Wafer processing process sensor data anomaly detection method and related device

Abstract

The application provides a wafer processing process sensor data anomaly detection method and related equipment, the method comprising: obtaining sensor data in a wafer processing process, and performing data preprocessing on the sensor data; inputting the data preprocessed sensor data into a pre-trained data anomaly detection model to obtain reconstructed data of the sensor data output by the data anomaly detection model; calculating the error between the sensor data and the reconstructed data to determine the anomaly degree, determining the abnormal data points according to the anomaly degree, and marking and displaying the abnormal data points. The application is modeled by deep learning, and the reconstructed data is generated by a deep learning model. The error between the real-time sensor data and the reconstructed data output by the model is calculated to intelligently determine the abnormal data points. The false positives or false negatives caused by the limitations of fixed threshold setting are avoided, the automation and intelligent level of anomaly detection are improved, and the method is more accurate and reliable compared with the traditional threshold setting method.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of semiconductor technology, and in particular to a wafer processing process sensor data anomaly detection method and related equipment. BACKGROUND

[0002] In the semiconductor manufacturing process, the sensors of the wafer processing machine will collect a large amount of real-time data, which is used to monitor various process parameters in the processing process. The existing fault detection and classification (FDC, Fault Detection and Classification) system usually controls the process by setting the control threshold (SPEC value) of the process parameters. For example, the maximum upper limit value of the temperature parameter is usually set, and when the real-time temperature data exceeds the threshold value, the system will trigger an alarm and notify the relevant engineers to handle it. For different machines (EQP), chambers (Chamber), recipes (Recipe) and parameters (Parameter), engineers need to set and adjust different SPEC values according to each specific situation. This process is not only tedious and time-consuming, but also puts a heavy burden on engineers. In addition, the SPEC value used by the existing system is usually a fixed value, which cannot cover all possible abnormal conditions. Therefore, when an abnormality occurs that is not set within the SPEC range, the existing system may not be able to identify it in time, or some normal conditions may be misjudged as abnormal, affecting the stability and efficiency of production. SUMMARY

[0003] In view of the deficiencies in the prior art, the present application provides a wafer processing process sensor data anomaly detection method and related equipment to at least solve the problem that the sensor data anomaly detection in the prior art relies on fixed threshold setting, resulting in inaccurate anomaly identification.

[0004] In order to achieve the above-mentioned purposes and other advantages, the present application adopts the following technical solutions:

[0005] In a first aspect, the present application provides a wafer processing process sensor data anomaly detection method, comprising:

[0006] Obtaining sensor data in a wafer processing process and performing data preprocessing on the sensor data;

[0007] Inputting the data preprocessed sensor data into a pre-trained data anomaly detection model to obtain reconstructed data of the sensor data output by the data anomaly detection model;

[0008] Calculating the error between the sensor data and the reconstructed data to determine the abnormality degree, determining the abnormal data points according to the abnormality degree, and marking and displaying the abnormal data points.

[0009] According to the wafer processing process sensor data anomaly detection method provided by the application, the step of obtaining sensor data in the wafer processing process and pre-processing the sensor data includes:

[0010] The sensor data is obtained in batch or streaming mode through the API interface of the FDC system, and the sensor data type is time series data.

[0011] The sensor data is cleaned, and the data cleaning includes at least one of missing value processing, denoising processing or data alignment.

[0012] The sensor data after data cleaning is subjected to minimum-maximum normalization processing to normalize the sensor data to a predetermined range.

[0013] According to the wafer processing process sensor data anomaly detection method provided by the application, the data anomaly detection model includes an encoder, a latent feature layer and a decoder.

[0014] The encoder is used to extract the feature representation of the sensor data and input to the latent feature layer.

[0015] The latent feature layer is used to linearly transform the extracted feature representation.

[0016] The decoder is used to reconstruct the sensor data according to the output of the latent feature layer.

[0017] According to the wafer processing process sensor data anomaly detection method provided by the application, the encoder includes a first fully connected layer and a first multi-head self-attention mechanism layer.

[0018] The first fully connected layer is used to convert the original input of the sensor data to more easily processed feature data through linear transformation.

[0019] The first multi-head self-attention mechanism layer is used to calculate the feature correlation of the feature data, capture the global dependency relationship between data points at different time steps, and calculate the information interaction between each time step based on the attention weight assignment, to generate a first context representation.

[0020] According to the wafer processing process sensor data anomaly detection method provided by the application, the decoder includes a second fully connected layer and a second multi-head self-attention mechanism layer.

[0021] The second multi-head self-attention mechanism layer is used to reconstruct the global dependency of the sensor data of the original input according to the output of the latent feature layer, calculate the attention weight between data points at different time steps, strengthen the feature relationship, and generate a second context representation for data reconstruction.

[0022] The second fully connected layer is used to receive the second context representation and map to the space of the original input data through linear transformation and nonlinear activation function, to generate reconstructed data.

[0023] According to the wafer processing process sensor data anomaly detection method provided by the application, the data anomaly detection model is trained based on an unsupervised deep learning method, and the training process includes:

[0024] Encoder training:

[0025] The first fully connected layer is used to perform linear transformation on the input sensor data, so that the data is converted into more easily processed feature data.

[0026] The first multi-head self-attention mechanism layer is used to calculate the attention score between the feature data and assign attention weights to identify the global dependency between data points at different time steps, and generate a first context representation.

[0027] The first context representation is mapped to a latent feature space through a feature conversion operation, to form a feature representation for representing the original input data.

[0028] Decoder training:

[0029] The second multi-head self-attention mechanism layer is used to reconstruct information of the feature representation, calculate the attention score between the feature representations, and assign attention weights to re-identify the global dependency between data points at different time steps, and generate a second context information for data reconstruction.

[0030] The second fully connected layer is used to map the decoded second context information, to convert the data back to the same dimension as the original input data, thereby completing the data reconstruction process.

[0031] Model optimization:

[0032] The loss value is generated by calculating the error between the original input data and the reconstructed data through the loss function, the parameters of the data anomaly detection model are updated by using the back propagation algorithm, and the training process is continuously iteratively optimized until the error converges or reaches the set training condition until convergence, to obtain the trained data anomaly detection model.

[0033] According to the wafer processing process sensor data anomaly detection method provided in the application, the first multi-head self-attention mechanism layer and the second multi-head self-attention mechanism layer identify the global dependency relationship between data points at different time steps, including:

[0034] The query vector, the key vector and the value vector are constructed, wherein the query vector represents the association degree between different data points, the key vector represents the data feature information of the data points, the association feature between the data points is matched with the query vector, and the value vector represents the data content information of the data points;

[0035] The dot product of the query vector and the key vector is calculated, and the attention score between the data points is obtained through scaling processing;

[0036] The attention score is converted into a probability distribution through a normalization function to serve as the attention weight of each data point;

[0037] The value vector is weighted and summed through the attention weight to obtain the context representation of each data point.

[0038] In a second aspect, the application provides an electronic device, which comprises:

[0039] One or more processors; and a memory storing computer program instructions which, when executed, cause the processor to perform the wafer processing process sensor data anomaly detection method as described above.

[0040] In a third aspect, the application provides a computer-readable storage medium, which stores computer programs / instructions, and the computer programs / instructions are executed by a processor to implement the wafer processing process sensor data anomaly detection method as described above.

[0041] In a fourth aspect, the application provides a computer program product, which comprises computer programs / instructions, and the computer programs / instructions are executed by a processor to implement the wafer processing process sensor data anomaly detection method as described above.

[0042] The wafer processing process sensor data anomaly detection method and related equipment provided by the application, by acquiring sensor data in the wafer processing process, and performing data preprocessing on the sensor data; the sensor data after data preprocessing is input into the pre-trained data anomaly detection model, and the reconstruction data of the sensor data output by the data anomaly detection model is obtained; the error between the sensor data and the reconstruction data is calculated to determine the anomaly degree, the abnormal data points are determined according to the anomaly degree, and the abnormal data points are marked and displayed. The application is modeled by deep learning, and the reconstruction data is generated by the deep learning model. The error between the real-time sensor data and the reconstruction data output by the model is calculated to intelligently determine the abnormal data points. Avoid false positives or false negatives caused by the limitations of fixed threshold setting, improve the automation and intelligence level of anomaly detection, and be more accurate and reliable than the traditional threshold setting method. In this way, the artificial operation workload can be reduced, the product yield can be improved, and economic benefits can be created. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other embodiments can be obtained by those skilled in the art without creative labor.

[0044] Figure 1 is a flowchart of the wafer processing process sensor data anomaly detection method provided by the embodiments of the application;

[0045] Figure 2 is a deep learning neural network structure diagram of the data anomaly detection model provided by the embodiments of the application;

[0046] Figure 3 is an example diagram of the parameter level sensor data provided by the embodiments of the application;

[0047] Figure 4 is a sensor data anomaly monitoring interface of the FDC system provided by the embodiments of the application;

[0048] Figure 5 is a structural schematic diagram of the electronic device provided by the embodiments of the application. DETAILED DESCRIPTION

[0049] The above description is only a summary of the technical solutions of the application, in order to more clearly understand the technical means of the application, the following preferred embodiments are described in detail according to the content of the specification, and in order to make the above and other purposes, characteristics and advantages of the application more obvious and easy to understand, the following preferred embodiments are described in detail as follows.

[0050] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the application, as claimed. It should be noted that the embodiments described herein are intended to be combined with other embodiments unless otherwise explicitly stated. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The singular terms "a," "an," and "the" include plural referents unless context clearly indicates otherwise. The terms "comprises," "comprising," "includes," "including," and "has," "having" as they refer to a described structure, do not preclude the presence or addition of one or more other structures or elements.

[0051] A fault detection and classification (FDC) system is an important system for real-time monitoring, detecting, and classifying process faults in semiconductor manufacturing processes. The FDC system collects process parameters, equipment states, and environmental data in real time through various sensors installed on production equipment, thereby continuously monitoring the production process. The system automatically determines whether the collected data exceeds the normal range by using preset thresholds and statistical models, to detect potential equipment faults or process abnormalities. Once an abnormality is detected, the FDC system not only alarms (setting a threshold for a parameter, if the threshold is exceeded, a warning action is triggered to notify the engineer to handle), but also classifies and analyzes the characteristics of the abnormal data to help engineers determine the nature and possible causes of the fault, thereby providing a basis for subsequent fault handling and process optimization.

[0052] Referring to Figure 1 The embodiments of the present application provide a wafer processing process sensor data anomaly detection method, comprising:

[0053] Step S1: Obtain sensor data in a wafer processing process, and perform data preprocessing on the sensor data.

[0054] In this embodiment, step S1 specifically comprises:

[0055] Step S101: Obtain sensor data in a batch or streaming manner through an API interface of an FDC system, and the sensor data is time series data;

[0056] Step S102: Perform data cleaning on the sensor data, and the data cleaning comprises at least one of missing value processing, denoising processing, or data alignment;

[0057] Step S103: Perform minimum-maximum normalization processing on the sensor data after data cleaning, to normalize the sensor data to a predetermined range.

[0058] The present application develops a data anomaly detection method based on an FDC system, which is a process monitoring platform for semiconductor manufacturing processes. It has integrated data management and acquisition functions and provides standardized data storage and calling mechanisms. The present application uses the existing data architecture and interface of the FDC system to achieve efficient and standardized data acquisition without additional data acquisition equipment, providing data support for subsequent anomaly detection.

[0059] Specifically, the sensor data in the FDC system is stored in a structured manner, which can cover multiple dimensions of process parameters involved in wafer processing, including: equipment state (EQP), chamber parameters (Chamber), process recipe (Recipe), and batch information (Batch ID), etc. Through the database interface or API interface provided by the FDC system, time series sensor data is obtained in batch or streaming mode, and the data obtained has timestamp information to provide complete time series data input. The sensor data type is time series data, which usually includes: equipment state information (machine ID, chamber ID), key process parameters (temperature, pressure, gas flow, voltage, current, etc.), production batch information (Batch ID, Wafer ID), etc.

[0060] The batch acquisition mode is suitable for offline analysis, and historical data within a certain period of time is obtained for model training or periodic analysis. The streaming acquisition mode is suitable for real-time monitoring, and the FDC system continuously receives sensor data and transmits it to the data anomaly detection model for real-time analysis.

[0061] The obtained sensor data is cleaned, mainly including: missing value processing, noise removal processing and data alignment, etc. For missing value processing, the sensor may lose signal, and the data may contain null or invalid data points. By interpolation filling to complete the data or deleting records with too many missing values, the data quality is ensured. For the noise of sensor data caused by factors such as equipment jitter and electromagnetic interference, the noise is removed by methods such as moving average or low-pass filtering to improve the stability of the data. For different sampling frequencies of different sensors, the data is aligned based on the timestamp to ensure that the time steps of different sensors are consistent, avoiding data alignment problems in subsequent model calculation. The cleaned data is more complete, accurate and smooth, providing high-quality data input for normalization processing.

[0062] The normalized data after data cleaning is to scale it to a specific range, usually between 0 and 1, or between -1 and 1. The purpose of this is to eliminate the influence of the dimensions of different features, so that different features have the same scale, thereby improving the convergence speed and detection accuracy of the model. The present method adopts the minimum-maximum normalization method for processing:

[0063]

[0064] wherein, is the normalized data, is the original input data, is the minimum value of the data, is the maximum value of the data.

[0065] Data normalization can reduce the influence between different parameter magnitudes, avoid certain feature values being too large and dominating model calculation. It allows the model to learn the characteristics of each process parameter more fairly. It improves the generalization ability of the model and ensures stable detection capability in different machine and different batch production data. The normalized data can be input into the pre-trained data anomaly detection model to improve the stability and convergence speed of model calculation.

[0066] Step S2: input the pre-processed sensor data into the pre-trained data anomaly detection model to obtain reconstructed data of the sensor data output by the data anomaly detection model.

[0067] As shown in Figure 2 , in the embodiment, the data anomaly detection model includes an encoder, a latent feature layer, and a decoder.

[0068] The encoder is used to extract the feature representation of the sensor data and input it to the latent feature layer.

[0069] The latent feature layer is used to perform linear transformation on the extracted feature representation.

[0070] The decoder is used to reconstruct the sensor data according to the output of the latent feature layer.

[0071] In the embodiment, the encoder includes a first fully connected layer and a first multi-head self-attention mechanism layer.

[0072] The first fully connected layer is used to convert the original input of the sensor data into more easily processed feature data through linear transformation.

[0073] The first multi-head self-attention mechanism layer is used to calculate the feature correlation of the feature data, capture the global dependency relationship between data points at different time steps, and calculate the information interaction between each time step based on the attention weight assignment, to generate a first context representation.

[0074] In the embodiment, the decoder includes a second fully connected layer and a second multi-head self-attention mechanism layer.

[0075] The second multi-head self-attention mechanism layer is used to reconstruct the global dependency of the original input sensor data according to the output of the latent feature layer, calculate the attention weight between data points at different time steps, strengthen the feature relationship, and generate a second context representation for data reconstruction.

[0076] The second fully connected layer is used to receive the second context representation and map to the space of the original input data through a linear transformation and a nonlinear activation function, to generate reconstructed data.

[0077] Specifically, the data anomaly detection model is a deep learning model trained in an autoencoder structure. The training method adopts unsupervised learning, and the training data is unlabeled, that is, there is no explicit "normal" or "abnormal" label. The model learns the normal mode of the data and constructs a network that can reconstruct the input data. In the autoencoder structure, the structure of the encoder is designed as a fully connected layer + multi-head self-attention mechanism, and the structure of the decoder is designed as a multi-head self-attention mechanism + fully connected layer. The reconstructed data is compressed by the encoder, and then restored by the decoder to a form as close as possible to the original data. The latent feature layer stores the features of the encoded sensor data and provides them to the decoder for reconstruction. Compared with a simple fully connected layer or a convolutional layer, the multi-head attention mechanism can better capture the time series relationship of the data, so that the decoded data can be closer to the original data. In the unsupervised learning anomaly detection task, the latent feature layer stores the key patterns of the normal data. If the pattern of the new input data is too different from the training data, the latent feature layer cannot effectively represent the data, resulting in a larger reconstruction error of the decoder, so that the data point can be judged as abnormal, which is helpful for anomaly detection.

[0078] In this embodiment, the data anomaly detection model is trained based on an unsupervised deep learning method, and the training process includes:

[0079] Encoder training:

[0080] The first fully connected layer is used to perform linear transformation on the input sensor data, so that the data is converted into more easily processed feature data.

[0081] The first multi-head self-attention mechanism layer is used to calculate the attention score between the feature data and assign attention weights to identify the global dependency between data points at different time steps and generate a first context representation.

[0082] The feature conversion operation is used to map the extracted first context representation to a latent feature space to form a feature representation for representing the original input data.

[0083] Specifically, step 1, assuming that the input data is entering the first fully connected layer of the encoder, can be represented as:

[0084]

[0085] wherein x1 is the feature data, w1 and b1 are the weight matrix and bias term of the first fully connected layer respectively, and f is the activation function Relu of the first fully connected layer.

[0086] The role of the fully connected layer is to convert the input data The feature extraction and dimension conversion are performed through linear transformation to convert the original input of the sensor data into a more easily processed feature representation. The feature dimension of the data can be adjusted to adapt to the subsequent self-attention mechanism calculation.

[0087] The first multi-head self-attention mechanism layer identifies the global dependency relationship between data points at different time steps, including:

[0088] The query vector, key vector and value vector are constructed, wherein the query vector represents the degree of association between different data points, the key vector represents the data feature information of the data points to provide the association feature between the data points for matching with the query vector, and the value vector represents the data content information of the data points;

[0089] The dot product of the query vector and the key vector is calculated and processed by scaling to obtain the attention score between the data points;

[0090] The attention score is converted into a probability distribution by a normalization function to serve as the attention weight of each data point;

[0091] The value vector is weighted and summed by the attention weight to obtain the context representation of each data point.

[0092] Specifically, in the wafer processing process, the data collected by the sensor is usually time series data, such as time series of process parameters such as temperature, air pressure, current, etc. When the model calculates the features of a certain time step , it not only considers the data of this time step, but also refers to other time steps in the entire sequence to determine which historical data is important to the current time step. In the self-attention mechanism, the data points at each time step (the position of each element in the data) are mapped into a query vector Query (Q), a key vector Key (K) and a value vector Value (V). The query vector Query represents the attention degree of the current time step data point to other time step data. The key vector Key represents the data feature information of other positions, which is used to match with the query vector Query to measure the relevance of the current data point and other data points. The value vector Value represents the data content information of other positions, which is finally used for weighted summation according to the calculated attention weight, thereby providing an enhanced information representation for the current position.

[0093] Step 2, for each data point in the sequence, a Query vector, a Key vector, and a Value vector are generated through different linear transformations (usually weight matrix multiplication). After the feature data enters the first multi-head self-attention mechanism layer, it is divided into multiple heads, and each head independently calculates the query vector Query (Q), the key vector Key (K), and the value vector Value (V).

[0094] Step 3, for each element position i in the data j , calculate the attention score between it and other positions in the sequence i (in fact, calculate the relevance between time step t i and time step t j , measure the similarity between the current data point and other data points, the larger the value, the more attention the data point in position j pays to the data point in position ), which is completed by the dot product operation of the query vector Q and the key vector K and then scaling:

[0095]

[0096] where, is the query vector of position i , is the key vector of position j , k is the dimension of the key vector K, d is the scaling factor to prevent gradient vanishing or explosion.

[0097] Step 4, convert the calculated attention score into a probability distribution through the softmax function to serve as the attention weight of each position data point :

[0098]

[0099] where, is the weight distribution of position i at each position, is the matrix inversion of the key vector.

[0100] Step 5, weight the value vector Value of all positions with the calculated attention weight to obtain the first context representation i of the current position :

[0101]

[0102] Step 6, the first context representation Further mapped to the latent feature space through linear transformation to form the final feature representation for representing the core features of the input data:

[0103]

[0104] where, is the final latent feature layer representation, w2, b2 are the weight matrix and bias term of the latent feature layer respectively, is the activation function of the latent feature layer.

[0105] Decoder training:

[0106] Through the second multi-head self-attention mechanism layer, the information of the feature representation is reconstructed, the attention score between the feature representations is calculated, and the attention weight is allocated to re-identify the global dependency relationship between the data points at different time steps, and generate the second context information for data reconstruction;

[0107] Through the second fully connected layer, the decoded second context information is mapped to convert the data back to the same dimension as the original input data, so as to complete the data reconstruction process.

[0108] After feature extraction by the encoder, the input data has been converted into a feature representation of the latent feature layer The task of the decoder is to reconstruct the input data based on the feature representation And generate the final reconstructed data.

[0109] Step 7, the decoder receives the output from the latent feature layer and starts to reconstruct the original input.

[0110] Step 8, the decoder also processes the input data through the multi-head self-attention mechanism layer, and its logic is the same as the encoder. Map each data point in the data into a query vector Query, a key vector Key and a value vector Value, calculate the attention weight, and weighted sum to generate a context representation.

[0111] Step 9, assuming the output of the decoder is , the output result is entered into a second fully connected layer, which can be represented as:

[0112]

[0113] where, represents the context information received by the decoder from the latent feature layer, w, b are the weight matrix and bias term of the second fully connected layer respectively, is the activation function of the second fully connected layer.

[0114] Step 10, the is linearly transformed to the output layer, which can be represented as:

[0115]

[0116] wherein, is the final reconstructed data after decoding, and are the weight matrix and bias term of the decoder, respectively, is the activation function of the output layer.

[0117] Model optimization:

[0118] The loss value is generated by calculating the error between the original input data and the reconstructed data through the loss function, and the parameters of the data anomaly detection model are updated using the back propagation algorithm. The training process is continuously iteratively optimized until the error converges or reaches the set training condition until convergence, so as to obtain the trained data anomaly detection model.

[0119] Step 11, the model training target is to minimize the error between the original input data and the reconstructed data , and the loss function used is the mean square error:

[0120]

[0121] During the training process, the parameters of the model are continuously adjusted from step 1 to step 11 to make the error as small as possible, until the error converges or reaches the set training condition until convergence, the model training is completed, and it is ensured that the model can accurately reconstruct normal data. If the reconstructed data is very close to the original input data , it means that the model successfully reconstructs the normal data, and the error is small. If the reconstructed data and the original input data have large differences, it means that the model has not seen similar patterns, and the error can be used to judge abnormal data points.

[0122] Compared with traditional convolutional neural network CNN or recurrent neural network RNN, which only focuses on local windows or short-time dependencies, the multi-head self-attention mechanism can directly capture the global features of the entire time series data, and is particularly suitable for anomaly detection of sensor data in the wafer processing process. The multi-head self-attention mechanism can also dynamically allocate attention weights, so that the model automatically focuses on the most important features for anomaly detection, rather than considering all input information equally, thereby improving detection accuracy. Therefore, the design of the autoencoder structure fully combines the feature mapping ability of the fully connected layer and the global modeling ability of the self-attention mechanism, forming an optimized structure that balances computational efficiency and detection accuracy.

[0123] Step S3: Calculate the error between the sensor data and the reconstructed data to determine the degree of abnormality, determine the abnormal data points according to the degree of abnormality, and mark and display the abnormal data points.

[0124] In the detection phase, the abnormal situation is judged by comparing the error between the input data and the output data reconstructed by the model. If the error is small, it means that the data conforms to the normal mode and is normal data. If the error is large, it means that the data deviates greatly from the normal mode and may be an abnormal data point. By setting a threshold or based on error ranking, the data point with the largest degree of abnormality is found and marked. Visualization support is provided in the FDC system, and abnormal data points are displayed through abnormal curves, color identification, alarm push, etc. This helps engineers quickly locate the cause of the anomaly, reduces the impact of anomalies in the wafer processing process, and improves product quality. Abnormal data points and associated time steps, sensor IDs, device IDs, etc. can also be written into the log by the data storage system.

[0125] Taking the etching process as an example, the sensor data at the parameter level is obtained from the FDC system, as shown in FIG. 1. Figure 3 The sensor data is modeled and trained to obtain a data anomaly detection model. The latest monitored data of the sensor is input into the data anomaly detection model after data preprocessing (data cleaning, normalization processing, etc.), and the model outputs reconstructed data, which is compared with the original data to calculate the degree of abnormality (measured by mean square error). Data points with a large degree of abnormality exceeding the threshold are marked and displayed on the front-end interface through a chart, as shown in FIG. 2. Figure 4 As shown in FIG. 2, a warning line (warning threshold T) is set, and if the degree of abnormality is greater than T, the system determines that the data point is abnormal and triggers an alarm. Detailed information such as EQP, Chamber, Recipe, Lot, Wafer, Parameter, Product, etc. is displayed in the alarm details (Alarm Detail). Engineers can check the Alarm Detail data to help engineers quickly locate the problem and decide whether to take further action, such as adjusting process parameters or checking device status.

[0126] The method is based on an unsupervised learning autoencoder model and can be widely used in various process scenarios in semiconductor manufacturing processes, such as CVD (Chemical Vapor Deposition), PVD (Physical Vapor Deposition), CMP (Chemical Mechanical Polishing), etc. Deep learning does not require sample labeling, the model learns automatically, and the relevant information is sent to the corresponding area engineer. It has efficient and automatic anomaly detection capability, reduces manual intervention, and improves the work efficiency of engineers.

[0127] In a semiconductor manufacturing process, an area usually contains multiple EQPs, each of which is responsible for a specific process step and performs the same or similar process (such as etching, deposition, polishing, etc.). During the operation of the EQP, multiple sensors monitor the process parameters in real time. When abnormal conditions are based on multiple process parameters, the abnormal state of the entire device can be identified. For EQPs with many abnormal parameters or high abnormality degree of key parameters, the system will increase the abnormality degree of the EQP level. By cross-EQP comparison and analysis, it is determined whether the abnormality of the EQP is an individual problem or a whole area problem. If the abnormality degrees of multiple EQPs increase, it indicates that the entire area (such as the etching area, the deposition area, and the polishing area) may have a global abnormality. Therefore, this method is suitable for abnormality identification at different levels of Area (area), EQP (machine), and Parameter (parameter) in semiconductor manufacturing. Single parameter abnormalities can be detected first to avoid false positives and unnecessary downtime. The impact of the EQP level is then analyzed to determine whether the abnormality is an individual device problem or a whole process problem. Finally, it is determined whether the area level is affected to ensure process stability and support process optimization and adjustment.

[0128] In summary, the wafer processing process sensor data anomaly detection method provided by the present application acquires sensor data in the wafer processing process, and performs data preprocessing on the sensor data; inputs the sensor data after data preprocessing into a pre-trained data anomaly detection model to obtain reconstructed data of the sensor data output by the data anomaly detection model; calculates the error between the sensor data and the reconstructed data to determine an abnormality degree, determines abnormal data points according to the abnormality degree, and marks and displays the abnormal data points. The present application uses deep learning modeling to generate reconstructed data through a deep learning model. The error between real-time sensor data and reconstructed data output by the model is calculated to intelligently determine abnormal data points. This avoids false positives or false negatives caused by the limitations of fixed threshold settings, improves the automation and intelligence level of anomaly detection, and is more accurate and reliable than traditional threshold setting methods. In this way, the workload of manual operation can be reduced, product yield can be improved, and economic benefits can be created.

[0129] Those skilled in the art can understand that in the above method of the specific implementation, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0130] In addition, some embodiments of the present application also provide an electronic device. The electronic device can be various forms of digital computers, such as a laptop computer, a desktop computer, a workstation, a personal digital assistant, a server, a blade server, a mainframe computer, and the like. The electronic device can also be various forms of mobile devices, such as a personal digital processing, a cellular phone, a smart phone, a wearable device, and other similar computing devices.

[0131] The electronic device includes one or more processors, and a memory storing computer program instructions which, when executed, cause the processor to perform the wafer processing process sensor data anomaly detection method provided by any one or more embodiments described above. Figure 5 An exemplary structural diagram of the electronic device is disclosed. As shown in Figure 5 The electronic device includes one or more processors 1101, a memory 1102, and an interface for connecting various components, including a high-speed interface and a low-speed interface. Various components are connected to each other by different buses, and can be installed on a common motherboard or otherwise installed as desired. The processor can process instructions executed within the electronic device, including instructions stored in the memory or on the memory to display a GUI on an external input / output device (such as a display device coupled to the interface). In some other embodiments, multiple processors and / or buses can be used with multiple memories and multiple storage devices, if desired. Also, multiple electronic devices can be connected, each device providing part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Among them, the components shown herein, their connections and relationships, and their functions are only examples, and are not intended to limit the implementation of the present application described and / or claimed herein.

[0132] The electronic device can also include an input device 1103 and an output device 1104. The processor 1101, the memory 1102, the input device 1103, and the output device 1104 can be connected by a bus or other means, Figure 5 For example, in the middle, by way of example.

[0133] The input device 1103 can receive input of a number or character information, and generate a key signal input relating to user settings and function control of the electronic device, such as a touch screen, a keypad, a mouse, a trackpad, a touchpad, a pointing stick, one or more mouse buttons, a trackball, a joystick, etc. The output device 1104 can include a display device, an auxiliary lighting device (e.g., an LED), a haptic feedback device (e.g., a vibration motor), etc. The display device can include, but is not limited to, a liquid crystal display (LCD), a light emitting diode (LED) display, and a plasma display. In some embodiments, the display device can be a touch screen.

[0134] To provide for interaction with a user, the electronic device can be a computer. The computer has a display device (e.g., a cathode-ray tube (CRT) or LCD monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0135] In the embodiments of the present application, the computer program / instruction is stored on the computer readable medium, and the computer program / instruction is executed by the processor to implement the wafer processing process sensor data anomaly detection method provided by any one or more of the above embodiments. The computer readable medium can be included in the electronic device described in the above embodiments; or can exist separately and not be assembled into the device. The computer readable medium carries one or more computer readable instructions.

[0136] The memory 1102 can be used as a non-transitory computer readable storage medium, and can be used to store non-transitory software programs, non-transitory computer executable programs and modules. The processor 1101 executes various function applications and data processing of the server by running the non-transitory software programs, instructions and modules stored in the memory 1102, so as to implement the program instructions / modules corresponding to the method provided by any one or more of the above embodiments in the embodiments of the present application.

[0137] The memory 1102 can include a program storage area and a data storage area. The program storage area can store an operating system, application programs, and / or data required by at least one function. The data storage area can store data created by the electronic device, etc. In addition, the memory 1102 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one disk memory device, a flash memory device, or other non-volatile solid-state memory device. In some embodiments, the memory 1102 can optionally include a memory that is remotely located from the processor 1101 and can be connected to the electronic device via a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0138] It should be noted that more specific examples of the computer-readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present application, a computer-readable storage medium can be any tangible medium that contains or stores a program used by an instruction execution system, apparatus, or device to function or be combined with it.

[0139] Computer-readable storage media includes permanent and non-permanent, removable and non-removable media, which can be implemented by any method or technology to store information. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM, Phase-Change Random-Access Memory), static random access memory (SRAM, Static Random-Access Memory), dynamic random access memory (DRAM, Dynamic Random-Access Memory), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM, Electrically Erasable Programmable Read-Only Memory), flash memory or other memory technologies, compact discs (CD-ROM), digital versatile discs (DVD) or other optical storage, magnetic cassette, magnetic tape disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device.

[0140] Computer program code for carrying out operations of the present application can be written in one or more programming languages or combinations of languages including object oriented programming languages such as Java, Smalltalk, C++ or conventional procedural programming languages such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0141] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. For example, an Application Specific Integrated Circuit (ASIC), a general purpose computer or any other similar hardware device can be used. In some embodiments, the software program of the present application can be executed by a processor to implement the above steps or functions. Similarly, the software program of the present application (including related data structures) can be stored in a computer readable recording medium, such as a RAM memory, a magnetic or optical drive or a soft disk and the like. In addition, some steps or functions of the present application can be implemented by hardware, for example, as a circuit cooperating with the processor to perform the respective steps or functions.

[0142] The computer program product provided by the embodiments of the present application includes one or more computer programs / instructions, which, when executed by a processor, generate all or part of the processes or functions described in the embodiments of the present application. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) mode. The computer readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media. The available media can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid state disk, SSD, solid state disk) and the like.

[0143] The computer program product of the present application can be a computer program embodied on a non-transitory computer readable medium. Such a medium includes, but is not limited to, a compact disc, a memory device, a hard disk drive, a solid state drive, a magnetic-based carrier, and an optical based carrier. The computer program product can also be a propagated signal as described above. The computer program product can also be a computer program, which, when invoked, carries out operations of the present application on one or more appropriate data processing devices.

[0144] The above description is only specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be limited by the protection scope of the claims, and the above-mentioned embodiments should be regarded as exemplary and non-limiting.

Claims

1. A wafer processing process sensor data anomaly detection method, characterized by, The method comprises the following steps: obtaining sensor data in a wafer processing process and performing data preprocessing on the sensor data; inputting the data preprocessed sensor data into a pre-trained data anomaly detection model to obtain reconstructed data of the sensor data output by the data anomaly detection model, wherein the data anomaly detection model comprises an encoder, a latent feature layer and a decoder; the encoder is used to extract a feature representation of the sensor data and input it into the latent feature layer; the latent feature layer is used to map the feature representation to a latent feature space through linear transformation, form a latent feature layer representation for representing core features of the feature representation through an activation function of the latent feature layer, and output it to the decoder for reconstruction; the decoder is used to reconstruct the sensor data according to the output of the latent feature layer; calculate the error between the sensor data and the reconstructed data to determine the degree of abnormality, determine the abnormal data points according to the degree of abnormality, and mark and display the abnormal data points; the data anomaly detection model is trained based on an unsupervised deep learning method. In the unsupervised learning anomaly detection task, the latent feature layer stores the key mode of normal data. If the mode of the newly input sensor data and the training data is too different, the latent feature layer cannot effectively represent the newly input sensor data, resulting in a larger reconstruction error of the decoder, so that the newly input sensor data can be determined as the abnormal data point.

2. The wafer processing procedure sensor data anomaly detection method of claim 1, wherein, The step of obtaining sensor data in a wafer processing process and performing data preprocessing on the sensor data comprises: obtaining sensor data in a wafer processing process and performing data preprocessing on the sensor data comprises: obtaining sensor data in a wafer processing process and performing data preprocessing on the sensor data comprises: obtaining sensor data in a wafer processing process and performing data preprocessing on the sensor data comprises:

3. The wafer fabrication process sensor data anomaly detection method of claim 1, wherein, the encoder comprises a first fully connected layer and a first multi-head self-attention mechanism layer; the first fully connected layer is used to convert the original input of the sensor data into more easily processed feature data through linear transformation; the first multi-head self-attention mechanism layer is used to calculate the feature correlation of the feature data, capture the global dependency relationship between data points at different time steps, and calculate the information interaction between each time step based on attention weight assignment to generate a first context representation.

4. The wafer processing procedure sensor data anomaly detection method of claim 3, wherein, the decoder comprises a second fully connected layer and a second multi-head self-attention mechanism layer; the second multi-head self-attention mechanism layer is used to reconstruct the global dependency relationship of the original input of the sensor data according to the output of the latent feature layer, calculate the attention weight between data points at different time steps, strengthen the feature relationship, and generate a second context representation for data reconstruction; the second fully connected layer is used to receive the second context representation and map it to the space of the original input data through linear transformation and nonlinear activation function to generate reconstructed data.

5. The wafer processing procedure sensor data anomaly detection method of claim 4, wherein, The training process of the data anomaly detection model comprises: Encoder training: Linearly transforming the input sensor data through a first fully connected layer to convert the data into more easily processed feature data; Calculating the attention scores between the feature data through a first multi-head self-attention mechanism layer and assigning attention weights to identify the global dependency between data points at different time steps and generate a first context representation; Mapping the extracted first context representation to a latent feature space through a feature conversion operation to form a feature representation for representing the original input data; Decoder training: Reconstructing information for the feature representation through a second multi-head self-attention mechanism layer, calculating the attention scores between the feature representation, and assigning attention weights to re-identify the global dependency between data points at different time steps and generate second context information for data reconstruction; Mapping the decoded second context information through a second fully connected layer to convert the data back to the same dimension as the original input data, thereby completing the data reconstruction process; Model optimization: Calculating the error between the original input data and the reconstructed data through a loss function to generate a loss value, updating the parameters of the data anomaly detection model using a backpropagation algorithm, and iteratively optimizing during the training process until the error converges or the set training conditions are met to obtain the trained data anomaly detection model.

6. The wafer processing procedure sensor data anomaly detection method of claim 5, wherein, The first multi-head self-attention mechanism layer and the second multi-head self-attention mechanism layer identify the global dependency between data points at different time steps, comprising: Constructing query vectors, key vectors, and value vectors, wherein the query vectors represent the degree of association between different data points, the key vectors represent the data feature information of the data points to provide the association features between the data points for matching with the query vectors, and the value vectors represent the data content information of the data points; Calculating the dot product of the query vectors and the key vectors and performing scaling processing to obtain the attention scores between the data points; Converting the attention scores into probability distributions through a normalization function to serve as the attention weights for each data point; Weighted summing the value vectors through the attention weights to obtain the context representation of each data point.

7. An electronic device, comprising: The electronic device comprises: One or more processors; and a memory storing computer program instructions that, when executed, cause the processors to perform the wafer processing process sensor data anomaly detection method of any one of claims 1-6.

8. A computer readable storage medium having stored thereon computer programs / instructions, characterized in that, The computer program / instructions, when executed by the processor, implement the wafer processing process sensor data anomaly detection method of any one of claims 1-6.

9. A computer program product comprising computer programs / instructions, characterized in that, The computer program / instructions, when executed by the processor, implement the wafer processing process sensor data anomaly detection method of any one of claims 1-6.

Images