Silicon wafer drying improvement method and apparatus based on d-LKA network model

Abstract

A silicon wafer drying improvement method and apparatus based on a D-LKA network model. The method comprises: acquiring image data of a dried silicon wafer and drying parameter data; using a D-LKA network model to preprocess the image data so as to extract water marks and other non-compliance features on the surface of the silicon wafer; using the D-LKA network model to perform feature learning on the preprocessed image data; determining whether the features of the surface state of the silicon wafer satisfy a condition; and acquiring control parameters of a corresponding silicon wafer drying process on the basis of different features of the surface state of the silicon wafer, and when the silicon wafer needs to be re-dried, invoking corresponding drying parameters by means of the D-LKA network model on the basis of the features of the surface of the silicon wafer, optimizing the drying parameters and transmitting the optimized drying parameters to a circulating hot air system, and drying the silicon wafer again. The D-LKA network model is applied to the silicon wafer drying process, so that unmanned silicon wafer screening is achieved, the drying parameters are optimized, the labor and time costs are low, the silicon wafer production efficiency is high, and the quality of the silicon wafer is high.

Description

A silicon wafer drying improvement method and device based on D-LKA network model Technical Field

[0001] The present invention relates to the technical field of silicon wafer processing, and in particular to a method and device for improving silicon wafer drying based on a D-LKA network model. Background Art

[0002] Currently, during the cleaning process of silicon wafer manufacturing, the wafers are cleaned with an aqueous solution to remove surface debris. After cleaning, the wafers need to be dried to remove moisture from the surface. Existing drying technology uses a circulating hot air system to dry out the moisture from the wafer surface, thereby removing debris from the wafer surface.

[0003] When drying each batch of silicon wafers using a circulating hot air system, the surface moisture content of each wafer varies within each batch. This can result in some wafers remaining undried after drying, resulting in water marks on the wafer surfaces. In this case, each wafer must be manually inspected to identify those with water marks. Drying parameters are then readjusted and the wafers re-dried. This process must be repeated to ensure that all wafers are free of water marks. However, due to the human factor involved, this process is subject to significant uncertainty, making it impossible to guarantee that all wafers produced will be free of water marks. Furthermore, the process is time-consuming and labor-intensive. Furthermore, when re-drying the wafers, the drying parameters are set manually based on experience, making it difficult to establish a unified standard.

[0004] Summary of the Invention

[0005] In order to solve the problem of low recognition accuracy in the existing technology, the purpose of the present invention is to provide a silicon wafer drying improvement method and device based on the D-LKA network model. This method applies the D-LKA network model to the silicon wafer drying process, which not only realizes unmanned silicon wafer screening, but also optimizes the drying parameters, thereby saving a lot of manpower and time costs and improving the efficiency and quality of silicon wafer production.

[0006] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0007] A method for improving silicon wafer drying based on a D-LKA network model, the method comprising the following steps:

[0008] Obtain image data of the silicon wafer after drying, as well as drying parameter data corresponding to the circulating hot air system, as input for model training data;

[0009] Train the deep learning framework to obtain the D-LKA network model;

[0010] The acquired image data is used as data input and preprocessed and analyzed using the D-LKA network model to extract water marks and other unqualified features on the silicon wafer surface;

[0011] Use the D-LKA network model to perform feature learning on the preprocessed image data;

[0012] The image data after feature learning is used to determine whether the characteristics of the silicon wafer surface state meet the cleaning requirements;

[0013] According to the different characteristics of the silicon wafer surface state, the control parameters of the corresponding silicon wafer drying process are obtained. When the silicon wafer needs to be re-dried, the corresponding drying parameters are retrieved according to the characteristics of the silicon wafer surface through the D-LKA network model, the drying parameters are optimized, and the optimized drying parameters are sent to the circulating hot air system through the message queue. The circulating hot air system re-dries the silicon wafer according to the optimized drying parameters.

[0014] According to a silicon wafer drying improvement method based on a D-LKA network model provided by the present invention, the image data is an image of the silicon wafer after drying obtained by taking a photo with a high-precision camera, and the image is used to show the shape and surface features of the silicon wafer.

[0015] According to a silicon wafer drying improvement method based on a D-LKA network model provided by the present invention, when performing image data preprocessing, the pixel values ​​of the image are linearly scaled and mapped to a specified range of 0, 10, 1 or between -1, 1 and 1, 1.

[0016] According to a method for improving silicon wafer drying based on a D-LKA network model provided by the present invention, the drying parameters are various control parameters required in the silicon wafer drying process, including drying temperature, drying time, wind speed of the circulating hot air system, etc.; the D-LKA network model is used to learn the various control parameters required for drying silicon wafers with varying surface moisture.

[0017] According to a silicon wafer drying improvement method based on a D-LKA network model provided by the present invention, the training of the D-LKA network model includes:

[0018] Data acquisition: During the data acquisition phase, image data of silicon wafers with and without water marks are collected;

[0019] Label data: assign corresponding labels to each image sample;

[0020] Data partitioning: Divide the entire dataset into training set, validation set, and test set;

[0021] Data augmentation;

[0022] Build a D-LKA network model: Use a deep learning framework to build a deep learning model, where the input of the model is the silicon wafer image and the corresponding drying parameter data, and the output is the prediction result of whether the silicon wafer cleaning meets the standards.

[0023] According to a silicon wafer drying improvement method based on a D-LKA network model provided by the present invention, the construction of the D-LKA network model includes:

[0024] Image input: the image of the dried silicon wafer is input into the D-LKA network model;

[0025] After one convolutional layer;

[0026] After a GELU function;

[0027] After a large convolution kernel layer;

[0028] Merge the output of the GELU function with the output of the large convolution kernel layer and output them;

[0029] Model training: Use the training set to train the convolutional and attention mechanism hybrid neural network;

[0030] Model evaluation and testing: Use the test set to evaluate and test the trained model.

[0031] According to a silicon wafer drying improvement method based on a D-LKA network model provided by the present invention, the data enhancement operation includes:

[0032] Random rotation: images of the silicon wafer at different angles.

[0033] Random flipping: Different situations of the silicon wafer in horizontal or vertical directions.

[0034] Random scaling: Silicon wafer images of different sizes.

[0035] Brightness and contrast adjustment: Silicon wafer images under different lighting conditions.

[0036] According to a silicon wafer drying improvement method based on a D-LKA network model provided by the present invention, the D-LKA network model is used to perform feature learning on pre-processed image data, including:

[0037] Forward propagation: The pre-processed image data is input into the D-LKA network, and the output of each layer is calculated layer by layer through the forward propagation process of the network;

[0038] Feature map extraction: In the convolutional layers and attention mechanisms, the network learns to extract feature maps of the image data and represent them as a high-dimensional feature space;

[0039] Pooling operation: perform pooling operation;

[0040] Fully connected layer: The pooled feature map is input into the fully connected layer to learn higher-level semantic information;

[0041] Loss function calculation: Using label information, calculate the loss between the network output and the true label;

[0042] Backpropagation and optimization: Through the backpropagation algorithm, the contribution of each layer to the loss is calculated and the network parameters are updated to minimize the loss function.

[0043] Repeated training: The above steps will be repeated multiple times on the training dataset.

[0044] Feature learning results: The D-LKA network model learns features that can effectively judge the surface state of silicon wafers.

[0045] According to a silicon wafer drying improvement method based on a D-LKA network model provided by the present invention, the D-LKA model analyzes the characteristics of the silicon wafer that needs to be re-dried based on the feature learning results, that is, the changing trend of the drying parameters corresponding to the change from the presence of water marks on the silicon wafer surface to the absence of water marks on the silicon wafer surface, and then outputs the drying parameters of the silicon wafer with respect to the absence of water marks.

[0046] A silicon wafer drying improvement device based on a D-LKA network model, comprising:

[0047] The data acquisition module is used to obtain image data of the silicon wafer after drying, as well as the drying parameter data corresponding to the circulating hot air system, as input for model training data;

[0048] Model building module, used to train the deep learning framework to obtain the D-LKA network model;

[0049] The data processing module is used to take the acquired image data as data input and perform preprocessing and analysis using the D-LKA network model to extract water marks and other unqualified features on the silicon wafer surface;

[0050] Feature learning module, used to perform feature learning on preprocessed image data using the D-LKA network model;

[0051] A judgment module is used to judge whether the characteristics of the silicon wafer surface state meet the cleaning requirements based on the image data after feature learning;

[0052] The result output module is used to obtain the control parameters of the silicon wafer's corresponding drying process according to the different characteristics of the silicon wafer's surface state. When the silicon wafer needs to be re-dried, the D-LKA network model is used to retrieve the corresponding drying parameters based on the characteristics of the silicon wafer's surface, optimize the drying parameters, and send the optimized drying parameters to the circulating hot air system through a message queue. The circulating hot air system then refroze the silicon wafer according to the optimized drying parameters.

[0053] It can be seen that compared with the prior art, the present invention has the following beneficial effects:

[0054] 1. The present invention uses the D-LKA network model to automatically judge the dried silicon wafers, screen out silicon wafers with water marks, realize unmanned silicon wafer screening, and save a lot of manpower and time.

[0055] 2. The present invention optimizes the drying parameters through the D-LKA network model. By learning a large amount of data, the D-LKA model optimizes the drying parameters, so that the silicon wafers after drying can achieve higher quality.

[0056] 3. Under the wireless network, the present invention uses the information transmission method of the message queue to realize the mutual interaction between the D-LKA network model and the circulating hot air system, ensuring the automation of the parameter adjustment of the circulating hot air system while meeting the production rhythm.

[0057] The present invention further provides an electronic device, comprising:

[0058] a memory storing computer-executable instructions;

[0059] a processor configured to execute the computer-executable instructions,

[0060] When the computer executable instructions are executed by the processor, the steps of any one of the above-mentioned methods for improving silicon wafer drying based on the D-LKA network model are implemented.

[0061] The present invention also provides a storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program is used to implement the steps of any one of the above-mentioned methods for improving silicon wafer drying based on the D-LKA network model.

[0062] Thus, the present invention also provides an electronic device and storage medium for a silicon wafer drying improvement method based on a D-LKA network model, comprising: one or more memories and one or more processors. The memories are used to store program code and intermediate data generated during program execution, model output results, and the model and its parameters; the processors are used to manage the processor resources occupied by code execution and the multiple processor resources occupied during model training.

[0063] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] FIG1 is a flow chart of an embodiment of a method for improving silicon wafer drying based on a D-LKA network model according to the present invention.

[0065] FIG2 is a schematic diagram of an embodiment of a silicon wafer drying improvement device based on a D-LKA network model according to the present invention.

[0066] FIG3 is a schematic diagram of drying parameters corresponding to the change from “with water marks” to “without water marks” in an embodiment of a method for improving silicon wafer drying based on a D-LKA network model of the present invention. DETAILED DESCRIPTION

[0067] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0068] References to "embodiments" herein mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0069] 1 , the present invention provides a method for improving silicon wafer drying based on a D-LKA network model, the method comprising the following steps:

[0070] Step S1, obtaining image data of the silicon wafer after drying and various drying parameter data corresponding to the circulating hot air system as input for model training data;

[0071] Step S2, training the deep learning framework to obtain a D-LKA network model;

[0072] In step S3, the acquired image data is used as data input and preprocessed and analyzed using the D-LKA network model (which has both SA attention and large convolution kernel capabilities) to extract water marks and other unqualified features on the silicon wafer surface; among them, the preprocessing includes image processing methods such as denoising and contrast enhancement to optimize the performance of the D-LKA method.

[0073] In step S4, the D-LKA network model is used to perform feature learning on the preprocessed image data. In this embodiment, D-LKA utilizes a combination of large-kernel convolutions to adaptively adjust the shape and position of the convolution kernels to better capture the unstructured features of the silicon wafer surface. Based on the D-LKA results, the attention mechanism assigns different importance to different parts of the image, effectively focusing on areas with potential problems.

[0074] Step S5: judging whether the characteristics of the surface state of the silicon wafer meet the cleaning requirements through the image data after feature learning.

[0075] In step S6, based on the D-LKA results and the different characteristics of the silicon wafer's surface condition, control parameters for the wafer's drying process, such as heating temperature and time, are obtained. When the wafer needs to be re-dried, the D-LKA network model retrieves the corresponding drying parameters based on the wafer's surface characteristics and optimizes them based on previous feature learning. These optimized drying parameters are then transmitted to the circulating hot air system via a message queue over the wireless communication network. The circulating hot air system then re-dries the wafer according to the optimized parameters, achieving higher wafer quality and a higher yield rate.

[0076] As shown in Figure 3, the D-LKA model analyzes the characteristics of the silicon wafer that needs to be re-dried based on previous learning, analyzes how the drying parameters change from "with water marks" to "without water marks," and then outputs the drying parameters for the silicon wafer to be "without water marks." For example, when the silicon wafer is characterized as "with water marks," the corresponding drying parameters are a drying temperature of 120°C, a drying time of 200 seconds, and a drying speed of level 3. When the silicon wafer is characterized as "without water marks," the corresponding drying parameters are a drying temperature of 115°C, a drying time of 180 seconds, and a drying speed of level 2.

[0077] The D-LKA model learns the drying parameters corresponding to the "water mark-free" silicon wafer. When the silicon wafer needs to be re-dried, the D-LKA model will output the new drying parameters for the silicon wafer based on the learning of the drying parameters of the "water mark-free" silicon wafer, which are the optimized drying parameters.

[0078] Among them, the wireless communication network can be an operator network or a wireless local area network based on the WiFi standard. For example, 5G technology can be combined with the information transmission method of the message queue to realize the interaction between the D-LKA network model and the circulating hot air system, truly realizing unmanned operations from silicon wafer screening to parameter setting and re-drying, thereby saving a lot of manpower and time costs and improving the efficiency and quality of silicon wafer production.

[0079] Through the above steps, we can see that, using the D-LKA network model combined with message queue transmission on a 5G network, we can improve the drying process through image processing and pattern recognition technology. This method not only identifies and processes defective silicon wafers with water marks, but also optimizes drying parameters, improving drying efficiency and the yield rate of finished products.

[0080] In step S1, the image data is an image of the dried silicon wafer, captured by a high-precision camera. This image is used to display the shape and surface features of the silicon wafer. This image data is input into the D-LKA network model for automatic analysis. The image can reveal surface features of the silicon wafer, such as water marks or other information such as shape and color.

[0081] When preprocessing image data, the pixel values ​​of the image are linearly scaled and mapped to the specified range 0, 10, 1 or -1, 1-1, 1.

[0082] Specifically, a common image preprocessing method is image normalization, which maps the pixel values ​​of an image to a specific range, usually 0, 10, 1 or -1, 1-1, 1, through linear scaling. This helps eliminate brightness and contrast differences between different images, making it easier for the model to learn common features of the image.

[0083] In this embodiment, the steps of image normalization are as follows:

[0084] 1. Obtain image data: Collect image data after the silicon wafer is dried.

[0085] 2. Pixel value scaling: linearly scale the pixel values ​​of the image so that their range is between 0, 10, 1 or -1, 1-1, 1.

[0086] As you can see, after image normalization preprocessing, the pixel values ​​of the image will be within the specified range (0, 10, 1 or -1, 1-1, 1), which helps speed up the model training process because neural networks can more easily process input data with a consistent range. In addition, image normalization also helps improve the model's robustness to image features and reduce the impact of changes in brightness and contrast.

[0087] In step S1 above, the drying parameters are the various control parameters required for the silicon wafer drying process, including drying temperature, drying time, and the wind speed of the circulating hot air system. These parameters are the control parameters required for the silicon wafer drying process, and the D-LKA network model is used to learn the various control parameters required for drying silicon wafers with varying surface moisture.

[0088] In the above step S2, the training of the D-LKA network model includes:

[0089] Data acquisition: During the data acquisition phase, image data of silicon wafers with and without water marks are collected.

[0090] Label data: Assign a corresponding label to each image sample.

[0091] Data partitioning: Divide the entire dataset into training set, validation set, and test set.

[0092] Data augmentation.

[0093] Build a D-LKA network model: Use a deep learning framework to build a deep learning model, where the input of the model is the silicon wafer image and the corresponding drying parameter data, and the output is the prediction result of whether the silicon wafer cleaning meets the standards.

[0094] Specifically, the detailed process of the D-LKA network model is as follows:

[0095] Data Collection: Collect silicon wafer image samples. These images should include the wafer's shape and surface features for training the model for classification and detection. This ensures that the dataset covers various conditions encountered in a production environment, such as varying lighting conditions, angles, and wafer shapes. This ensures that the dataset is diverse and representative, improving the model's generalization capabilities.

[0096] Labeled data: Provides accurate label information for model training. For each wafer image, key features are annotated, and each image sample is assigned a corresponding label. In this case, the label can be binary, indicating whether the wafer has water marks. The accuracy of the labeled data is crucial for model training.

[0097] Data partitioning: Divide the entire dataset into training, validation, and test sets. Typically, most of the data is used for training, a small portion is used to validate the model’s performance, and the test set is used to ultimately evaluate the model’s generalization performance.

[0098] Data augmentation: Because the surface condition of silicon wafers can be affected by a variety of factors, data augmentation can be performed to increase the diversity and reliability of training samples. Using data augmentation techniques can increase data diversity and improve model robustness. For example, the dataset can be expanded by performing operations such as rotation, flipping, scaling, and translation on images. This helps improve the robustness and generalization of the model.

[0099] Build a D-LKA network model: Use a deep learning framework (Python) to build a deep learning model. The model inputs are silicon wafer images and corresponding drying parameter data, and the output is a prediction result on whether the silicon wafer has met the cleaning standards.

[0100] In this embodiment, the construction of the D-LKA network model includes:

[0101] Image input: the image of the dried silicon wafer is input into the D-LKA network model;

[0102] After one convolution layer (9x9 convolution kernel size, stride 1);

[0103] After a GELU function;

[0104] After a large convolution kernel layer (21x21 convolution kernel size, stride 2);

[0105] The output of the GELU function is combined with the output of the large convolution kernel layer (concatenated) and output;

[0106] Model training: Use the training set to train the convolutional and attention mechanism hybrid neural network;

[0107] Model evaluation and testing: Use the test set to evaluate and test the trained model.

[0108] Specifically, the data enhancement operations in this embodiment include:

[0109] Random rotation: images of the silicon wafer at different angles.

[0110] Random flipping: Different situations of the silicon wafer in horizontal or vertical directions.

[0111] Random scaling: Silicon wafer images of different sizes.

[0112] Brightness and contrast adjustment: Silicon wafer images under different lighting conditions.

[0113] In the above step S4, the D-LKA network model is used to perform feature learning on the pre-processed image data, including:

[0114] Input image: Obtain image data of the silicon wafer after drying and undergo necessary preprocessing.

[0115] Build a D-LKA network: Design and build a D-LKA neural network model, which typically includes convolutional layers, attention mechanisms (such as large convolution kernel attention in D-LKA), pooling layers, fully connected layers, etc. D-LKA is designed to fully understand contextual information within the receptive field while adapting to different data modes.

[0116] Forward propagation: The preprocessed image data is fed into the D-LKA network. Through the network's forward propagation process, the output of each layer is calculated layer by layer. The attention mechanism (D-LKA) helps the network focus on important areas in the image and learn more discriminative features.

[0117] Feature map extraction: In the convolutional layers and attention mechanisms, the network learns to extract feature maps from the image data. These feature maps capture information such as texture, shape, and structure in the image and represent it as a high-dimensional feature space.

[0118] Pooling operation: After some layers, pooling operations may be performed to reduce the spatial resolution of the feature map, reduce the amount of computation, and extract higher-level abstract features.

[0119] Fully connected layers: The pooled feature maps are fed into fully connected layers to learn higher-level semantic information. These layers help map the extracted features to the final output.

[0120] Loss function calculation: Using label information, calculate the loss between the network output and the true label. Common loss functions include cross entropy loss for classification problems.

[0121] Backpropagation and optimization: Through the backpropagation algorithm, the contribution of each layer to the loss is calculated and the network parameters are updated to minimize the loss function. Optimization algorithms such as stochastic gradient descent (SGD) or its variants are used to adjust the network weights.

[0122] Repeated training: The above steps will be iterated multiple times on the training dataset to continuously optimize the network weights and improve model performance.

[0123] Feature Learning Results: Ultimately, the D-LKA network model learned features that effectively determined the surface condition of silicon wafers. These features were learned using weights and parameters within the network hierarchy, enabling the model to automatically determine the surface condition of silicon wafers after drying.

[0124] In this embodiment, the D-LKA model analyzes the characteristics of silicon wafers that need re-drying based on feature learning results, specifically the trend in drying parameters from when water marks appear on the wafer surface to when the wafer surface is free of water marks. The model then outputs the drying parameters for the wafer to achieve a water-free state. For example, the drying temperature is 80°C and the drying time is 1 hour.

[0125] In the above step S5, whether the cleaning requirement is met is determined based on the label "with water marks" or "without water marks".

[0126] An embodiment of a silicon wafer drying improvement device based on a D-LKA network model:

[0127] As shown in Figure 2, a silicon wafer drying improvement device based on the D-LKA network model includes:

[0128] Data acquisition module 1 is used to obtain image data of the silicon wafer after drying, as well as various drying parameter data corresponding to the circulating hot air system, as input for model training data;

[0129] Model building module 2 is used to train the deep learning framework to obtain the D-LKA network model;

[0130] Data processing module 3, used to take the acquired image data as data input, and perform preprocessing and analysis using the D-LKA network model to extract water marks and other unqualified features on the silicon wafer surface;

[0131] Feature learning module 4, used for performing feature learning on the preprocessed image data using the D-LKA network model;

[0132] A judgment module 5 is used to judge whether the characteristics of the surface state of the silicon wafer meet the cleaning requirements based on the image data after feature learning;

[0133] The result output module 6 is used to obtain the control parameters of the silicon wafer corresponding to the drying process according to the different characteristics of the silicon wafer surface state. When the silicon wafer needs to be re-dried, the corresponding drying parameters are retrieved according to the characteristics of the silicon wafer surface through the D-LKA network model, the drying parameters are optimized, and the optimized drying parameters are sent to the circulating hot air system through the message queue. The circulating hot air system re-dries the silicon wafer according to the optimized drying parameters.

[0134] The circulating hot air system includes a fan and a heater arranged at the air outlet of the fan. The fan is also connected to a wind speed control device, and the heater is also connected to a heating temperature control device. The wind speed control device includes a frequency converter and an analog generator for setting the fan speed; the heating temperature control device includes a temperature controller and a thermal relay; the control mode of the temperature controller and the thermal relay is PID mode; a time relay is provided on the fan, and the time relay uses a delayed closing mode to control the operation of the fan.

[0135] In data acquisition module 1, image data is obtained by a high-precision camera from a dried silicon wafer. This image is used to display the shape and surface features of the silicon wafer. This image data is input into the D-LKA network model for automatic judgment. The image can reveal surface features of the silicon wafer, such as water marks or other information such as shape and color.

[0136] When preprocessing image data, the pixel values ​​of the image are linearly scaled and mapped to the specified range 0, 10, 1 or -1, 1-1, 1.

[0137] In data acquisition module 1, drying parameters are the various control parameters required for the silicon wafer drying process, including drying temperature, drying time, and the wind speed of the circulating hot air system. These parameters are the control parameters required for the silicon wafer drying process. The D-LKA network model learns the control parameters required for drying silicon wafers with varying surface moisture content.

[0138] In model building module 2, the training of the D-LKA network model includes:

[0139] Data acquisition: During the data acquisition phase, image data of silicon wafers with and without water marks are collected.

[0140] Label data: Assign a corresponding label to each image sample.

[0141] Data partitioning: Divide the entire dataset into training set, validation set, and test set.

[0142] Data augmentation.

[0143] Build a D-LKA network model: Use a deep learning framework to build a deep learning model, where the input of the model is the silicon wafer image and the corresponding drying parameter data, and the output is the prediction result of whether the silicon wafer cleaning meets the standards.

[0144] Specifically, the construction of the D-LKA network model includes:

[0145] Image input: the image of the dried silicon wafer is input into the D-LKA network model;

[0146] After one convolution layer (9x9 convolution kernel size, stride 1);

[0147] After a GELU function;

[0148] After a large convolution kernel layer (21x21 convolution kernel size, stride 2);

[0149] The output of the GELU function is combined with the output of the large convolution kernel layer (concatenated) and output;

[0150] Model training: Use the training set to train the convolutional and attention mechanism hybrid neural network;

[0151] Model evaluation and testing: Use the test set to evaluate and test the trained model.

[0152] In feature learning module 4, the D-LKA network model is used to perform feature learning on the preprocessed image data, including:

[0153] Input image: Obtain image data of the silicon wafer after drying and undergo necessary preprocessing.

[0154] Build a D-LKA network: Design and build a D-LKA neural network model, which typically includes convolutional layers, attention mechanisms (such as large convolution kernel attention in D-LKA), pooling layers, fully connected layers, etc. D-LKA is designed to fully understand contextual information within the receptive field while adapting to different data modes.

[0155] Forward propagation: The preprocessed image data is fed into the D-LKA network. Through the network's forward propagation process, the output of each layer is calculated layer by layer. The attention mechanism (D-LKA) helps the network focus on important areas in the image and learn more discriminative features.

[0156] Feature map extraction: In the convolutional layers and attention mechanisms, the network learns to extract feature maps from the image data. These feature maps capture information such as texture, shape, and structure in the image and represent it as a high-dimensional feature space.

[0157] Pooling operation: After some layers, pooling operations may be performed to reduce the spatial resolution of the feature map, reduce the amount of computation, and extract higher-level abstract features.

[0158] Fully connected layers: The pooled feature maps are fed into fully connected layers to learn higher-level semantic information. These layers help map the extracted features to the final output.

[0159] Loss function calculation: Using label information, calculate the loss between the network output and the true label. Common loss functions include cross entropy loss for classification problems.

[0160] Backpropagation and optimization: Through the backpropagation algorithm, the contribution of each layer to the loss is calculated and the network parameters are updated to minimize the loss function. Optimization algorithms such as stochastic gradient descent (SGD) or its variants are used to adjust the network weights.

[0161] Repeated training: The above steps will be iterated multiple times on the training dataset to continuously optimize the network weights and improve model performance.

[0162] Feature Learning Results: Ultimately, the D-LKA network model learned features that effectively determined the surface condition of silicon wafers. These features were learned using weights and parameters within the network hierarchy, enabling the model to automatically determine the surface condition of silicon wafers after drying.

[0163] In this embodiment, the D-LKA model analyzes the characteristics of silicon wafers that need re-drying based on feature learning results, specifically the trend in drying parameters from when water marks appear on the wafer surface to when the wafer surface is free of water marks. The model then outputs the drying parameters for the wafer to achieve a water-free state. For example, the drying temperature is 80°C and the drying time is 1 hour.

[0164] In the judgment module 5, whether the cleaning standard is met is judged based on the label "with water marks" or "without water marks".

[0165] In summary, this embodiment uses the D-LKA network model to automatically judge the silicon wafers after drying, screen out silicon wafers with water marks, realize unmanned silicon wafer screening, and save a lot of manpower and time; this embodiment uses the D-LKA network model to optimize the drying parameters. By learning a large amount of data, the D-LKA model optimizes the drying parameters, so that the dried silicon wafers can achieve higher quality; under the wireless network, this embodiment uses the information transmission method of the message queue to realize the mutual interaction between the D-LKA network model and the circulating hot air system, ensuring that the parameter adjustment of the circulating hot air system is automated while meeting the production rhythm.

[0166] In one embodiment, an electronic device is provided, which may be a server. The computer device includes a processor, a memory, and a network interface connected via a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic 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 computer program in the non-volatile storage medium. The database of the electronic device is used to store data. The network interface of the electronic device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for improving silicon wafer drying based on a D-LKA network model is implemented.

[0167] Those skilled in the art will understand that the electronic device structure shown in this embodiment is only a partial structure related to the scheme of the present application, and does not constitute a limitation on the electronic device to which the scheme of the present application is applied. The specific electronic device may include more or fewer components than those shown in this embodiment, or combine certain components, or have a different component arrangement.

[0168] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0169] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented 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 used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0170] In addition, the logical instructions in the above-mentioned memory can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0171] Thus, this embodiment also provides an electronic device and storage medium for a silicon wafer drying improvement method based on a D-LKA network model, comprising: one or more memories and one or more processors. The memories are used to store program code and intermediate data generated during program execution, model output results, and the model and its parameters; the processors are used to manage the processor resources occupied by code execution and the multiple processor resources occupied by model training.

[0172] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0173] The above embodiments are only preferred embodiments of the present invention and cannot be used to limit the scope of protection of the present invention. Any non-substantial changes and replacements made by technicians in this field on the basis of the present invention fall within the scope of protection required by the present invention.

Claims

1. A method for improving silicon wafer drying based on a D-LKA network model, characterized in that: The method comprises the following steps: Obtain image data of the silicon wafer after drying, as well as drying parameter data corresponding to the circulating hot air system, as input for model training data; Train the deep learning framework to obtain the D-LKA network model; The acquired image data is used as data input and preprocessed and analyzed using the D-LKA network model to extract water marks and other unqualified features on the silicon wafer surface; Use the D-LKA network model to perform feature learning on the preprocessed image data; The image data after feature learning is used to determine whether the characteristics of the silicon wafer surface state meet the cleaning requirements; According to the different characteristics of the silicon wafer surface state, the control parameters of the corresponding silicon wafer drying process are obtained. When the silicon wafer needs to be re-dried, the corresponding drying parameters are retrieved according to the characteristics of the silicon wafer surface through the D-LKA network model, the drying parameters are optimized, and the optimized drying parameters are sent to the circulating hot air system through the message queue. The circulating hot air system re-dries the silicon wafer according to the optimized drying parameters.

2. The method according to claim 1, wherein: The image data is an image of the dried silicon wafer obtained by taking a photo with a high-precision camera, and the image is used to show the shape and surface features of the silicon wafer.

3. The method according to claim 1, wherein: When preprocessing image data, the pixel values ​​of the image are linearly scaled and mapped to the specified range 0, 10, 1 or -1, 1-1, 1.

4. The method according to claim 1, wherein: The drying parameters are various control parameters required in the silicon wafer drying process, including drying temperature, drying time, wind speed of the circulating hot air system, etc. The D-LKA network model is used to learn the various control parameters required for drying silicon wafers with different surface moisture.

5. The method according to claim 1, wherein: The training of the D-LKA network model includes: Data acquisition: During the data acquisition phase, image data of silicon wafers with and without water marks are collected; Label data: assign corresponding labels to each image sample; Data partitioning: Divide the entire dataset into training set, validation set, and test set; Data augmentation; Build a D-LKA network model: Use a deep learning framework to build a deep learning model, where the input of the model is the silicon wafer image and the corresponding drying parameter data, and the output is the prediction result of whether the silicon wafer cleaning meets the standards.

6. The method according to claim 5, characterized in that: The construction of the D-LKA network model includes: Image input: the image of the dried silicon wafer is input into the D-LKA network model; After one convolutional layer; After one GELU function; After a large convolution kernel layer; Merge the output of the GELU function with the output of the large convolution kernel layer and output them; Model training: Use the training set to train the convolutional and attention mechanism hybrid neural network; Model evaluation and testing: Use the test set to evaluate and test the trained model.

7. The method according to claim 5, characterized in that: The data enhancement operations include: Random rotation: images of the silicon wafer at different angles. Random flipping: Different situations of the silicon wafer in horizontal or vertical directions. Random scaling: Silicon wafer images of different sizes. Brightness and contrast adjustment: Silicon wafer images under different lighting conditions.

8. The method according to claim 1, wherein: The method of using the D-LKA network model to perform feature learning on the pre-processed image data includes: Forward propagation: The pre-processed image data is input into the D-LKA network, and the output of each layer is calculated layer by layer through the forward propagation process of the network; Feature map extraction: In the convolutional layers and attention mechanisms, the network learns to extract feature maps of the image data and represent them as a high-dimensional feature space; Pooling operation: perform pooling operation; Fully connected layer: The pooled feature map is input into the fully connected layer to learn higher-level semantic information; Loss function calculation: Using label information, calculate the loss between the network output and the true label; Backpropagation and optimization: Through the backpropagation algorithm, the contribution of each layer to the loss is calculated and the network parameters are updated to minimize the loss function. Repeated training: The above steps will be repeated multiple times on the training dataset. Feature learning results: The D-LKA network model learns features that can effectively judge the surface state of silicon wafers.

9. The method according to claim 1, wherein: The D-LKA model analyzes the characteristics of silicon wafers that need to be re-dried based on the feature learning results, that is, the changing trend of the drying parameters corresponding to the change from water marks on the silicon wafer surface to the absence of water marks on the silicon wafer surface, and then outputs the drying parameters for the silicon wafer with no water marks.

10. A silicon wafer drying improvement device based on the D-LKA network model, characterized in that: include: The data acquisition module is used to obtain image data of the silicon wafer after drying, as well as the drying parameter data corresponding to the circulating hot air system, as input for model training data; Model building module, used to train the deep learning framework to obtain the D-LKA network model; The data processing module is used to take the acquired image data as data input and perform preprocessing and analysis using the D-LKA network model to extract water marks and other unqualified features on the silicon wafer surface; Feature learning module, used to perform feature learning on preprocessed image data using the D-LKA network model; The judgment module is used to judge the characteristics of the silicon wafer surface state through the image data after feature learning. Whether it meets the cleaning requirements; The result output module is used to obtain the control parameters of the silicon wafer's corresponding drying process according to the different characteristics of the silicon wafer's surface state. When the silicon wafer needs to be re-dried, the D-LKA network model is used to retrieve the corresponding drying parameters based on the characteristics of the silicon wafer's surface, optimize the drying parameters, and send the optimized drying parameters to the circulating hot air system through a message queue. The circulating hot air system then refroze the silicon wafer according to the optimized drying parameters.

Images