change rate is extracted to obtain the dynamic temperature change rate matrix of the substrate. Traverse the dynamic temperature change rate matrix of the substrate and extract the cell grids with temperature change rates greater than a preset temperature change rate threshold to obtain temperature anomaly grids. The adaptive mesh substrate surface is updated after the temperature anomaly mesh is encrypted and reconstructed.

4. The method for monitoring the execution of an NCVM non-conductive vacuum coating process as described in claim 1, characterized in that, Also includes: Calculate the edge compensation coefficient based on the radius of curvature of the mesh in the edge region of the substrate; The three-dimensional compensation vector is adjusted according to the edge compensation coefficient to obtain the three-dimensional enhancement compensation vector; After performing compensation on the edge region mesh according to the three-dimensional strengthening compensation vector, the edge secondary verification is initiated, and the edge region mesh is re-measured using a laser thickness gauge to obtain the re-measurement result. The edge region mesh is optimized by coating based on the retest results.

5. A monitoring system for the execution of an NCVM non-conductive vacuum coating process, characterized in that, The system is used to implement the NCVM non-conductive vacuum coating process execution monitoring method according to any one of claims 1-4, the system comprising: The detection unit grid dynamic division module is used to dynamically divide the substrate surface inside the vacuum coating cavity into detection unit grids to obtain an adaptive grid substrate surface. An infrared thermal imaging scanning module is used to scan the surface of the adaptive grid substrate in real time using an infrared thermal imager to obtain a dynamic temperature distribution matrix of the substrate. The coating uniformity prediction module is used to analyze and predict the dynamic temperature distribution matrix of the substrate based on the mapping relationship between coating thickness and temperature gradient, and obtain a coating uniformity prediction map. The alignment and fusion processing module is used to align and fuse the dynamic temperature distribution matrix of the substrate and the coating uniformity prediction map to obtain a cross-validation grid. The three-dimensional compensation vector generation module is used to generate a three-dimensional compensation vector when any one of the cross-validation grids fails the validation; the process parameter control module is used to adjust the process parameters of the vacuum coating equipment in real time according to the three-dimensional compensation vector.

Images