Modular defect detection method and system for IC (integrated circuit) carrier plate and computer equipment

Abstract

The invention relates to a modular defect detection method and system for an IC carrier plate and computer equipment, and the method comprises the steps: obtaining original image data in the production process of the IC carrier plate, carrying out the preprocessing of the original image data, obtaining the image data of the IC carrier plate, constructing a defect recognition model, introducing an attention mechanism into the defect recognition model, and carrying out the recognition of the defect of the IC carrier plate. Obtaining an identification result of the IC carrier plate; the method comprises the following steps: acquiring process parameter data and equipment operation state data in IC support plate production, performing association analysis with historical defect data, mining a potential relationship between the data, and determining key parameters having great influence on defects; constructing a defect prediction model, and judging whether the IC carrier plate has defects or not; and fusing the results of the defect identification model and the defect prediction model, and generating a corresponding process parameter adjustment suggestion or equipment maintenance strategy in combination with a preset rule, so as to achieve the purposes of accurately detecting the defects of the IC carrier plate, predicting potential risks, optimizing process and equipment maintenance and improving product quality and production efficiency.

Description

Technical Field

[0001] The present invention relates to the field of IC substrate detection technology, and in particular to a modular defect detection method, system and computer equipment for IC substrates. Background Art

[0002] In the field of IC substrate production, the complexity of the process, the diversity of material properties and the interference of environmental factors make it very easy for various defects such as pinholes, scratches, stains, etc. to appear on the surface of the substrate. Once these defects are not detected in a timely and accurate manner, they will have a serious impact on the electrical performance and reliability of the substrate, and may even cause the downstream chips to fail, thereby causing a large amount of economic losses. At present, the traditional defect detection methods mainly include manual visual inspection and dedicated optical inspection equipment. However, manual visual inspection has obvious disadvantages. Its detection speed is slow and it is easily interfered by human subjective factors, which makes it difficult to guarantee the accuracy and stability of the detection results. Although dedicated optical inspection equipment has improved the detection efficiency to a certain extent, it lacks the ability to intelligently identify and classify defect types of various forms. In the actual operation process, a large number of manual participation in judgment and screening is still required, which not only increases labor costs, but also reduces the overall detection efficiency.

[0003] In addition, the production process parameters of IC substrates are extremely numerous, and there are large differences in material properties, dimensional accuracy, etc. between different batches of products, which undoubtedly puts forward higher requirements for the cause judgment and prevention of defects. Relying solely on the post-detection method of defects that have occurred, it is impossible to obtain the dynamic change trend of defects in a timely manner, and it is difficult to achieve effective prediction and early warning of defects. The massive test data generated during the production process has not been fully and effectively utilized, and its data value urgently needs to be deeply mined. Summary of the invention

[0004] The main purpose of the present invention is to provide a modular defect detection method, system and computer equipment for IC substrates, so as to achieve the purpose of accurately detecting IC substrate defects, predicting potential risks, optimizing process and equipment maintenance to improve product quality and production efficiency. To achieve the above purpose, the present invention provides a modular defect detection method for IC substrates, comprising the following steps:

[0005] Acquire original image data in the production process of the IC substrate, preprocess the original image data, and obtain preprocessed IC substrate image data;

[0006] According to the IC substrate image data, a defect recognition model is constructed, and an attention mechanism is introduced into the defect recognition model to generate an attention weight map to obtain an IC substrate recognition result;

[0007] Acquire process parameter data and equipment operation status data in the production process of the IC substrate, perform correlation analysis on the process parameter data, the equipment operation status data and historical defect data, mine the potential relationship between the data, and determine the key parameters that have a greater impact on defect formation;

[0008] Constructing a defect prediction model based on the process parameter data, the equipment operation status data and the key parameters, and determining whether the IC substrate has defects;

[0009] The results of the defect recognition model and the defect prediction model are fused, and based on the fused results and in combination with pre-set rules, corresponding process parameter adjustment suggestions or equipment maintenance strategies are generated, wherein the process parameter adjustment suggestions include specific adjustment values ​​of temperature and pressure, and the equipment maintenance strategy includes equipment maintenance time and maintenance items.

[0010] Furthermore, the step of acquiring original image data in the IC substrate production process, preprocessing the original image data, and obtaining preprocessed IC substrate image data comprises:

[0011] Acquire raw image data in the IC substrate production process, and preprocess the raw image data, including denoising, enhancement, and standardization operations;

[0012] According to a preset range of IC substrate geometric characteristic parameters, determining whether each connected area is a qualified IC substrate area;

[0013] The IC substrate area judged as qualified is extracted to obtain pre-processed IC substrate image data.

[0014] Furthermore, the step of constructing a defect recognition model according to the IC substrate image data, introducing an attention mechanism into the defect recognition model, generating an attention weight map, and obtaining an IC substrate recognition result includes:

[0015] According to the preprocessed IC substrate image data, a defect recognition model is constructed by using a convolutional neural network, an attention mechanism is introduced into the convolutional neural network, and the weights of different areas are automatically learned through the attention mechanism to highlight the defective areas;

[0016] During the training process of the defect recognition model, the attention weight is calculated based on the marked defect area to obtain an attention weight map, where the area with a high weight in the attention weight map indicates that there may be defects;

[0017] Fusing the attention weight map with the original image data to obtain fused image data, wherein the defect area is highlighted in the fused image data;

[0018] Inputting the fused image data into the defect recognition model for prediction;

[0019] Output the prediction results of the defect recognition model, including the location, type and confidence level of the defect, to determine whether the IC substrate has defects.

[0020] Furthermore, the step of fusing the attention weight map with the original image data to obtain fused image data, and after highlighting the defect area in the fused image data, further includes:

[0021] Using data enhancement technology, the original image is rotated, translated, and scaled to generate training samples and expand the training data set;

[0022] Adjusting the model structure and hyperparameters in the convolutional neural network model according to the fused image data and the expanded training data set;

[0023] The convolutional neural network model is optimized through an adaptive learning rate adjustment algorithm.

[0024] Furthermore, the step of obtaining process parameter data and equipment operation status data in the production process of the IC substrate, correlating the process parameter data, the equipment operation status data and historical defect data, mining the potential relationship between the data, and determining the key parameters that have a greater impact on defect formation includes:

[0025] Obtain process parameter data and equipment operation status data during IC substrate production, and pre-process the data, including data cleaning, data integration and data transformation operations

[0026] According to the characteristics of the IC substrate production process, the corresponding data mining algorithm is selected to calculate the influence of various process parameters and equipment operating status on the formation of IC substrate defects, and determine the key parameters based on the influence degree. The data mining algorithm is used to analyze the correlation between process parameters and equipment operating status and historical defect data.

[0027] Furthermore, the step of constructing a defect prediction model based on the process parameter data, the equipment operation status data and the key parameters, and determining whether the IC substrate has defects includes:

[0028] Preprocessing the data according to the process parameter data, the equipment operation status data and the key parameters to obtain a preprocessed data set;

[0029] Using feature selection algorithm, feature selection is performed on the preprocessed data set to select feature subsets with high correlation with substrate defects;

[0030] A decision tree algorithm is used to construct a model, and the model is trained based on the feature subset to obtain a defect prediction model;

[0031] Acquire process parameter data, equipment operation status data and key parameters of the IC substrate, perform preprocessing operations on the acquired data, and obtain preprocessed data to be predicted;

[0032] The preprocessed data to be predicted is input into the trained defect prediction model. Through the operation of the model, it is determined whether the IC substrate has defects. If the defect probability output by the model is greater than a preset threshold, it is determined that the IC substrate has defects.

[0033] Furthermore, the step of fusing the results of the defect recognition model and the defect prediction model, and generating corresponding process parameter adjustment suggestions or equipment maintenance strategies based on the fused results and in combination with pre-set rules, includes:

[0034] Obtain the output results of the defect recognition model and the defect prediction model, perform feature fusion on the results of the two models, and obtain the fused defect analysis results;

[0035] According to the preset process parameter adjustment rules, determine whether the fused defect analysis results meet the conditions for triggering process parameter adjustment. If so, determine the specific adjustment values ​​of temperature and pressure based on the defect type and severity, and generate process parameter adjustment suggestions;

[0036] According to the preset equipment maintenance strategy rules, determine whether the fused defect analysis results meet the conditions for triggering equipment maintenance. If so, determine the equipment maintenance time and maintenance items based on the defect type and severity as well as the equipment's historical maintenance records, and generate an equipment maintenance strategy.

[0037] The present invention also provides a modular defect detection system for an IC substrate, comprising:

[0038] An image preprocessing module, used to obtain original image data in the production process of the IC substrate, preprocess the original image data, and obtain preprocessed IC substrate image data;

[0039] A recognition model building module is used to build a defect recognition model according to the IC substrate image data, introduce an attention mechanism into the defect recognition model, generate an attention weight map, and obtain an IC substrate recognition result;

[0040] A data analysis module, used to obtain process parameter data and equipment operation status data in the production process of the IC substrate, associate and analyze the process parameter data, the equipment operation status data and historical defect data, mine the potential relationship between the data, and determine the key parameters that have a greater impact on defect formation;

[0041] A prediction model building module, used to build a defect prediction model based on the process parameter data, the equipment operation status data and the key parameters, and determine whether the IC substrate has defects;

[0042] The strategy generation module is used to fuse the results of the defect recognition model and the defect prediction model, and based on the fused results and in combination with pre-set rules, generate corresponding process parameter adjustment suggestions or equipment maintenance strategies.

[0043] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned modular defect detection method for IC substrates when executing the computer program.

[0044] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-mentioned modular defect detection method for IC substrates.

[0045] The modular defect detection method, system and computer equipment for IC substrates provided by the present invention have the following beneficial effects: The present invention first uses a defect recognition model to quickly process image data, screen out surface defective products in a timely manner, avoid them from entering subsequent processes, prevent waste of resources, and greatly improve production efficiency. In this process, by introducing an attention mechanism, the defect location, type and confidence level are accurately judged to provide reliable samples for defect identification, thereby reducing the overall detection misjudgment and missed judgment rate. Based on the recognition results, combined with process and equipment data, key parameters are mined to build a prediction model, and potential defects are warned in advance, so as to facilitate timely adjustment of the process and prevent defects from occurring again. The results of the two models are integrated to generate process adjustment suggestions and equipment maintenance strategies according to preset rules, accurately adjust parameters, and reasonably arrange maintenance to improve product yield and ensure production stability and product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flow chart of a modular defect detection method for an IC substrate in one embodiment of the present invention;

[0047] Figure 2 is a block diagram of a modular defect detection system for IC substrates according to an embodiment of the present invention;

[0048] Figure 3It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0049] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0051] Reference Figure 1 , is a flow chart of a modular defect detection method for IC substrates proposed by the present invention, comprising the following steps:

[0052] S1, acquiring original image data in the production process of the IC substrate, and preprocessing the original image data to obtain preprocessed IC substrate image data;

[0053] S2, constructing a defect recognition model according to the IC substrate image data, and introducing an attention mechanism into the defect recognition model to generate an attention weight map to obtain an IC substrate recognition result;

[0054] S3, obtaining process parameter data and equipment operation status data in the production process of the IC substrate, performing correlation analysis on the process parameter data, the equipment operation status data and the historical defect data, mining the potential relationship between the data, and determining the key parameters that have a greater impact on defect formation;

[0055] S4, constructing a defect prediction model according to the process parameter data, the equipment operation status data and the key parameters, and determining whether the IC substrate has defects;

[0056] S5, fusing the results of the defect recognition model and the defect prediction model, and based on the fused results and in combination with pre-set rules, generating corresponding process parameter adjustment suggestions or equipment maintenance strategies, wherein the process parameter adjustment suggestions include specific adjustment values ​​of temperature and pressure, and the equipment maintenance strategies include equipment maintenance time and maintenance items.

[0057] As described in step S1 above, the original image data in the production process of the IC carrier is obtained, and the original image data is preprocessed to obtain the preprocessed IC carrier image data. The image data of the IC carrier in the production process is collected. These images are unprocessed raw data and may contain various noises, uneven illumination and other problems. Denoising, enhancement and standardization operations are performed on the original image data. Due to factors such as the production environment and equipment, the original image may contain noise. The denoising operation can reduce noise interference, improve image quality, and make the subsequent feature extraction of the IC carrier more accurate. For example, algorithms such as Gaussian filtering are used to remove Gaussian noise in the image, enhance certain features of the image, make the details of the IC carrier clearer, and facilitate subsequent identification. For example, the contrast of the image is enhanced by histogram equalization, so that different areas in the image are easier to distinguish. The image data is normalized according to certain standards so that the image data obtained in different batches and under different conditions have unified specifications and ranges, which is convenient for subsequent model processing. According to the preset range of geometric feature parameters of the IC carrier, each connected area in the image is judged to determine whether it is a qualified IC carrier area. The qualified IC substrate areas are extracted, and the image data composed of these areas is the pre-processed IC substrate image data, which can be used to build defect recognition models and other operations later.

[0058] As described in step S2 above, a defect recognition model is constructed according to the IC carrier image data, and an attention mechanism is introduced into the defect recognition model to generate an attention weight map to obtain the recognition result of the IC carrier. Based on the IC carrier image data preprocessed in step S1, a defect recognition model is constructed using a convolutional neural network (CNN). The attention mechanism is introduced into the constructed convolutional neural network. The core function of the attention mechanism is to allow the model to automatically pay attention to the importance of different areas when processing images, that is, to assign different weights to different areas of the image through learning. In IC carrier defect recognition, the attention mechanism can highlight the defective area in the image, so that the model can focus more on the part where defects may exist, and improve the sensitivity and recognition ability of defects. In the training process of the defect recognition model, the attention weight is calculated based on the marked defective area. The marked defective area is the part of the image that does have defects that is marked in advance by artificial or other means. Based on these annotation information, the model calculates the importance of each area for identifying defects through a specific algorithm, and then generates an attention weight map. In this weight map, the area with a high weight means that the area is more likely to have defects. The generated attention weight map is fused with the original image data to obtain the fused image data. In this fused image, since the attention weight map highlights the area where defects may exist, the defective area will be more clearly displayed in the image. The fused image data is input into the constructed defect recognition model for prediction. The model judges whether there are defects in the IC substrate in the image based on the learned features and patterns, and outputs the prediction results. The prediction results include the location, type and confidence of the defect. For example, the model may output that there is a scratch defect at a specific coordinate position of the IC substrate, and the confidence in this judgment is 90%.

[0059] As described in step S3 above, the process parameter data and equipment operation status data in the production process of the IC carrier are obtained, and the process parameter data, the equipment operation status data and the historical defect data are correlated and analyzed to mine the potential relationship between the data and determine the key parameters that have a greater impact on the formation of defects. The process parameter data in the production process of the IC carrier are collected, such as temperature, pressure, time, etc. At the same time, the equipment operation status data, such as the speed, vibration of the equipment, the working status of each component, etc. are obtained. The obtained process parameter data and equipment operation status data are preprocessed, including data cleaning, data integration and data transformation operations. According to the characteristics of the IC carrier production process, a suitable data mining algorithm is selected. These algorithms are used to analyze the correlation between process parameters, equipment operation status and historical defect data. For example, a decision tree algorithm, an association rule mining algorithm, etc. may be selected to analyze a large amount of historical data through an algorithm to quantify the contribution of each factor to the formation of defects. For example, through analysis, it is found that the fluctuation of temperature within a certain range is positively correlated with the probability of pinhole defects in the IC carrier, and the specific influence degree of temperature change on the formation of pinhole defects is calculated. According to the calculated influence, the key parameters with the greatest influence on defect formation are determined. For example, if the analysis results show that the pressure parameter and the vibration frequency of a certain component of the equipment have a greater influence on the scratch defect of the IC substrate than other factors, then the pressure and the vibration frequency of the component are determined as key parameters, and these parameters need to be paid attention to and managed in subsequent production to reduce the possibility of scratch defects.

[0060] As described in step S4 above, a defect prediction model is constructed based on the process parameter data, the equipment operation status data and the key parameters, and it is determined whether the IC substrate has defects. Based on the acquired process parameter data, equipment operation status data and the key parameters determined in step S3, these data are preprocessed to obtain a data set suitable for model processing. A feature selection algorithm is used to conduct an in-depth analysis of the preprocessed data set, and a feature subset with a high correlation with substrate defects is selected from a large number of features (i.e., various process parameters, equipment operation status parameters, etc.), thereby removing redundant or irrelevant features, reducing data dimensions, and reducing the complexity and amount of model training; and the model can be more focused on key features, improving the prediction accuracy and efficiency of the model. A decision tree algorithm is used to construct a defect prediction model. Based on the previously selected feature subset, the decision tree model is trained. During the training process, the model will learn the intrinsic relationship between different feature combinations and IC substrate defects from the data, and adjust its own parameters and structure to achieve the best prediction performance.

[0061] Obtain the process parameter data, equipment operation status data and key parameters of the current IC substrate to be tested, and also preprocess the acquired data to ensure that it is consistent with the data format and quality when the model is trained. Input the preprocessed data to be predicted into the defect prediction model that has been trained. The model calculates and analyzes the input data based on the rules and patterns learned during the training process. The model outputs a defect probability value and compares the value with the preset threshold. If the defect probability output by the model is greater than the preset threshold, this indicates that according to the current process parameters, equipment operation status and other data conditions, the probability of the IC substrate being defective is high, so the IC substrate is judged to be defective; conversely, if the defect probability is less than or equal to the preset threshold, it is judged that the IC substrate is not defective. In this way, the prediction and judgment of whether the IC substrate is defective is achieved.

[0062] As described in step S5 above, the results of the defect recognition model and the defect prediction model are fused, and based on the fused results and combined with pre-set rules, corresponding process parameter adjustment suggestions or equipment maintenance strategies are generated, wherein the process parameter adjustment suggestions include specific adjustment values ​​of temperature and pressure, and the equipment maintenance strategy includes equipment maintenance time and maintenance items. The output results of the defect recognition model and the defect prediction model are obtained. The defect recognition model mainly outputs information related to the defects that have appeared on the IC substrate, such as the location, type, and confidence of the defects; the defect prediction model determines whether the IC substrate has defects and outputs the probability of defects based on data such as process parameters and equipment operating status.

[0063] The results of the two models are feature fused. For example, the feature vectors output by the two models are concatenated, or weights are assigned according to the importance of the results of different models and then weighted summed. Through feature fusion, a fused defect analysis result that comprehensively reflects the defects of the IC substrate is obtained. This result integrates the judgment information of the IC substrate defects from different angles of the two models, which is more comprehensive and accurate than the results of a single model. According to the preset process parameter adjustment rules, the fused defect analysis results are judged to see whether the conditions for triggering process parameter adjustment are met. These preset rules are formulated based on production experience, experimental data and understanding of process principles. For example, if the fusion result shows that a certain type of defect has occurred on the IC substrate, and the defect is closely related to abnormal changes in temperature and pressure in historical data, the conditions for triggering process parameter adjustment are met. When the conditions are met, the specific adjustment values ​​of temperature and pressure are further determined according to the defect type and severity. Different types of defects may require process parameter adjustments in different directions and amplitudes. For example, if the defect is caused by material deformation due to excessive temperature, the temperature may need to be reduced by a certain degree; the higher the severity of the defect, the greater the adjustment amplitude may be. According to the preset equipment maintenance strategy rules, determine whether the fused defect analysis results meet the conditions for triggering equipment maintenance. These rules are also formulated based on actual production conditions and equipment characteristics. For example, when the defect analysis results suggest that a certain defect occurs frequently and is related to a specific operating state or component of the equipment, the trigger condition is met. If the conditions are met, the equipment maintenance time and maintenance items are determined based on the defect type and severity and the historical maintenance records of the equipment. Different types of defects may indicate problems with different parts of the equipment. For example, if the defect is related to pressure instability of the equipment, and the historical maintenance records show that the pressure control component has had similar failures, then the maintenance items may include inspection and repair of the pressure control component; the higher the severity of the defect, the more timely maintenance may be required, that is, the maintenance time will be earlier.

[0064] In one embodiment, a high-definition camera is used to obtain image data during the production process of an IC carrier board, with an image resolution of 1920×1080 pixels and a color depth of 24 bits. The color image is converted into a grayscale image using the OpenCV image processing library, and the conversion formula is Gray=R×299+G×587+B×114. A 5×5 median filter algorithm is used to denoise the grayscale image to remove salt and pepper noise and Gaussian noise. According to the grayscale characteristics of the IC carrier board, the binarization threshold is set to 128, and the pixels with grayscale values ​​greater than or equal to 128 are set to 255, and the pixels with grayscale values ​​less than 128 are set to 0 to obtain a black and white binary image. First, a 3×3 rectangular structural element is used to corrode the binary image to eliminate small white noise points. Then a 5×5 rectangular structural element is used to dilate the image to fill black breakpoints and holes. The connected regions in the binary image are extracted by the connected domain analysis algorithm, and the geometric feature parameters such as the area, perimeter and circularity of each connected region are calculated. According to the IC substrate design specifications, the area range of the qualified IC substrate area is set to 100-400 square millimeters, the perimeter range is 40-80 millimeters, and the circularity range is 85-95. The connected areas that meet the parameter range are extracted to obtain the preprocessed qualified IC substrate image data.

[0065] The histogram equalization algorithm is used to enhance the original image data of the IC substrate to improve the contrast and clarity. The median filter algorithm is used to remove noise and retain the image edge and texture features. The image is normalized and the pixel value is scaled to between 0 and 1.

[0066]

[0067] I norm represents the normalized image, I represents the original image, and I min and I max Represent the minimum and maximum pixel values ​​of the image respectively. This formula is used to scale the pixel value to between 0 and 1. Based on the preprocessed image data, the ResNet-50 convolutional neural network is used to build a defect recognition model. The channel attention mechanism and spatial attention mechanism are introduced after the last convolutional layer of the network to automatically learn the weights of different regions and highlight the defective areas. During the model training process, the attention weight is calculated according to the marked defective area. The weight value of the defective area is set to 8, and the weight value of the non-defective area is set to 2, and an attention weight map of the same size as the original image is obtained. The attention weight map is weightedly fused with the original image at the pixel level, and the formula is:

[0068] I fused =I·(1+αW)

[0069] I fusedRepresents the fused image, I represents the original image, W represents the attention weight map, and α represents the fusion coefficient. This formula is used to perform pixel-level weighted fusion of the attention weight and the original image. The pixel values ​​in the defective area are enlarged, and the pixel values ​​in the non-defective area are reduced to obtain the fused image data. The fused image data is input into the defect recognition model for prediction, and the defect category probability is output through the softmax function. The confidence threshold is set to 8. When the probability of a certain category is greater than 8, it is judged that this type of defect exists in the area, and the defect location coordinates are output. 500 training data of different types of IC substrates are added every week. Use data enhancement technology to expand the data set, such as rotating the original image (within ±10 degrees), translating (within ±10 pixels), and scaling (within 8 to 2 times). The grid search algorithm is used to adjust the model hyperparameters, such as the learning rate, batch size, and regularization coefficient. At the same time, an adaptive learning rate adjustment algorithm is introduced to prevent the model from overfitting. The formula is:

[0070]

[0071] η t represents the learning rate of the tth iteration, η0 represents the initial learning rate, β represents the decay rate, and t represents the number of iterations. This formula is used to dynamically adjust the learning rate to prevent the model from overfitting and improve the defect recognition accuracy and generalization ability of the model.

[0072] In the production process of IC substrates, process parameter data (such as temperature, pressure, speed, etc.) and equipment operation status data are collected in real time through online sensors with a sampling frequency of 1000Hz and transmitted to the database server for storage via industrial Ethernet. The maximum entropy model is used to estimate and fill missing data. The local anomaly factor algorithm is used to detect and eliminate abnormal data. Data standardization and normalization are performed to convert data from different sources and scales into a unified format and range. The Apriori association rule mining algorithm is used to perform association analysis on process parameters, equipment status and defect data. The support threshold is set to 0.5, the confidence threshold is set to 0.8, and the minimum association rule length is 3. For example, it is found that in the electroplating process of IC substrates, the current density exceeds 5A / dm 2 When the plating solution temperature exceeds 60°C, pinhole defects are likely to appear on the substrate surface, with a defect rate of over 5%. Through experimental optimization analysis, it was determined that the current density should be controlled at 2A / dm 2Below, the plating solution temperature is controlled below 55°C, which can reduce the pinhole defect rate to less than 1%. The current density and plating solution temperature are the key parameters. Data cleaning is performed on process parameter data, equipment operation status data and key parameters to remove missing values, outliers (using the 3σ criterion to remove data exceeding plus or minus 3 times the standard deviation) and other noise data. The minimum-maximum normalization method is used to map the data to the [0,1] interval to eliminate dimensional differences. A recursive feature elimination algorithm is used, and the number of recursions is set to 10 times. 10% of the features are eliminated each iteration, and the top 20 features with the greatest impact on product defects are selected as feature subsets (based on the Pearson correlation score between features and defects, features with correlations less than 1 are iteratively deleted).

[0073]

[0074] r represents the Pearson correlation coefficient, which is used to evaluate the correlation between the feature and the defect. i and i represent the characteristic value and defect value respectively, and They represent their average values ​​respectively.

[0075] Based on the feature subset, the CART decision tree algorithm is used to build a defect prediction model, the maximum depth of the decision tree is set to 5, the minimum number of leaf node samples is set to 10, and through 10-fold cross validation, a model with an average prediction accuracy of 92% is obtained (or the support vector machine algorithm is used to build a model, and the hyperparameters are optimized through 5-fold cross validation and grid search, such as the penalty coefficient C = 10, and the kernel function is a Gaussian kernel, so that the prediction accuracy reaches more than 95%). Obtain the relevant data of the product to be predicted, perform the same data preprocessing as the training data, and input the trained decision tree model for prediction. The formula is:

[0076]

[0077] This formula represents the defect determination rule. When the defect probability P(defect) output by the model is greater than 6, the product is judged to have defects (Defect=1), otherwise it is judged to be non-defective (Defect=0). For products predicted to be defective, the key factors causing the defects are determined according to the importance scores of each feature in the feature subset (such as Gini importance score), such as temperature exceeding 450℃, pressure below 2MPa, etc., and defect location analysis is performed. According to the analysis results, the corresponding process parameters or equipment status are adjusted, such as controlling the temperature below 420℃ and increasing the pressure to 5MPa, and predicting the defect probability again to reduce it to below 2, and the product qualification rate is increased to more than 98%. The output results of the defect recognition model and the defect prediction model are obtained. Assume that the defect recognition model outputs the defect type as crack and the severity is 8; the defect prediction model outputs the defect type as pore and the severity is 6. The weighted average method is used for feature fusion, the crack weight is set to 7, and the pore weight is set to 3. The defect analysis results after fusion are obtained: the crack severity is 7.6 and the pore severity is 1.8.

[0078] According to the preset process parameter adjustment rules, when the crack severity is greater than 7, the temperature needs to be reduced by 20℃ and the pressure needs to be increased by 5MPa; when the pore severity is greater than 2, the temperature needs to be increased by 10℃ and the pressure needs to be reduced by 8MPa. Since the crack severity is 7.6 and meets the conditions, the initial process parameter adjustment suggestion is to reduce the temperature by 20℃ and increase the pressure by 5MPa. Further combined with the real-time operating parameters of the equipment and the historical trend data of the past 24 hours, the gradient descent algorithm is used to optimize the adjustment range (the learning rate is set to 0.1, and iterates 100 times), and the final process parameter adjustment suggestion is to reduce the temperature by 15℃ and increase the pressure by 2.8MPa. According to the preset equipment maintenance strategy rules, when the crack severity is greater than 7.5 and the equipment operation time exceeds 1000 hours, the equipment needs to be overhauled within 3 days, and the maintenance items include replacing parts and calibrating parameters. Query the historical maintenance records of the equipment. The equipment has been running for 1200 hours. The initial equipment maintenance strategy is to overhaul the equipment within 3 days, replace parts and calibrate parameters. Combined with the equipment health status assessment results and risk prediction results, the maintenance time and maintenance items are dynamically adjusted through the reinforcement learning algorithm (with minimizing the risk of equipment failure as the reward function). After 500 iterations, the optimal equipment maintenance strategy is to perform equipment maintenance, replace parts, calibrate parameters, and add lubricating oil within 1 day.

[0079] Reference Figure 2 , is a structural block diagram of a modular defect detection system for IC substrates in one embodiment of the present invention, comprising:

[0080] An image preprocessing module, used to obtain original image data in the production process of the IC substrate, preprocess the original image data, and obtain preprocessed IC substrate image data;

[0081] A recognition model building module is used to build a defect recognition model according to the IC substrate image data, introduce an attention mechanism into the defect recognition model, generate an attention weight map, and obtain an IC substrate recognition result;

[0082] A data analysis module, used to obtain process parameter data and equipment operation status data in the production process of the IC substrate, associate and analyze the process parameter data, the equipment operation status data and historical defect data, mine the potential relationship between the data, and determine the key parameters that have a greater impact on defect formation;

[0083] A prediction model building module, used to build a defect prediction model based on the process parameter data, the equipment operation status data and the key parameters, and determine whether the IC substrate has defects;

[0084] The strategy generation module is used to fuse the results of the defect recognition model and the defect prediction model, and based on the fused results and in combination with pre-set rules, generate corresponding process parameter adjustment suggestions or equipment maintenance strategies.

[0085] For the specific implementation of each module in the above device example, please refer to the above method embodiment, which will not be repeated here.

[0086] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0087] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0088] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0089] In summary, the original image data in the production process of the IC substrate is obtained, and the original image data is preprocessed to obtain the preprocessed IC substrate image data; according to the IC substrate image data, a defect recognition model is constructed, and an attention mechanism is introduced into the defect recognition model to generate an attention weight map to obtain the recognition result of the IC substrate; the process parameter data and the equipment operation status data in the production process of the IC substrate are obtained, and the process parameter data, the equipment operation status data and the historical defect data are correlated and analyzed to mine the potential relationship between the data and determine the key parameters that have a greater impact on the formation of defects; according to the process parameter data, the equipment operation status data and the key parameters, a defect prediction model is constructed, and it is determined whether the IC substrate has defects; the results of the defect recognition model and the defect prediction model are fused, and based on the fused results and combined with pre-set rules, corresponding process parameter adjustment suggestions or equipment maintenance strategies are generated, the process parameter adjustment suggestions include specific adjustment values ​​of temperature and pressure, and the equipment maintenance strategy includes equipment maintenance time and maintenance items, so as to achieve the purpose of accurately detecting IC substrate defects, predicting potential risks, and optimizing process and equipment maintenance to improve product quality and production efficiency.

[0090] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0091] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0092] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A modular defect detection method for IC substrates, characterized in that: The following steps are involved: Acquire original image data in the production process of the IC substrate, preprocess the original image data, and obtain preprocessed IC substrate image data; According to the IC substrate image data, a defect recognition model is constructed, and an attention mechanism is introduced into the defect recognition model to generate an attention weight map to obtain an IC substrate recognition result; Acquire process parameter data and equipment operation status data in the production process of the IC substrate, perform correlation analysis on the process parameter data, the equipment operation status data and historical defect data, mine the potential relationship between the data, and determine the key parameters that have a greater impact on defect formation; Constructing a defect prediction model based on the process parameter data, the equipment operation status data and the key parameters, and determining whether the IC substrate has defects; The results of the defect recognition model and the defect prediction model are fused, and based on the fused results and in combination with pre-set rules, corresponding process parameter adjustment suggestions or equipment maintenance strategies are generated, wherein the process parameter adjustment suggestions include specific adjustment values ​​of temperature and pressure, and the equipment maintenance strategy includes equipment maintenance time and maintenance items.

2. The modular defect detection method for IC substrate according to claim 1, characterized in that: The step of acquiring original image data in the IC substrate production process, preprocessing the original image data, and obtaining preprocessed IC substrate image data comprises: Acquire raw image data in the IC substrate production process, and preprocess the raw image data, including denoising, enhancement, and standardization operations; According to a preset range of IC substrate geometric feature parameters, determining whether each connected area is a qualified IC substrate area; The IC substrate area judged as qualified is extracted to obtain pre-processed IC substrate image data.

3. The modular defect detection method for IC substrate according to claim 1, characterized in that: The step of constructing a defect recognition model according to the IC substrate image data, introducing an attention mechanism into the defect recognition model, generating an attention weight map, and obtaining an IC substrate recognition result comprises: According to the preprocessed IC substrate image data, a defect recognition model is constructed by using a convolutional neural network, an attention mechanism is introduced into the convolutional neural network, and the weights of different areas are automatically learned through the attention mechanism to highlight the defective areas; During the training process of the defect recognition model, the attention weight is calculated based on the marked defect area to obtain an attention weight map, where the area with a high weight in the attention weight map indicates that there may be defects; Fusing the attention weight map with the original image data to obtain fused image data, wherein the defect area is highlighted in the fused image data; Inputting the fused image data into the defect recognition model for prediction; Output the prediction results of the defect recognition model, including the location, type and confidence level of the defect, to determine whether the IC substrate has defects.

4. The modular defect detection method for IC substrate according to claim 3, characterized in that: The method further comprises: fusing the attention weight map with the original image data to obtain fused image data, and after highlighting the defect area in the fused image data, the method further comprises: Using data enhancement technology, the original image is rotated, translated, and scaled to generate training samples and expand the training data set; Adjusting the model structure and hyperparameters in the convolutional neural network model according to the fused image data and the expanded training data set; The convolutional neural network model is optimized through an adaptive learning rate adjustment algorithm.

5. The modular defect detection method for IC substrate according to claim 1, characterized in that: The step of obtaining process parameter data and equipment operation status data in the production process of the IC substrate, correlating the process parameter data, the equipment operation status data and historical defect data, mining the potential relationship between the data, and determining the key parameters that have a greater impact on defect formation includes: Obtain process parameter data and equipment operation status data during IC substrate production, and pre-process the data, including data cleaning, data integration and data transformation operations According to the characteristics of the IC substrate production process, the corresponding data mining algorithm is selected to calculate the influence of various process parameters and equipment operating status on the formation of IC substrate defects, and determine the key parameters based on the influence degree. The data mining algorithm is used to analyze the correlation between process parameters and equipment operating status and historical defect data.

6. The modular defect detection method for IC substrate according to claim 1, characterized in that: The step of constructing a defect prediction model based on the process parameter data, the equipment operation status data and the key parameters, and determining whether the IC substrate has defects includes: Preprocessing the data according to the process parameter data, the equipment operation status data and the key parameters to obtain a preprocessed data set; Using feature selection algorithm, feature selection is performed on the preprocessed data set to select feature subsets with high correlation with substrate defects; A decision tree algorithm is used to construct a model, and the model is trained based on the feature subset to obtain a defect prediction model; Acquire process parameter data, equipment operation status data and key parameters of the IC substrate, perform preprocessing operations on the acquired data, and obtain preprocessed data to be predicted; The preprocessed data to be predicted is input into the trained defect prediction model. Through the operation of the model, it is determined whether the IC substrate has defects. If the defect probability output by the model is greater than a preset threshold, it is determined that the IC substrate has defects.

7. The modular defect detection method for IC substrate according to claim 1, characterized in that: The step of fusing the results of the defect recognition model and the defect prediction model, and generating corresponding process parameter adjustment suggestions or equipment maintenance strategies based on the fused results and in combination with pre-set rules, comprises: Obtain the output results of the defect recognition model and the defect prediction model, perform feature fusion on the results of the two models, and obtain the fused defect analysis results; According to the preset process parameter adjustment rules, determine whether the fused defect analysis results meet the conditions for triggering process parameter adjustment. If so, determine the specific adjustment values ​​of temperature and pressure based on the defect type and severity, and generate process parameter adjustment suggestions; According to the preset equipment maintenance strategy rules, determine whether the fused defect analysis results meet the conditions for triggering equipment maintenance. If so, determine the equipment maintenance time and maintenance items based on the defect type and severity as well as the equipment's historical maintenance records, and generate an equipment maintenance strategy.

8. A modular defect detection system for IC substrates, characterized in that: include: An image preprocessing module, used to obtain original image data in the production process of the IC substrate, preprocess the original image data, and obtain preprocessed IC substrate image data; A recognition model building module is used to build a defect recognition model according to the IC substrate image data, introduce an attention mechanism into the defect recognition model, generate an attention weight map, and obtain an IC substrate recognition result; A data analysis module, used to obtain process parameter data and equipment operation status data in the production process of the IC substrate, associate and analyze the process parameter data, the equipment operation status data and historical defect data, mine the potential relationship between the data, and determine the key parameters that have a greater impact on defect formation; A prediction model building module, used to build a defect prediction model based on the process parameter data, the equipment operation status data and the key parameters, and determine whether the IC substrate has defects; The strategy generation module is used to fuse the results of the defect recognition model and the defect prediction model, and based on the fused results and in combination with pre-set rules, generate corresponding process parameter adjustment suggestions or equipment maintenance strategies.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the modular defect detection method for IC substrates according to any one of claims 1 to 7 are implemented.

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

Images