Control method, device, computer device and storage medium of semiconductor manufacturing system
By using virtual measurement sampling rate and measurement prediction model, the timeliness and efficiency of physical measurement in semiconductor manufacturing systems are solved, achieving efficient measurement control and reducing costs and the generation of defective products.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CSMC TECH FAB2 CO LTD
- Filing Date
- 2024-09-12
- Publication Date
- 2026-07-31
AI Technical Summary
In existing semiconductor manufacturing systems, physical measurements cannot be performed in a timely manner, resulting in time lags and high measurement costs.
The current process parameters are obtained by virtual measurement sampling rate, and the trained measurement prediction model is used to make prediction measurements. Combined with failure rate calculation, the process equipment is controlled to stop to prevent the generation of abnormal products.
It enables timely physical measurement, improves measurement efficiency, reduces costs, and prevents the generation of large quantities of abnormal products.
Smart Images
Figure CN121657586B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of semiconductor manufacturing technology, and in particular to a control method, apparatus, computer equipment, and storage medium for a semiconductor manufacturing system. Background Technology
[0002] In semiconductor manufacturing, process measurement is the cornerstone of ensuring process stability and product quality. For semiconductor manufacturing systems, efficiently managing and implementing diverse process measurements for hundreds or even thousands of product types is a crucial issue for high-quality operation.
[0003] Physical measurements in semiconductor manufacturing include thickness, stripe width, mechanical properties, and deformation measurements. Physical measurement stations are typically located after a certain stage of the process flow. Traditionally, measurement relies on statistical principles for sampling inspection, indirectly monitoring the process stability of equipment over a specific period by sampling a portion of the product. However, traditional measurement methods directly measure the physical quantities of the sampled wafers using measurement equipment, which inevitably introduces a time lag, preventing timely measurement of these physical quantities. Summary of the Invention
[0004] Therefore, it is necessary to provide a control method, apparatus, computer equipment, and storage medium for a semiconductor manufacturing system that can measure physical quantities in a timely manner, addressing the aforementioned technical problems.
[0005] In a first aspect, this application provides a control method for a semiconductor manufacturing system. The method includes: acquiring multiple current process parameters based on a virtual measurement sampling rate; wherein the virtual measurement sampling rate is the proportion of virtual measurements to all measurements; inputting each of the current process parameters into a trained measurement prediction model to obtain multiple predicted measurement results; wherein the measurement prediction model is trained from a training dataset, the training dataset including training process parameters and their corresponding training measurement results; determining a failure rate based on each of the predicted measurement results; wherein the failure rate is the proportion of data exceeding the target measurement range in all predicted measurement results to all predicted measurement results; and controlling the process equipment to shut down when the failure rate exceeds a preset threshold.
[0006] In one embodiment, the method further includes: obtaining the determination coefficient of the trained measurement prediction model; if the determination coefficient is less than or equal to a preset coefficient, updating the training dataset to retrain the measurement prediction model until the determination coefficient of the trained measurement prediction model is greater than the preset coefficient.
[0007] In one embodiment, after the step of obtaining the determination coefficient of the trained measurement prediction model, the method further includes: reducing the virtual measurement sampling rate if the determination coefficient is less than or equal to a preset coefficient.
[0008] In one embodiment, the method further includes: setting model training parameters; wherein the model training parameters include: the number of decision trees, the maximum depth of each tree, the minimum number of samples per node, the minimum number of samples per leaf node, and the maximum number of features; training multiple decision trees based on the model training parameters and the training dataset to obtain the trained measurement prediction model.
[0009] In one embodiment, the method further includes: acquiring a historical dataset; wherein the historical dataset includes historical process parameters and corresponding historical measurement results; cleaning the historical dataset to obtain a cleaned dataset; and refining the cleaned dataset based on the correlation between the process parameters and the measurement results to obtain the training dataset.
[0010] In one embodiment, the step of training multiple decision trees based on the model training parameters and the training dataset to obtain the trained measurement prediction model includes: performing multiple random samplings on the training dataset to obtain multiple subsample sets; training one decision tree based on the model training parameters and each subsample set; and integrating the trained multiple decision trees to obtain the trained measurement prediction model.
[0011] In one embodiment, the training process parameters include: process gas concentration, process pressure, process temperature, and process time, and the training measurement results include: actual measured film thickness.
[0012] Secondly, this application also provides a control device for a semiconductor manufacturing system. The device includes: a process parameter acquisition module, used to acquire multiple current process parameters based on a virtual measurement sampling rate; wherein the virtual measurement sampling rate is the proportion of virtual measurements to all measurements; a measurement result prediction module, used to input each of the current process parameters into a trained measurement prediction model to obtain multiple predicted measurement results; wherein the measurement prediction model is obtained by training a training dataset, and the training dataset includes training process parameters and their corresponding training measurement results; a failure rate calculation module, used to determine the failure rate based on each of the predicted measurement results; wherein the failure rate is the proportion of data exceeding the target measurement range in all the predicted measurement results to all the predicted measurement results; and a process equipment control module, used to control the process equipment to stop when the failure rate is greater than a preset threshold.
[0013] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the above-described method.
[0014] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.
[0015] The aforementioned semiconductor manufacturing system's control method, apparatus, computer equipment, and storage medium acquire multiple current process parameters based on a virtual measurement sampling rate. These parameters are then input into a trained measurement prediction model to obtain multiple predicted measurement results. Since the current process parameters are immediately available after the corresponding process flow, the measurement prediction model quickly yields the corresponding predicted measurement results. Compared to measurement using measuring machines, this method offers higher measurement efficiency and allows for timely measurement of physical quantities. Furthermore, after obtaining multiple predicted measurement results, the system determines the failure rate, thereby identifying the proportion of data exceeding the target measurement range in all predicted measurement results. If the failure rate exceeds a preset threshold, the system shuts down the process equipment for timely maintenance, preventing the generation of large batches of defective products. Attached Figure Description
[0016] Figure 1 This is a flowchart illustrating a control method for a semiconductor manufacturing system in one embodiment;
[0017] Figure 2 This is a flowchart illustrating a process where the determination coefficient is less than or equal to a preset coefficient in one embodiment.
[0018] Figure 3 This is a flowchart illustrating the training process of a measurement prediction model in one embodiment.
[0019] Figure 4 This is a schematic diagram illustrating the process of obtaining the training dataset in one embodiment;
[0020] Figure 5 This is a flowchart illustrating the training process of a measurement prediction model in another embodiment;
[0021] Figure 6 This is a flowchart illustrating the training process of a measurement prediction model in yet another embodiment;
[0022] Figure 7 This is a comparison chart of predicted and actual measurement results in one embodiment;
[0023] Figure 8 for Figure 7 A schematic diagram showing the difference between the predicted measurement results and the actual measurement results;
[0024] Figure 9 This is a comparison chart of predicted and actual measurement results in another embodiment;
[0025] Figure 10 for Figure 9 A schematic diagram showing the difference between the predicted measurement results and the actual measurement results;
[0026] Figure 11 This is a schematic diagram of the control device of a semiconductor manufacturing system in one embodiment;
[0027] Figure 12 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation
[0028] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0029] The control method for a semiconductor manufacturing system provided in this application can be applied to semiconductor manufacturing systems. Semiconductor manufacturing systems are used for wafer fabrication, packaging, and testing. During wafer fabrication, the semiconductor manufacturing system needs to control different process equipment to perform different process flows (such as oxidation, photolithography, etching, deposition, ion implantation, etc.). After the corresponding process flow is completed, sampling inspection is generally used to measure the physical quantities of the wafer (such as thickness, stripe width, mechanical properties, deformation, etc.) to indirectly monitor the process stability of the equipment. However, during product sampling inspection, there is inevitably a time lag, making it impossible to measure the physical quantities of the wafer in a timely manner. Furthermore, at high production volumes, this method occupies a large amount of measurement equipment capacity, resulting in high measurement costs.
[0030] In one embodiment, such as Figure 1 As shown, a control method for a semiconductor manufacturing system is provided. Taking the application of this method to a semiconductor manufacturing system as an example, the method includes the following steps:
[0031] Step S110: Obtain multiple current process parameters based on the virtual measurement sampling rate.
[0032] Specifically, the semiconductor manufacturing system first determines the number of virtual measurements required based on a preset virtual measurement sampling rate, and then acquires a corresponding number of current process parameters based on this number of virtual measurements. The virtual measurement sampling rate is the proportion of virtual measurements to all measurements, which also includes actual measurements performed using measurement equipment. For example, if a batch of products is processed by multiple identical process devices in a certain process flow, requiring a total of 20 measurements, and the virtual measurement sampling rate is set to 50%, then both 10 virtual measurements and 10 actual measurements are required.
[0033] When obtaining a corresponding number of current process parameters based on the determined number of virtual measurements, a corresponding number of process devices are randomly selected based on the number of virtual measurements, and then the corresponding process parameters are directly obtained from the corresponding process devices, thereby obtaining multiple current process parameters.
[0034] Step S120: Input each current process parameter into the trained measurement prediction model to obtain multiple predicted measurement results.
[0035] Specifically, after acquiring multiple current process parameters, the semiconductor manufacturing system inputs each parameter into a trained measurement prediction model. The trained model then processes these parameters to obtain multiple predicted measurement results. The measurement prediction model is trained using a training dataset, which includes the training process parameters and their corresponding training measurement results. The measurement prediction model can be trained using regression models such as decision tree regression, support vector machine regression, neural network regression, random forest regression, and deep forest regression.
[0036] Step S130: Determine the failure rate based on the results of each predicted measurement.
[0037] Specifically, after obtaining multiple predicted measurement results, the semiconductor manufacturing system determines the failure rate based on each predicted measurement result. The failure rate is the proportion of data exceeding the target measurement range among all predicted measurement results. To determine the failure rate, first, it is determined whether each predicted measurement result exceeds the target measurement range; then, the number of predicted measurement results exceeding the target measurement range is determined; finally, the proportion of data exceeding the target measurement range among all predicted measurement results is calculated, thus obtaining the failure rate.
[0038] Step S140: If the failure rate is greater than a preset threshold, control the process equipment to stop.
[0039] Specifically, after determining the failure rate, the semiconductor manufacturing system compares the failure rate with a preset threshold. If the failure rate exceeds the preset threshold, it indicates that the current batch of products has experienced numerous anomalies after processing through the current process flow, potentially leading to an increased scrap rate. Therefore, the process equipment executing the current process flow is shut down. It is understood that when controlling the shutdown of process equipment, only equipment whose predicted measurement results exceed the target measurement range may be shut down; only the process equipment generating anomalies may be stopped. In some embodiments, the preset threshold is set to 1%.
[0040] The control method of the aforementioned semiconductor manufacturing system acquires multiple current process parameters based on a virtual measurement sampling rate. These parameters are then input into a trained measurement prediction model to obtain multiple predicted measurement results. Since the current process parameters are immediately available after the completion of the corresponding process flow, the prediction model quickly yields the corresponding predicted measurement results. Compared to measurement using measuring instruments, this method is more efficient and allows for timely measurement of physical quantities. Furthermore, after obtaining multiple predicted measurement results, the system determines the failure rate, thereby identifying the proportion of data exceeding the target measurement range in all predicted measurement results. If the failure rate exceeds a preset threshold, the system shuts down the process equipment for timely maintenance, preventing the generation of large batches of defective products.
[0041] In one embodiment, such as Figure 2 As shown, the control method for semiconductor manufacturing systems also includes:
[0042] Step S210: Obtain the determination coefficients of the trained measurement prediction model.
[0043] Specifically, in this embodiment, after the measurement prediction model is trained using the training dataset, the coefficient of determination (R²) of the trained measurement prediction model is first obtained. The coefficient of determination (R²) is an indicator used to measure the goodness of fit of a measurement prediction model, ranging from 0 to 1. The closer the R² is to 1, the better the model fit, meaning the independent variable explains a higher degree of variation in the dependent variable, and the higher the percentage of variation caused by the independent variable. Conversely, the closer the R² is to 0, the worse the model fit.
[0044] Step S220: If the determination coefficient is less than or equal to the preset coefficient, update the training dataset to retrain the measurement prediction model until the determination coefficient of the trained measurement prediction model is greater than the preset coefficient.
[0045] Specifically, if the coefficient of determination is less than or equal to a preset coefficient, it indicates that the currently trained measurement prediction model has a poor fit, and the reliability of the predicted measurement results output by the measurement prediction model is not high. In this case, the training dataset is updated based on the actual measurement results and their corresponding process parameters. The updated training dataset is then used to retrain the measurement prediction model, resulting in an updated measurement prediction model. The updated measurement prediction model also has a corresponding coefficient of determination. The determination coefficient of the updated measurement prediction model is then checked again to see if it is less than or equal to the preset coefficient. By repeatedly executing the above steps, the measurement prediction model can be continuously trained until the coefficient of determination of the trained measurement prediction model is greater than the preset coefficient. If the coefficient of determination is greater than the preset coefficient, it indicates that the model has a good fit, and steps S110 to S140 can then be executed. In some embodiments, the preset coefficient is set to 90%.
[0046] In one embodiment, after obtaining the determination coefficient of the trained measurement prediction model in step S210, the control method of the semiconductor manufacturing system further includes: reducing the virtual measurement sampling rate when the determination coefficient is less than or equal to a preset coefficient.
[0047] Specifically, in this embodiment, steps S110 to S140 can be executed if the determination coefficient of the measurement prediction model is less than or equal to a preset coefficient. However, since the measurement prediction model has a poor fit, it is necessary to reduce the virtual measurement sampling rate and increase the proportion of actual measurements. When reducing the virtual measurement sampling rate, it can be gradually reduced by a preset ratio, or the virtual measurement sampling rate can be reduced to a lower preset ratio. While reducing the virtual measurement sampling rate, the training dataset is updated using actual measurement data to retrain the measurement prediction model until the determination coefficient of the trained measurement prediction model is greater than the preset coefficient.
[0048] In one embodiment, such as Figure 3 As shown, the control method for semiconductor manufacturing systems also includes:
[0049] Step S310: Set the model training parameters.
[0050] Specifically, in this embodiment, the measurement prediction model is a random forest model comprising multiple decision trees. When training the measurement prediction model, it is first necessary to set the model training parameters, which are hyperparameters used for model training. These parameters include: the number of decision trees (n_estimators), with a set value ranging from 100 to 500 to ensure sufficient complexity and generalization ability of the model; the maximum depth of each tree (max_depth), which can be set to None, indicating that each tree can grow indefinitely to fully learn the complex relationships in the training data; the minimum number of samples per node (min_samples_split), with a set value ranging from 3 to 20, meaning each internal node contains at least 3 to 20 samples; the minimum number of samples per leaf node (min_samples_leaf), with a set value ranging from 2 to 10, meaning each leaf node contains at least 2 to 10 samples to avoid over-pruning and producing overly simplistic tree structures; and the maximum number of features (max_features), which can be set to auto, meaning that all features are considered for optimal splitting at each node.
[0051] Step S320: Train multiple decision trees based on model training parameters and training dataset to obtain a trained measurement prediction model.
[0052] Specifically, after setting the model training parameters, multiple decision trees are trained based on the set parameters and training dataset to obtain a trained measurement prediction model. When making predictions, the multiple decision trees in the trained measurement prediction model obtain the predicted measurement results using the following formula:
[0053]
[0054] in, This represents the predicted measurement result output by the measurement prediction model, where T is the number of decision trees in the measurement prediction model. It is the predicted value of the i-th decision tree. At each leaf node, the output of the decision tree is the average value of the samples in that node.
[0055] In one embodiment, such as Figure 4 As shown, the control method for semiconductor manufacturing systems also includes:
[0056] Step S410: Obtain historical dataset.
[0057] Specifically, in determining the training dataset, this embodiment first acquires the historical dataset. The historical dataset is a pre-collected dataset that includes historical process parameters and their corresponding historical measurement results. The historical measurement results are the measurement results obtained through actual measurements using measurement equipment.
[0058] Step S420: Clean the historical dataset to obtain a cleaned dataset.
[0059] Specifically, after obtaining the historical dataset, it is first cleaned to obtain a cleaned dataset. During data cleaning, processes such as filling missing values, data transformation, and data normalization can be performed to ensure the data can be correctly identified and utilized by subsequent measurement and prediction models. In some embodiments, when the historical process parameters are those used to generate the film layer, the collected historical process parameters need to be physically correlated with the actual position of the measurement piece within the equipment cavity, retaining the process temperature and process gas concentration at the actual measurement piece location. Then, the historical dataset is grouped to prepare for building multiple mathematical prediction models. If outliers are found in the collected historical dataset, further data cleaning is performed to preserve the data's integrity and accuracy.
[0060] Step S430: Based on the correlation between process parameters and measurement results, the cleaning dataset is purified to obtain a training dataset.
[0061] Specifically, after obtaining the cleaning dataset, the data is purified based on the correlation between process parameters and measurement results to obtain the training dataset. When determining the correlation between process parameters and measurement results, feature selection algorithms, variance selection methods, and correlation coefficient methods can be used to screen out process parameters with high correlation to the measurement result prediction. For example, when the measurement prediction model is used for film thickness prediction, it is necessary to screen out feature parameters with high correlation to film thickness prediction from the process parameters. For instance, the correlation coefficient method can be used to rank the correlation between multiple feature parameters (process gas concentration, process pressure, process temperature, etc.) and film thickness, and the ranking results can be evaluated based on the importance of the features to select the top-ranked feature parameters, which are then used as the training dataset as input data for training the measurement prediction model.
[0062] The training process of the measurement prediction model of this application is described in detail below with a specific embodiment. For example... Figure 5As shown, first, obtain the historical data set, then clean the historical data set to obtain the cleaned data set, and then purify the cleaned data set to obtain the training data set. After obtaining the training data set, train the measurement prediction model in combination with the set model training parameters. After training the measurement prediction model, determine whether the coefficient of determination of the current measurement prediction model is greater than the preset coefficient. If the coefficient of determination is greater than the preset coefficient, it indicates that the fitting degree of the model is good. At this time, it can be used as the target measurement prediction model, and the measurement result can be predicted through it. If the coefficient of determination is less than or equal to the preset coefficient, it indicates that the fitting degree of the model is poor. At this time, reduce the virtual measurement sampling rate to obtain more actual measurement results, and update the training data set with the actual measurement data to retrain the measurement prediction model until the coefficient of determination of the trained measurement prediction model is greater than the preset coefficient.
[0063] In one embodiment, as Figure 6 shown, step S320 includes:
[0064] Step S321, perform multiple random samplings on the training data set to obtain multiple sub-sample sets.
[0065] Specifically, in the process of training the measurement prediction model in this embodiment, first perform multiple random samplings on the training data set to obtain multiple sub-sample sets. For a specific example, the training data set D = {(x1, y1), (x2, y2),..., (x M , y M ), where M represents the total number of samples. x i represents the feature vector of the i-th sample, and y i represents the label of the i-th sample, where i represents the index of the sample, and the value range of i is [1, M]. For each decision tree, a random sampling with replacement needs to be performed on the training data set to obtain a sub-sample set D k , and the size of the sub-sample set is m, m << M.
[0066] Step S322, train a decision tree based on the model training parameters and each sub-sample set.
[0067] Specifically, after obtaining multiple sub-sample sets, train a decision tree according to each sub-sample set, and the number of sub-sample sets and decision trees is the same. When training a decision tree with a sub-sample set, select a feature subset F k from the sub-sample set, and the size of the feature subset is f, f << N, where N is the total number of features. After determining the feature subset, train a decision tree T k , denoted as T k = train(Dk F k ).
[0068] Step S323: Integrate the trained decision trees to obtain the trained measurement prediction model.
[0069] Specifically, after all decision trees have been trained, the trained decision trees are integrated to obtain a trained measurement prediction model. For example, the outputs of multiple decision trees are averaged and used as the output of the measurement prediction model. Specifically, for the first input test sample x, for each decision tree T... k (k=1,2,3,...,K), decision tree T k The prediction result is denoted as y k =Predict(T k The prediction result y output by the measurement prediction model is the average of the prediction results of all decision trees, i.e.:
[0070]
[0071] By following the steps above, multiple decision trees can be integrated into a measurement prediction model, which can then predict measurement results based on the current process parameters.
[0072] In one embodiment, the training process parameters include: process gas concentration, process pressure, process temperature, and process time, and the training measurement results include: the actual measured film thickness. Specifically, the measurement prediction model in this embodiment is used for virtual measurement of film thickness. Correspondingly, the current process parameters obtained also include: process gas concentration, process pressure, process temperature, and process time, and the obtained predicted measurement result is the predicted film thickness. In some embodiments, the training process parameters are the process parameters for generating a silicon nitride (Si3N4) film layer through a diffusion process. In this case, the process pressure is defined as x1; the process gas concentration (NH3, SiH2CL4) is defined as x2, x3; the process temperatures at four uniformly distributed positions on the outer tube are defined as x4, x5, x6, x7; and the process temperatures at four uniformly distributed positions on the inner tube are defined as x8, x9, x1, x2, x3, x4, x5, x6, x7. 10 x 11 The process time is x. 12 The actual measured film thickness is y. It's understandable that historical datasets contain a wider variety of process parameters, and the training dataset obtained after data purification contains process parameters that are highly correlated with the measurement results.
[0073] The measurement prediction model of this application is described below with a specific embodiment as an example. The measurement prediction model is used for virtual measurement of film thickness. The trained measurement prediction model includes 205 decision trees, and the training dataset contains 48,056 data points. The mean squared error (MSE) of the trained measurement prediction model is 0.33, and the coefficient of determination is 0.91, which is greater than the preset coefficient of 0.9. This indicates that the measurement prediction model has strong interpretability of parameter characteristics, the model's predicted values are relatively consistent with the actual measured values, and the prediction confidence is strong. At this time, multiple current process parameters can be input into the trained measurement prediction model to obtain multiple predicted measurement results. Simultaneously, to determine the accuracy of the predicted measurement results, actual measurements are also performed on the corresponding wafers to obtain multiple actual measurement results. For example... Figure 7 The image shown is a comparison chart of predicted and actual measurement results in one embodiment. Figure 8 for Figure 7 The difference between the predicted measurement result and the actual measurement result. For example... Figure 9 As shown, this is a comparison chart of predicted measurement results and actual measurement results in another embodiment. Figure 10 for Figure 9 The difference between the predicted and actual measurement results is shown in the comparison chart above. It can be seen that the maximum error between the predicted and actual measurement results does not exceed 20 Å, indicating that the predicted measurement results obtained through the measurement prediction model are relatively accurate.
[0074] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.
[0075] Based on the same inventive concept, this application also provides a control device for implementing the control method described above. The solution provided by this device is similar to the solution described in the above method; therefore, specific limitations in one or more control device embodiments provided below can be found in the limitations of the control method described above, and will not be repeated here.
[0076] In one embodiment, such as Figure 11As shown, a control device for a semiconductor manufacturing system is provided, including: a process parameter acquisition module 510, a measurement result prediction module 520, a failure rate calculation module 530, and a process equipment control module 540, wherein:
[0077] The process parameter acquisition module 510 is used to acquire multiple current process parameters based on the virtual measurement sampling rate; wherein, the virtual measurement sampling rate is the proportion of virtual measurements to all measurements.
[0078] The measurement result prediction module 520 is used to input each current process parameter into the trained measurement prediction model to obtain multiple predicted measurement results; wherein, the measurement prediction model is obtained by training the training dataset through model training, and the training dataset includes the training process parameters and their corresponding training measurement results;
[0079] The failure rate calculation module 530 is used to determine the failure rate based on each predicted measurement result; wherein, the failure rate is the proportion of data that exceeds the target measurement range in all predicted measurement results to all predicted measurement results.
[0080] The process equipment control module 540 is used to control the process equipment to shut down when the failure rate is greater than a preset threshold.
[0081] In one embodiment, the control device of the semiconductor manufacturing system further includes: a determination coefficient judgment module, used to obtain the determination coefficient of the trained measurement prediction model; and when the determination coefficient is less than or equal to a preset coefficient, to update the training dataset to retrain the measurement prediction model until the determination coefficient of the trained measurement prediction model is greater than the preset coefficient.
[0082] In one embodiment, the determination coefficient judgment module is further used to reduce the virtual measurement sampling rate when the determination coefficient is less than or equal to a preset coefficient.
[0083] In one embodiment, the control device of the semiconductor manufacturing system further includes: a model training module for setting model training parameters; wherein the model training parameters include: the number of decision trees, the maximum depth of each tree, the minimum number of samples per node, the minimum number of samples per leaf node, and the maximum number of features; and multiple decision trees are trained based on the model training parameters and the training dataset to obtain a trained measurement prediction model.
[0084] In one embodiment, the control device of the semiconductor manufacturing system further includes: a data filtering module for acquiring historical datasets; wherein the historical datasets include historical process parameters and their corresponding historical measurement results; the historical datasets are cleaned to obtain cleaned datasets; and the cleaned datasets are purified based on the correlation between process parameters and measurement results to obtain training datasets.
[0085] In one embodiment, the model training module is further configured to perform multiple random samplings on the training dataset to obtain multiple subsample sets; train a decision tree based on the model training parameters and each subsample set; and integrate the trained decision trees to obtain a trained measurement prediction model.
[0086] In one embodiment, the training process parameters include: process gas concentration, process pressure, process temperature, and process time, and the training measurement results include: actual measured film thickness.
[0087] The various modules in the control device of the aforementioned semiconductor manufacturing system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each module.
[0088] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 12 As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a control method for a semiconductor manufacturing system. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.
[0089] Those skilled in the art will understand that Figure 12The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0090] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.
[0091] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method embodiments.
[0092] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
[0093] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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.
[0094] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.
Claims
1. A control method for a semiconductor manufacturing system, characterized in that, The method includes: Multiple current process parameters are obtained based on the virtual measurement sampling rate; wherein, the virtual measurement sampling rate is the proportion of virtual measurements to all measurements; Each of the current process parameters is input into the trained measurement prediction model to obtain multiple predicted measurement results; wherein, the measurement prediction model is obtained by training the training dataset through model training, and the training dataset includes the training process parameters and their corresponding training measurement results; The failure rate is determined based on each of the predicted measurement results; wherein, the failure rate is the proportion of data exceeding the target measurement range in all the predicted measurement results to all the predicted measurement results. If the failure rate exceeds a preset threshold, the process equipment will be shut down. Obtain the determination coefficient of the trained measurement prediction model; wherein the determination coefficient is an index used to measure the goodness of fit of the measurement prediction model; If the determination coefficient is less than or equal to the preset coefficient, update the training dataset to retrain the measurement prediction model until the determination coefficient of the trained measurement prediction model is greater than the preset coefficient. If the determination coefficient is less than or equal to the preset coefficient, the virtual measurement sampling rate is reduced.
2. The control method for a semiconductor manufacturing system according to claim 1, characterized in that, The method further includes: Set the model training parameters; wherein the model training parameters include: the number of decision trees, the maximum depth of each tree, the minimum number of samples per node, the minimum number of samples per leaf node, and the maximum number of features; Multiple decision trees are trained based on the model training parameters and the training dataset to obtain the trained measurement prediction model.
3. The control method for a semiconductor manufacturing system according to claim 2, characterized in that, The method further includes: Obtain historical datasets; wherein, the historical datasets include historical process parameters and their corresponding historical measurement results; The historical dataset is cleaned to obtain a cleaned dataset; The cleaning dataset is purified based on the correlation between process parameters and measurement results to obtain the training dataset.
4. The control method for a semiconductor manufacturing system according to claim 2, characterized in that, The step of training multiple decision trees based on the model training parameters and the training dataset to obtain the trained measurement prediction model includes: Multiple random samplings are performed on the training dataset to obtain multiple subsets of samples; A decision tree is trained based on the model training parameters and each of the sub-sample sets; The trained decision trees are integrated to obtain the trained measurement prediction model.
5. The control method for a semiconductor manufacturing system according to claim 2, characterized in that, The training process parameters include: process gas concentration, process pressure, process temperature, and process time. The training measurement results include: actual measured film thickness.
6. A control device for a semiconductor manufacturing system, characterized in that, The device includes: The process parameter acquisition module is used to acquire multiple current process parameters based on the virtual measurement sampling rate; wherein, the virtual measurement sampling rate is the proportion of virtual measurements to all measurements; The measurement result prediction module is used to input each of the current process parameters into the trained measurement prediction model to obtain multiple predicted measurement results; wherein, the measurement prediction model is obtained by training the model through the training dataset, and the training dataset includes the training process parameters and their corresponding training measurement results; A failure rate calculation module is used to determine the failure rate based on each of the predicted measurement results; wherein, the failure rate is the proportion of data exceeding the target measurement range in all the predicted measurement results to all the predicted measurement results. The process equipment control module is used to control the process equipment to shut down when the failure rate is greater than a preset threshold. The determination coefficient judgment module is used to obtain the determination coefficient of the trained measurement prediction model; wherein the determination coefficient is an index used to measure the goodness of fit of the measurement prediction model; if the determination coefficient is less than or equal to a preset coefficient, the training dataset is updated to retrain the measurement prediction model until the determination coefficient of the trained measurement prediction model is greater than the preset coefficient; if the determination coefficient is less than or equal to the preset coefficient, the virtual measurement sampling rate is reduced.
7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.