Semiconductor process parameter optimization model training method and device, and computer device

Abstract

The application relates to the technical field of deep learning, and discloses a semiconductor process parameter optimization model training method and device and computer equipment, wherein structured sample data is collected, each structured sample data contains process parameters recorded when a semiconductor device is produced, and a real performance index measured from the semiconductor device produced based on the process parameters; a plurality of basic learners are trained respectively by using the structured sample data; the basic learners are used for outputting a predicted performance index according to the process parameters; the prediction accuracy of each basic learner is calculated according to the real performance index and the predicted performance index, and a classification model is generated according to the prediction accuracy; the classification model is used for determining the basic learner for processing the process parameters; a process parameter optimization model is composed based on the classification model and the basic learners; and the process parameter optimization model is used for outputting a predicted performance index according to the process parameters. The application has the beneficial effect of improving the process parameter optimization efficiency.

Description

Technical Field

[0001] This application relates to the field of deep learning technology, and in particular to a method, apparatus and computer equipment for training a semiconductor process parameter optimization model. Background Technology

[0002] As semiconductor processes continue to evolve towards advanced nodes, the performance of complementary metal-oxide-semiconductor (CMOS) devices is increasingly constrained by process parameters.

[0003] Traditionally, optimizing process parameters has relied on a combination of simulation and experience, which is not only inefficient but also severely inadequate in adapting to new processes and designs. To overcome this bottleneck, machine learning technology has been introduced to replace traditional physical modeling methods.

[0004] However, single machine learning models often suffer from low prediction accuracy when dealing with such complex problems, thus failing to improve the efficiency of process parameter optimization based on the prediction results. Summary of the Invention

[0005] This application provides a method, apparatus, and computer equipment for training a semiconductor process parameter optimization model, which solves the technical problem of low efficiency in process parameter optimization and achieves the technical effect of improving the efficiency of process parameter optimization.

[0006] To achieve the above objectives, the main technical solutions adopted in this application include: In a first aspect, embodiments of this application provide a method for training a semiconductor process parameter optimization model, the method comprising: Collect structured sample data, each of which contains process parameters recorded during the production of semiconductor devices, as well as actual performance indicators measured on semiconductor devices produced based on the process parameters. Using the structured sample data, multiple basic learners are trained respectively; the basic learners are used to output predicted performance indicators based on process parameters. Based on the actual performance metrics and the predicted performance metrics, the prediction accuracy of each of the base learners is calculated, and a classification model is generated based on the prediction accuracy; the classification model is used to determine the base learners for processing parameters. Based on the classification model and each of the basic learners, a process parameter optimization model is formed; the process parameter optimization model is used to output predicted performance indicators according to the process parameters.

[0007] In this embodiment, structured sample data is collected and used to train multiple basic learners. Then, a classification model is generated based on the prediction accuracy of each basic learner. The classification model and the basic learners are used to form a process parameter optimization model. This method generates a process parameter optimization model and improves the prediction accuracy of the process parameter optimization model, thereby improving the efficiency of optimizing production process parameters based on the prediction results of the process parameter optimization model.

[0008] In one example, based on the true performance metric and the predicted performance metric, the prediction accuracy of each of the base learners is calculated, and a classification model is generated based on the prediction accuracy, including: The process parameters are classified based on a preset classification algorithm. The prediction accuracy of each of the aforementioned process parameters when using the respective basic learners is statistically analyzed. The classification model is generated based on the mapping relationship between the base learner with the highest prediction accuracy in each class and the class.

[0009] In this example, process parameters are classified using a preset classification algorithm. Based on the prediction accuracy of each base learner in each class, the mapping relationship between each class and the base learner is determined. Then, based on the preset classification algorithm and the mapping relationship between the classes, as well as the classification algorithm, a classification model is generated to classify the process parameters, thereby matching the optimal base learner and improving the prediction accuracy of the process parameters.

[0010] In one example, the process parameters are classified based on a preset classification algorithm, including: The process parameters are classified according to a preset numerical range.

[0011] In this example, by classifying process parameters according to preset numerical ranges and generating classification models, the complex and diverse process parameters are accurately divided to facilitate targeted analysis and processing.

[0012] In one example, based on a preset classification algorithm for the process parameters, the process parameters are classified to generate a classification model, including: The process parameters are clustered to obtain at least one cluster; each cluster contains multiple process parameters. Based on the process parameters contained in each cluster, the cluster center of each cluster is determined; the cluster center is used to generate the classification model.

[0013] In this example, by first clustering the process parameters into clusters, then calculating the cluster center of each cluster, and subsequently generating a classification model based on the cluster center, the process parameters are efficiently grouped and a classification model is constructed, thus improving data processing efficiency.

[0014] In one example, the method of calculating the prediction accuracy of the process parameters based on each of the base learners and generating a classification model based on the prediction accuracy includes: The predicted performance index is obtained by using each of the basic learners to predict all the process parameters; and the difference between the predicted performance index and the actual performance index of each process parameter is calculated. Training data is generated based on the process parameters and the difference values, and a preset classification algorithm is trained to obtain a classification model; the number of categories in the classification model is determined based on the number of base learners. The classes in the classification model correspond one-to-one with the base learners.

[0015] In this example, the difference between the predicted performance index and the actual performance index of each basic learner for predicting all process parameters is calculated, and the classification model is trained based on this difference. This method achieves the effect of matching the basic learner with its corresponding process parameters, thereby improving the utilization rate of the basic learner and the recognition accuracy.

[0016] In one example, training data is generated based on the process parameters and the difference values, and a preset classification algorithm is trained to obtain a classification model, including: The process parameters are input into the classification model for classification, and the class corresponding to each process parameter is determined. Based on the difference value of the process parameters contained in each class when predicted using the base learner corresponding to the class, a first total difference value for the class is calculated; and the first total difference values ​​of all classes are summarized to obtain a second total difference value. Based on the second total difference value, the process parameters of the classification model are optimized to minimize the second total difference value of the classification model.

[0017] In this example, by inputting process parameters into a classification model for classification, calculating the total difference value for each category based on the classification results, and optimizing the process parameters of the classification model based on the total difference value, the accuracy of the classification model for process parameters and the overall prediction accuracy are improved.

[0018] In one example, the method further includes: The process parameters generated using a genetic algorithm are input into a trained process parameter optimization model to generate a predicted performance index; wherein, the process parameters are obtained by optimizing the process parameters of the previous iteration using a genetic algorithm based on the predicted performance index and a preset target performance index; Repeat the above process until the predicted performance index matches the target performance index, then output the process parameters.

[0019] In this example, a genetic algorithm is used to generate process parameters, which are then input into a trained model to generate predicted performance indicators. Based on the difference between the predicted and target performance indicators, the genetic algorithm generates new process parameters, and this process is repeated until the predicted and target performance indicators match. This achieves the goal of accurately obtaining process parameters that meet performance requirements. Based on these process parameters, production line optimization can be implemented.

[0020] Secondly, embodiments of this application provide a semiconductor process parameter optimization model training apparatus, the apparatus comprising: The acquisition module is used to collect structured sample data. Each piece of structured sample data contains process parameters recorded during the production of semiconductor devices, as well as actual performance indicators measured on semiconductor devices produced based on the process parameters. A training module is used to train multiple basic learners using the structured sample data; the basic learners are used to output predicted performance indicators based on process parameters; based on the actual performance indicators and the predicted performance indicators, the prediction accuracy of each basic learner is calculated, and a classification model is generated based on the prediction accuracy; the classification model is used to determine the basic learner for processing the process parameters; the classification model is used to process the basic learner for the process parameters; based on the classification model and each basic learner, a process parameter optimization model is formed; the process parameter optimization model is used to output predicted performance indicators based on the process parameters.

[0021] In this embodiment, structured sample data is collected and used to train multiple basic learners. Then, a classification model is generated based on the prediction accuracy of each basic learner. The classification model and the basic learners are used to form a process parameter optimization model. This method generates a process parameter optimization model and improves the prediction accuracy of the process parameter optimization model, thereby improving the efficiency of optimizing production process parameters based on the prediction results of the process parameter optimization model.

[0022] In one example, the training module is used for: The process parameters are classified based on a preset classification algorithm. The prediction accuracy of each of the aforementioned process parameters when using the respective basic learners is statistically analyzed. The classification model is generated based on the mapping relationship between the base learner with the highest prediction accuracy in each class and the class.

[0023] In this example, process parameters are classified using a preset classification algorithm. Based on the prediction accuracy of each base learner in each class, the mapping relationship between each class and the base learner is determined. Then, based on the preset classification algorithm and the mapping relationship between the classes, as well as the classification algorithm, a classification model is generated to classify the process parameters, thereby matching the optimal base learner and improving the prediction accuracy of the process parameters.

[0024] In one example, the training module is used for: The process parameters are classified according to a preset numerical range.

[0025] In this example, by classifying process parameters according to preset numerical ranges and generating classification models, the complex and diverse process parameters are accurately divided to facilitate targeted analysis and processing.

[0026] In one example, the training module is used for: The process parameters are clustered to obtain at least one cluster; each cluster contains multiple process parameters. Based on the process parameters contained in each cluster, the cluster center of each cluster is determined; the cluster center is used to generate the classification model.

[0027] In this example, by first clustering the process parameters into clusters, then calculating the cluster center of each cluster, and subsequently generating a classification model based on the cluster center, the process parameters are efficiently grouped and a classification model is constructed, thus improving data processing efficiency.

[0028] In one example, the training module is used for: The predicted performance index is obtained by using each of the basic learners to predict all the process parameters; and the difference between the predicted performance index and the actual performance index of each process parameter is calculated. Training data is generated based on the process parameters and the difference values, and a preset classification algorithm is trained to obtain a classification model; the number of categories in the classification model is determined based on the number of base learners. The classes in the classification model correspond one-to-one with the base learners.

[0029] In this example, the difference between the predicted performance index and the actual performance index of each basic learner for predicting all process parameters is calculated, and the classification model is trained based on this difference. This method achieves the effect of matching the basic learner with its corresponding process parameters, thereby improving the utilization rate of the basic learner and the recognition accuracy.

[0030] In one example, the training module is used for: The process parameters are input into the classification model for classification, and the class corresponding to each process parameter is determined. Based on the difference value of the process parameters contained in each class when predicted using the base learner corresponding to the class, a first total difference value for the class is calculated; and the first total difference values ​​of all classes are summarized to obtain a second total difference value. Based on the second total difference value, the process parameters of the classification model are optimized to minimize the second total difference value of the classification model.

[0031] In this example, by inputting process parameters into a classification model for classification, calculating the total difference value for each category based on the classification results, and optimizing the process parameters of the classification model based on the total difference value, the accuracy of the classification model for process parameters and the overall prediction accuracy are improved.

[0032] In one example, the device further includes: The optimization module is used to input the process parameters generated by the genetic algorithm into the trained process parameter optimization model to generate a predicted performance index. The process parameters are obtained by optimizing the process parameters of the previous iteration using the genetic algorithm based on the predicted performance index and the preset target performance index. The above process is repeated until the predicted performance index matches the target performance index, and then the process parameters are output.

[0033] In this example, a genetic algorithm is used to generate process parameters, which are then input into a trained model to generate predicted performance indicators. Based on the difference between the predicted and target performance indicators, the genetic algorithm generates new process parameters, and this process is repeated until the predicted and target performance indicators match. This achieves the goal of accurately obtaining process parameters that meet performance requirements. Based on these process parameters, production line optimization can be implemented.

[0034] Thirdly, embodiments of this application provide a computer device, including: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the method described in any of the above embodiments.

[0035] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer instructions, which are used to cause a computer to perform the method described in any one of the above embodiments.

[0036] Fifthly, embodiments of this application provide a computer program product, including computer instructions, which are used to cause a computer to perform the method described in any of the above embodiments. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0038] Figure 1 A network model structure diagram provided in an embodiment of this application; Figure 2 A flowchart illustrating a semiconductor process parameter optimization model training method provided in this application embodiment; Figure 3 A flowchart illustrating a semiconductor process parameter optimization method based on a process parameter optimization model, provided as an embodiment of this application; Figure 4 A flowchart illustrating a semiconductor process parameter optimization model training method provided in this application embodiment; Figure 5 A flowchart illustrating a semiconductor process parameter optimization method based on a process parameter optimization model, provided as an embodiment of this application; Figure 6 A block diagram of a semiconductor process parameter optimization model training device provided in this application embodiment; Figure 7 This is a structural diagram of a computer device provided in an embodiment of this application. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0040] As semiconductor processes continue to evolve towards advanced nodes, the performance of Complementary Metal Oxide Semiconductor (CMOS) devices is increasingly constrained by process parameters. Specifically, in advanced processes, key process parameters such as ion implantation, diffusion annealing, and gate dielectric deposition have complex and nonlinear effects on electrical characteristics such as threshold voltage (Vth), subthreshold swing (SS), on-state current (Ion), and off-state leakage current (Ioff).

[0041] Faced with such complex parameter relationships, the traditional method of optimizing process parameters by combining simulation and experience is not only inefficient but also severely lacks adaptability to new processes and designs. To overcome this bottleneck, machine learning technology has been introduced to replace traditional physical modeling methods. However, single machine learning models often suffer from underfitting or overfitting when dealing with such complex problems, leading to unstable prediction accuracy and a lack of sufficient robustness to cope with various changes in actual processes. The application of machine learning models can be exemplified by... Figure 1 As shown.

[0042] To address this, this application proposes a method for training a semiconductor process parameter optimization model. This method aims to accurately construct a nonlinear mapping relationship between process parameters and device electrical performance in highly complex and tightly coupled CMOS manufacturing processes. Using this mapping model, it achieves performance-oriented reverse optimization of process parameters, thereby improving overall device performance and process debugging efficiency.

[0043] First, computer equipment can perform data acquisition. A set of structured sample data can be obtained from simulation platforms or actual production. This structured sample data can cover process parameters and actual performance indicators. Process parameters may include ion doping concentration, ion implantation energy, annealing time, oxide layer thickness, etc. Actual performance indicators may include threshold voltage, subthreshold swing, on-state current, leakage current, etc.

[0044] Next, the computer equipment can construct output learners. After standardizing and cleaning the collected data, the computer equipment can use this structured sample data to train multiple pre-constructed base learners with complementary modeling capabilities. This training process is used to fit the relationship between process parameters and true performance indicators in the structured sample data.

[0045] For example, the base learner can be eXtremeGradient Boosting (XGBoost), Support Vector Regression (SVR), Multilayer Perceptron (MLP), or Random Forest.

[0046] Then, the computer device can introduce an ensemble learning mechanism. Based on the aforementioned basic learners, the computer device can use strategies such as weighted averaging, stacking regression, or dynamic selection to merge multiple basic learners into a global regression model that has both high robustness and high accuracy.

[0047] The strategy employed in this application can be a dynamic selection mechanism. During model prediction, the computer device can automatically select the most suitable base learner for local prediction based on the features of the input sample or its position in the input space. This mechanism is essentially a local expert strategy, which records the error distribution of each sub-model in different regions of the training data and combines this with a sample neighborhood evaluation method to determine in real time which base learner should dominate the prediction of the current input. Compared to traditional static weighted ensemble, this mechanism can effectively improve the model's ability to fit complex local relationships and avoid the spread of prediction errors caused by global averaging.

[0048] After completing the ensemble model training, the computer equipment can enter the parameter optimization phase. Based on user-defined target performance indicators, the computer equipment can use a genetic algorithm (GA) or other black-box optimization strategies to search for the optimal combination of process parameters within the model input space. During the optimization process, the aforementioned dynamic selection mechanism is invoked at each objective function evaluation to ensure that the prediction results are always provided by locally optimal sub-models or sub-model clusters, thereby improving the stability and accuracy of the optimization path.

[0049] Using the methods described above, this invention not only enables high-precision modeling of complex process-performance mapping relationships but also supports performance-oriented reverse engineering and parameter tuning, significantly shortening the process development cycle. Compared to traditional single-model methods, this invention demonstrates significant advantages in prediction accuracy, robustness, and generalization ability.

[0050] According to an embodiment of this application, a method for training a semiconductor process parameter optimization model is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed on a computer device via a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here. The computer device can be a mobile terminal, a personal computer, a server, etc.

[0051] Figure 2 A flowchart of a semiconductor process parameter optimization model training method provided in this application embodiment is shown below. Figure 2 As shown, with a computer device as the execution subject, the process includes the following steps: S101. Collect structured sample data. Each structured sample data contains process parameters recorded during the production of semiconductor devices, as well as the actual performance indicators measured on the semiconductor devices produced based on the process parameters.

[0052] For example, the computer device first initiates a data acquisition program to collect process parameters related to semiconductor device production from the semiconductor manufacturing process. The computer device also initiates a performance testing program to test the completed semiconductor devices and obtain their actual performance indicators. The computer device can structure the process parameters and actual performance indicators of each semiconductor device to obtain structured sample data.

[0053] In one implementation, structured sample data refers to data stored according to a certain structure, which facilitates computer storage, processing, and analysis.

[0054] In one implementation, process parameters are various parameters that control the production process and product characteristics during semiconductor device manufacturing. For example, process parameters may include doping concentration, implantation energy, annealing time, etc.

[0055] In one implementation, the computer device can obtain the process parameter by directly acquiring real-time data from the production equipment's data interface. Alternatively, the computer device can obtain the process parameter by batch extracting historical data from a production database. Or, the computer device can obtain the process parameter through simulation.

[0056] In one implementation, the true performance index is performance data obtained by actual measurement of the manufactured semiconductor device. For example, true performance indexes may include threshold voltage, subthreshold swing, on-state current, leakage current, etc. Alternatively, computer equipment can perform simulations based on process parameters to obtain the true performance index.

[0057] In one implementation, computer equipment can use specialized testing instruments to test semiconductor devices and obtain their true performance indicators.

[0058] S102. Using structured sample data, train multiple basic learners. The basic learners are used to output predicted performance indicators based on process parameters.

[0059] For example, the computer device trains multiple different types of basic learners with the collected structured sample data, enabling the basic learners to learn the mapping relationship between process parameters and predicted performance indicators, thereby predicting the predicted performance indicators based on the process parameters.

[0060] In one implementation, the base learner is used to output predicted performance metrics based on process parameters. Optionally, the base learner is... In one implementation, the training set and the test set are two subsets obtained by dividing the original dataset according to a certain ratio. The training set is used to train the model, and the test set is used to evaluate the performance of the model.

[0061] In one implementation, the base learner can be a machine learning model. For example, the base learner can be multiple models such as gradient boosting trees, random forests, support vector regression, and multilayer perceptrons.

[0062] In one implementation, the multiple base learners selected in this embodiment can be complementary. For example, the multiple base learners can each have better prediction performance for different features. Or, the multiple base learners can each have better prediction performance for different types of data.

[0063] In one implementation, after acquiring a large amount of structured sample data, the computer device can divide the data into a training set and a test set. Subsequently, the computer device can use the data from the training set to train various basic learners in this step.

[0064] S103. Based on the actual performance indicators and predicted performance indicators, calculate the prediction accuracy of each basic learner, and generate a classification model based on the prediction accuracy. The classification model is used to determine the basic learner for the processing parameters.

[0065] For example, a computer device can calculate the prediction accuracy of each basic learner based on the predicted performance metrics output by each basic learner after processing process parameters, and the actual performance metrics contained in structured sample data. Furthermore, the computer device can generate a classification model for classifying the basic learners based on these prediction accuracies. This classification model is used to determine the basic learner with the highest prediction accuracy for a given process parameter, thereby enabling the selection of that basic learner.

[0066] In one implementation, prediction accuracy refers to the percentage of process parameters whose predicted performance indicators match the actual performance indicators among all process parameters predicted using the basic learner.

[0067] In one implementation, when calculating the prediction accuracy, the computer device can use common evaluation metrics such as Mean Squared Error (MSE) and Mean Absolute Error (MAE) to calculate the error between the actual performance metric and the predicted performance metric, thereby obtaining the accuracy.

[0068] In one implementation, the classification model is a machine learning model used to assign input process parameters to different base learners.

[0069] In one implementation, the classification model can use a decision tree algorithm, construct a decision tree based on the prediction accuracy, and assign samples to different leaf nodes according to the characteristics of the process parameters, with each leaf node corresponding to a basic learner.

[0070] In another implementation, the classification model can use a neural network algorithm, using prediction accuracy as an input feature to train a neural network model for classification.

[0071] In another implementation, the classification model can determine the base learner with the highest accuracy in each data range based on a preset partitioning rule, thereby determining the mapping relationship between process parameters and the base learner.

[0072] In one implementation, after collecting a large amount of structured sample data, the computer device can divide the data into a training set and a test set. Subsequently, in this step, the computer device can input the test set data into each basic learner, and then determine the prediction accuracy of each basic learner based on the prediction performance metrics output by each basic learner and the actual performance metrics in the test set.

[0073] S104. Based on the classification model and various basic learners, a process parameter optimization model is constructed. The process parameter optimization model is used to output predicted performance indicators based on the process parameters.

[0074] For example, the computer device integrates the generated classification model with the various trained basic learners to form a complete process parameter optimization model.

[0075] Based on this process parameter optimization model, when a computer device acquires a process parameter, it first determines the optimal base learner corresponding to that process parameter using a classification model. Then, the computer device can use the selected base learner to process the process parameter and output a predicted performance index.

[0076] In this embodiment, structured sample data is collected and used to train multiple basic learners. Then, a classification model is generated based on the prediction accuracy of each basic learner. The classification model and the basic learners are used to form a process parameter optimization model. This method generates a process parameter optimization model and improves the prediction accuracy of the process parameter optimization model, thereby improving the efficiency of optimizing production process parameters based on the prediction results of the process parameter optimization model.

[0077] In one example, in step S103 above, the computer device may first classify the process parameters and then determine the base learner corresponding to each classification. This process may include: S1031. Based on a preset classification algorithm for process parameters, classify the process parameters.

[0078] For example, the computer device first selects a preset classification algorithm. Based on this algorithm, the computer device can input process parameters from the collected structured sample data into the algorithm to classify the process parameters, resulting in multiple classes of process parameters. Each class can include multiple process parameters.

[0079] In one implementation, the preset classification algorithm is a pre-defined mathematical method and logical rule for classifying data. For example, the classification algorithm could be a decision tree algorithm, K-nearest neighbor algorithm, support vector machine algorithm, etc. Alternatively, the classification algorithm could also be an algorithm based on simple mapping for classification.

[0080] In one implementation, to avoid overfitting, the structured sample data used in this step can be data from the test set. This structured sample data in the test set can be separate from the structured sample data in the training set used to train the base learner in step S102. This avoids the problem of excessively high prediction accuracy caused by using the same data.

[0081] S1032. Statistically calculate the prediction accuracy of various process parameters when using each basic learner for prediction.

[0082] For example, the computer device inputs the process parameters for each class into separate base learners, which then output corresponding predicted performance metrics based on these parameters. These predicted performance metrics are then compared with the actual performance metrics to calculate the prediction accuracy for each class of process parameters under each base learner. This prediction accuracy can be calculated by methods such as the proportion of correctly predicted samples to the total number of samples in that class.

[0083] For example, when there are five basic learners, the computer device can input a set of process parameters corresponding to a class into each of the five basic learners. The computer device can then obtain the predicted performance index output by each basic learner. Based on this, the computer device can compare the predicted performance index output by each basic learner with the actual performance index in the set of process parameters to determine the prediction accuracy of the five basic learners.

[0084] For example, when there are 7 classes, the computer device can obtain the prediction accuracy of the 5 basic learners for each class based on the example above.

[0085] S1033. Based on the mapping relationship between the base learner with the highest prediction accuracy in each class and the class, generate the classification model.

[0086] For example, after the computer device completes the statistical analysis of the prediction accuracy of process parameters in each class under each basic learner, it compares the prediction accuracy of each contact learner in each class and identifies the basic learner with the highest prediction accuracy as the basic learner corresponding to that class. That is, the computer device can determine the mapping relationship between each class and the basic learner.

[0087] For example, when it is determined that the prediction accuracy of the basic learner 3 is the highest in class 1, in the subsequent prediction process, the computer device can input the process parameter into the basic learner 3 when determining the output of the process parameter class 1 to achieve the prediction of the performance index.

[0088] Furthermore, the computer device can generate the classification model based on the mapping relationship and the classification algorithm in step S1031 above.

[0089] In one implementation, the classification model is used to determine the class corresponding to the process parameter based on the classification algorithm, and to determine the base learner to process the process parameter based on the mapping relationship between the class and the base learner.

[0090] In this example, process parameters are classified using a preset classification algorithm. Based on the prediction accuracy of each base learner in each class, the mapping relationship between each class and the base learner is determined. Then, based on the preset classification algorithm and the mapping relationship between the classes, as well as the classification algorithm, a classification model is generated to classify the process parameters, thereby matching the optimal base learner and improving the prediction accuracy of the process parameters.

[0091] In one example, in step S1031 above, the computer device can classify process parameters based on numerical ranges. This process includes: Step 311: Classify the process parameters according to the preset numerical range.

[0092] For example, the computer device divides the collected process parameters according to a pre-defined series of numerical ranges, obtaining multiple process parameters contained in each numerical range. The process parameters in each numerical range are denoted as process parameters in a class.

[0093] In one implementation, the preset numerical range is a series of numerical ranges determined in advance based on factors such as the characteristics of process parameters, production needs, and past experience.

[0094] Optionally, when the process parameters include only one parameter, the computer equipment can divide the process parameter according to the numerical range of that single parameter. For example, for the temperature process parameter, the low temperature range may be set to 20℃-50℃, the medium temperature range to 51℃-80℃, and the high temperature range to 81℃-120℃.

[0095] Optionally, when the process parameters include two parameters, the computer equipment can divide the data into numerical ranges based on these two parameters. Multiple numerical ranges can be determined using a grid partitioning method based on these two parameter ranges. For example, if the first parameter includes three ranges and the second parameter includes four ranges, the computer equipment can generate 12 numerical ranges accordingly.

[0096] Optionally, when the process parameters include multiple parameters, the computer equipment can divide the data into numerical ranges for each parameter. Based on these numerical ranges, the computer equipment can divide the data into multiple numerical ranges in a multi-dimensional space. For example, if the first parameter includes 3 ranges, the second parameter includes 4 ranges, and the third parameter includes 5 ranges, the computer equipment can generate 60 numerical ranges accordingly.

[0097] In this example, by classifying process parameters according to preset numerical ranges and generating classification models, the complex and diverse process parameters are accurately divided to facilitate targeted analysis and processing.

[0098] In another example, in step S1031 above, the computer device can classify process parameters based on a clustering algorithm, including: Step 312: Cluster the process parameters to obtain at least one cluster. Each cluster contains multiple process parameters.

[0099] For example, the computer device uses a preset clustering algorithm to cluster the collected process parameters. This clustering algorithm automatically divides the process parameters into different clusters based on their degree of similarity. Each cluster contains multiple process parameters.

[0100] In one implementation, the clustering algorithm is an unsupervised learning method. Each cluster contains a set of process parameters with similar characteristics. For example, the clustering algorithm can be K-means clustering, hierarchical clustering, etc.

[0101] Step 313: Determine the cluster center of each cluster based on the process parameters contained in each cluster; the cluster center is used to generate the classification model.

[0102] For example, after obtaining the various clusters, the computer device calculates the cluster center of each cluster based on the process parameters contained in each cluster. The cluster center can be the feature vector that minimizes the sum of distances to each process parameter in that cluster.

[0103] In one implementation, the cluster center is a representative feature vector within a cluster. The dimension of this cluster center is the same as the process parameter, which is a vector composed of multiple parameters.

[0104] In one implementation, the distance can be calculated by Euclidean distance, Manhattan distance, Minkowski distance, etc., between the feature vector of the cluster center and the vector of process parameters.

[0105] In one implementation, during the subsequent determination of the classification model, the computer equipment can classify the process parameters after obtaining them based on the cluster centers of each cluster.

[0106] In this example, by first clustering the process parameters into clusters, then calculating the cluster center of each cluster, and subsequently generating a classification model based on the cluster center, the process parameters are efficiently grouped and a classification model is constructed, thus improving data processing efficiency.

[0107] In another example, in step S103 above, the computer device can classify the process parameters based on the prediction results of each basic learner. This process may include: S1034. Use each basic learner to predict all process parameters and obtain predicted performance indices. Calculate the difference between the predicted performance indices and the actual performance indices for each process parameter.

[0108] For example, the computer device first invokes the prepared basic learners to perform predictions on all process parameters. Each basic learner, based on its own algorithm logic and parameter settings, provides a prediction result for the process parameters, which is denoted as the prediction performance index.

[0109] Subsequently, the computer equipment compares these predicted performance metrics with the previously known actual performance metrics, and calculates the difference between the predicted and actual performance metrics of each process parameter based on the base learner using a specific calculation method.

[0110] In one implementation, the difference value can be used to characterize the deviation between the predicted performance index of the process parameters obtained based on the prediction of the base learner and the actual performance index.

[0111] In one implementation, the difference value can be obtained by calculating the difference between the predicted performance metric and the actual performance metric.

[0112] Since this performance metric can include multiple parameters, the difference is actually a vector composed of the differences between multiple parameters.

[0113] Therefore, the difference value can be obtained by calculating the mean value of the differences between the predicted and actual performance indicators, and then calculating the mean value of the differences in each parameter within that difference. Alternatively, the difference value can be obtained by calculating the standard deviation of the differences in each parameter within that difference. Alternatively, the difference value can be obtained by calculating the mean squared error of the differences in each parameter within that difference. Alternatively, the difference value can be obtained by calculating the mean absolute error of the differences in each parameter within that difference.

[0114] S1035. Generate training data based on process parameters and difference values, and train the preset classification algorithm to obtain a classification model. The number of classes in the classification model is determined by the number of base learners. Each class in the classification model corresponds one-to-one with a base learner.

[0115] For example, a computer device can determine the number of classes in a classification model based on the number of base learners. Each class in the classification model can correspond to one base learner.

[0116] Computer equipment can generate training data from process parameters and variance values. The process parameters can serve as samples, and the variance values ​​as labels.

[0117] The computer device can use this training data to train the classification model. Based on the discrepancies in the training data, it calculates the loss for each classification. The computer device can then optimize the classification model based on this loss to minimize it.

[0118] In one implementation, during the training process, after obtaining the classification result of each iteration, the computer device can acquire the process parameters included in each class. The computer device can determine the difference value of the process parameters for that class based on the corresponding base learner. The computer device can calculate the sum of all total difference values ​​as the loss. The computer device can then use this loss to back-optimize the model parameters of the classification model. The computer device can iterate the above optimization process until the loss converges.

[0119] In this example, the difference between the predicted performance index and the actual performance index of each basic learner for predicting all process parameters is calculated, and the classification model is trained based on this difference. This method achieves the effect of matching the basic learner with its corresponding process parameters, thereby improving the utilization rate of the basic learner and the recognition accuracy.

[0120] In one example, in step S1035 above, training data is generated based on process parameters and difference values, and a preset classification algorithm is trained to obtain a classification model, including: Step 351: Input the process parameters into the classification model for classification and determine the class corresponding to each process parameter.

[0121] For example, the computer device takes process parameters as input data and feeds them into a pre-built classification model. The classification model analyzes and judges these process parameters based on its internally learned classification rules and parameters, determining the class to which each process parameter belongs. Each class corresponds to a base learner.

[0122] In one implementation, the classification model is a mathematical model built based on a machine learning algorithm. This model can classify input process parameters. The categories assigned by the classification model can be determined based on a pre-set base learner. Each category typically corresponds to a base learner.

[0123] In one implementation, the classification model can be a decision tree model, a neural network model, etc.

[0124] Step 352: Calculate the first total difference value for each class based on the difference values ​​when using the base learner corresponding to the class to predict the process parameters contained in each class. Then, summarize the first total difference values ​​for each class to obtain the second total difference value.

[0125] For example, the computer device can obtain the process parameters for each class in step 351, and the difference values ​​when making predictions using the corresponding base learner for that class. The computer device can accumulate these difference values ​​to obtain a first total difference value for each class. Then, the computer device sums up the first total difference values ​​for all classes to obtain a second total difference value for the entire classification scenario. This second total difference value reflects the overall prediction difference under the current combination of the classification model and the base learner.

[0126] In one implementation, the first total difference value is the sum of the difference values ​​of all process parameters in each class defined by the classification model when predicted using the corresponding base learner. The first total difference value reflects the overall difference of each class of process parameters when predicted using the corresponding base learner.

[0127] In one implementation, the second total difference value is obtained by summing the first total difference values ​​for all classes. The second total difference value reflects the overall degree of difference in the predictions of all process parameters by the combination of the classification model and the base learner.

[0128] Step 353: Based on the second total difference value, optimize the process parameters of the classification model to minimize the second total difference value of the classification model.

[0129] For example, the computer device uses the second total difference value calculated in step 352 as a loss function to adjust the model parameters of the classification model. The computer device can repeat this process continuously until it finds the model parameters of the classification model that minimize the second total difference value, thereby optimizing the classification model and improving the accuracy of the combination of the classification model and the base learner in predicting process parameters.

[0130] In one implementation, the computer device can optimize the model parameters in the classification process based on the gradient descent method.

[0131] In this example, by inputting process parameters into a classification model for classification, calculating the total difference value for each category based on the classification results, and optimizing the process parameters of the classification model based on the total difference value, the accuracy of the classification model for process parameters and the overall prediction accuracy are improved.

[0132] Figure 3 A flowchart illustrating a semiconductor process parameter optimization method based on a process parameter optimization model, provided in this application embodiment, is shown below. Figure 1 and Figure 2 Based on the illustrated embodiment, Figure 2The generated process parameter optimization model can be used to predict process parameters generated by the genetic algorithm to obtain predicted performance indicators. During the iteration process, the genetic algorithm can generate process parameters based on the gap between the predicted performance indicators and the target performance indicators, so that the predicted performance indicators obtained by the process parameter optimization model based on these process parameters can approach the target performance indicators.

[0133] like Figure 3 As shown, with a computer device as the execution subject, the process includes the following steps: S201. Input the process parameters generated using the genetic algorithm into the trained process parameter optimization model to generate predicted performance indicators. The process parameters are obtained by optimizing the process parameters from the previous iteration using the genetic algorithm, based on the predicted performance indicators and preset target performance indicators.

[0134] For example, the computer device first activates the genetic algorithm module to generate a set of initial process parameters based on specific encoding rules and initial parameter settings. The computer device then inputs these generated process parameters into a pre-trained process parameter optimization model. The process parameter optimization model processes and analyzes these input process parameters based on the data patterns and mapping relationships it has learned internally, ultimately generating the corresponding predicted performance indicators.

[0135] In one implementation, the process parameters generated by the genetic algorithm are new process parameters obtained by optimizing and adjusting the process parameters obtained in the previous iteration based on the predicted performance index and the pre-set target performance index, aiming to gradually approach the optimal solution.

[0136] In one implementation, a genetic algorithm is an optimization search algorithm that simulates natural selection and genetic mechanisms. By simulating selection, crossover, and mutation processes in biological evolution, the genetic algorithm searches the solution space of a problem to find the optimal or near-optimal solution. In process parameter optimization, the genetic algorithm encodes process parameters as chromosomes and iteratively updates the chromosome population to gradually optimize the process parameters.

[0137] In one implementation, the process parameter optimization model uses... Figure 2 The model generated by the example shown.

[0138] In one implementation, the target performance index is a pre-defined, desired performance value. This target performance index is the objective of process parameter optimization. The genetic algorithm continuously adjusts the process parameters to make the predicted performance index gradually approach the target performance index.

[0139] In one implementation, the computer equipment divides the range of process parameters into several intervals, each interval corresponding to a binary code. A set of binary codes is randomly generated and then decoded into the corresponding process parameter values. In the selection operation, methods such as roulette wheel selection or tournament selection can be used to select superior individuals for the next generation based on the degree of difference between the predicted and target performance indicators. Crossover operations can employ single-point crossover, multi-point crossover, etc., while mutation operations can be implemented by randomly altering certain gene loci.

[0140] S202. Repeat the above process until the predicted performance index matches the target performance index, then output the process parameters.

[0141] For example, the computer device initiates a loop mechanism, repeatedly executing the process described in S201. In each loop, a genetic algorithm is used to generate new process parameters, which are then input into the process parameter optimization model to generate predicted performance indicators. The computer device then performs a detailed comparison and analysis of the currently obtained predicted performance indicators with pre-set target performance indicators. If a difference still exists, i.e., a mismatch, the computer device continues to use the genetic algorithm to further optimize and adjust the process parameters, and then enters the next loop. This process continues until the computer device determines that the predicted performance indicators and target performance indicators have reached a matching state. At this point, the loop stops, and the computer device outputs the finally obtained process parameters that meet the requirements.

[0142] In one implementation, matching refers to the predicted performance metric being numerically equal to or meeting a pre-defined error range with the target performance metric.

[0143] In one implementation, the computer device can determine whether the predicted performance metric matches the target performance metric by setting an error threshold. For example, an allowable error range can be set; if the difference between the predicted and target performance metrics falls within this error range, they are considered to match, and the loop stops; otherwise, the loop continues.

[0144] In another implementation, the computer device can set a maximum number of iterations. When the number of iterations reaches the maximum, the loop stops and the current optimal process parameters are output, regardless of whether the predicted performance index matches the target performance index.

[0145] In one implementation, after the loop stops, the computer device can output the process parameters obtained from the last iteration as the final result. The output can be achieved by storing the process parameters in a database for later use. Alternatively, the computer device can display the process parameters on its screen for operator viewing. Or, the computer device can transmit the process parameters to other devices or systems via a network.

[0146] In this embodiment, process parameters are generated using a genetic algorithm and input into a trained model to generate predicted performance indicators. Based on the difference between the predicted and target performance indicators, a new process parameter is generated using the genetic algorithm. This process is repeated until the predicted and target performance indicators match, thereby achieving the effect of accurately obtaining process parameters that meet performance requirements. Based on these process parameters, production line optimization can be achieved.

[0147] Figure 4 A flowchart illustrating a semiconductor process parameter optimization model training method provided in this application embodiment is shown below. Figures 1 to 3 Based on the illustrated embodiments, as Figure 4 As shown, the semiconductor process parameter optimization model provided in this application has the core objective of accurately establishing a nonlinear mapping relationship between process parameters and device electrical performance in a highly complex and strongly coupled CMOS manufacturing process, and using this mapping model to achieve target performance-oriented reverse optimization of process parameters, thereby improving the overall performance level of the device and enhancing process debugging efficiency.

[0148] Using computer devices as the execution subject, this process includes the following steps: S301, Data Collection and Preprocessing.

[0149] For example, a computer device can acquire a large amount of manufacturing simulation data or collected experimental data to obtain structured sample data of semiconductor devices. Each of these structured sample data includes process parameters recorded during the production of the semiconductor device, as well as actual performance indicators measured on the semiconductor device produced based on the process parameters.

[0150] For example, the semiconductor device can be a CMOS device.

[0151] In one implementation, the process parameters may specifically include information such as implantation energy, concentration, oxidation time, channel length, and oxide layer thickness. Optionally, the process parameters may include device structure parameters.

[0152] For example, this process parameter can be denoted as [Well Implantation Dose, LDD Implantation Dose, Halo LDD Implantation Dose, Source / Drain Implantation Dose, OxideThickness, Gate Length].

[0153] In one implementation, the actual performance metrics include multiple electrical performance metrics. For example, these actual performance metrics can be denoted as [Ion, Ioff, Vth].

[0154] In one implementation, the computer equipment can use Min-Max normalization or Z-Score normalization to process the process parameters.

[0155] S302, Construction of Multi-Based Learners.

[0156] For example, a computer device can construct multiple representative base learners. These base learners could be multiple learners such as Gradient Boosting Tree (XGBoost), Random Forest, Support Vector Regression (SVR), and Multilayer Perceptron (MLP).

[0157] Computer equipment can train each base learner independently and evaluate its prediction accuracy for different performance metrics.

[0158] S303, Dynamic Selection Mechanism.

[0159] For example, a computer device can dynamically evaluate the historical prediction performance of multiple sub-models near the input for each input sample, based on its feature distribution or local region attributes, and select the optimal sub-model for prediction.

[0160] Here, the characteristic distribution refers to the distribution of the input parameter space. For example, the value of the process parameter WellImplantationDose is concentrated in the range of 0.2-0.3. The value of the process parameter LDDImplantationDose is concentrated in the range of 0.4-0.6.

[0161] In this context, "local region" refers to the division of the aforementioned feature distribution into different local regions if the distribution is complex. The computer device can then select a different base learner for each local region to predict electrical performance.

[0162] The computer device can select the base learner with the highest prediction accuracy in a local area as the base learner used for this classification.

[0163] The process parameters contained in each local region can be considered as a class of process parameters.

[0164] The division of this local region allows for the gridding of different input parameter distributions. For example, parameter 1 can be divided into 10 intervals, parameter 2 into 10 intervals, and so on. Multiple intervals constitute more local regions. The number of local regions is the product of the number of intervals for parameter 1, the number of intervals for parameter 2, and so on.

[0165] In each local region, different electrical properties are predicted using different models. The computer can select the three models with the highest prediction accuracy based on Ion, Ioff, and Vth, respectively.

[0166] Figure 5 A flowchart illustrating a semiconductor process parameter optimization model training method provided in this application embodiment is shown below. Figures 1 to 4 Based on the illustrated embodiments, as Figure 5 As shown, with a computer device as the execution subject, the process includes the following steps: S401, Process Parameter Optimization For example, given a target electrical performance setting, a computer device can use a genetic algorithm (GA) to search for a combination of process parameters in the input space that satisfies the target. An ensemble model participates in the optimization as a black-box predictor.

[0167] Computer equipment can initialize multiple sets of process parameters using a genetic algorithm, then evaluate these parameters using a semiconductor process parameter optimization model to obtain predicted electrical performance. The genetic algorithm adjusts the process parameters based on the predicted and target electrical performance. Subsequently, the computer equipment can input the new process parameters back into the process parameter optimization model to output predicted electrical performance.

[0168] That is, computer equipment can perform iterative optimization of process parameters based on genetic algorithms.

[0169] The input process parameters can be denoted as [Well Implantation Dose, LDD Implantation Dose, Halo LDD Implantation Dose, Source / Drain Implantation Dose Oxide Thickness, Gate Length]. Each parameter can be a specific process parameter value.

[0170] Figure 6 A structural diagram of a semiconductor process parameter optimization model training device provided in this application embodiment is shown below. Figure 6As shown, the semiconductor process parameter optimization model training device 600 includes: The acquisition module 601 is used to collect structured sample data. Each structured sample data contains process parameters recorded during the production of semiconductor devices, as well as the actual performance indicators measured on the semiconductor devices produced based on the process parameters.

[0171] Training module 602 is used to train multiple base learners using structured sample data. The base learners output predicted performance indicators based on process parameters. The prediction accuracy of each base learner for process parameters is statistically analyzed, and a classification model is generated based on the prediction accuracy. The classification model processes the base learners for process parameters. Based on the classification model and the base learners, a process parameter optimization model is constructed. The process parameter optimization model outputs predicted performance indicators based on process parameters.

[0172] In one example, training module 602 is used for: Based on a pre-defined classification algorithm for process parameters, the process parameters are classified and a classification model is generated.

[0173] The prediction accuracy of various process parameters when using different basic learners is statistically analyzed.

[0174] The base learner with the highest prediction accuracy among all classes is used as the base learner for the process parameters of that class.

[0175] In one example, training module 602 is used for: Based on the preset numerical range, the process parameters are classified and a classification model is generated.

[0176] In one example, training module 602 is used for: The process parameters are clustered to obtain at least one cluster. Each cluster contains multiple process parameters.

[0177] A classification model is generated based on the cluster centers of each cluster. The cluster centers are determined based on the process parameters contained in the cluster.

[0178] In one example, training module 602 is used for: All process parameters are predicted using various basic learners to obtain predicted performance indices. The difference between the predicted and actual performance indices for each process parameter is then calculated.

[0179] Training data is generated based on process parameters and variance values. A pre-defined classification algorithm is then trained to obtain a classification model. The number of classes in the classification model is determined by the number of base learners.

[0180] In this model, each class corresponds one-to-one with a base learner.

[0181] In one example, training module 602 is used for: The process parameters are input into the classification model for classification, and the class corresponding to each process parameter is determined.

[0182] Based on the difference values ​​of the process parameters contained in each class when predicted using the base learner corresponding to that class, the first total difference value for each class is calculated. The first total difference values ​​for each class are then summed to obtain the second total difference value.

[0183] Based on the second total difference value, optimize the process parameters of the classification model to minimize the second total difference value of the classification model.

[0184] In one example, the device further includes: The optimization module 603 is used to input the process parameters generated by the genetic algorithm into the trained process parameter optimization model to generate predicted performance indicators. The process parameters are obtained by optimizing the process parameters from the previous iteration using the genetic algorithm, based on the predicted performance indicators and preset target performance indicators. This process is repeated until the predicted performance indicators match the target performance indicators, at which point the process parameters are output.

[0185] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0186] In this embodiment, the semiconductor process parameter optimization model training device is presented in the form of functional units. Here, a unit refers to an application-specific integrated circuit (ASIC), a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0187] Figure 7 A structural diagram of a computer device provided in an embodiment of this application, such as... Figure 7As shown, the computer device 700 includes one or more processors 701, memory 702, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interface). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 7 Take the 701 processor as an example.

[0188] Processor 701 may be a central processing unit, a network processor, or a combination thereof. Processor 701 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GPA), or any combination thereof.

[0189] The memory 702 stores instructions executable by at least one processor 701 to cause at least one processor 701 to perform the method shown in the above embodiments.

[0190] Memory 702 may include a program storage area and a data storage area. The program storage area may store the operating system and application programs required for at least one function. The data storage area may store data created based on the use of the computer device. Furthermore, memory 702 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, memory 702 may optionally include memory remotely located relative to processor 701, which can be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0191] The memory 702 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 702 may also include a combination of the above types of memory.

[0192] The computer device also includes a communication interface 703 for communicating with other devices or communication networks.

[0193] This application also provides a computer-readable storage medium. The methods described in this application can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded over a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor computer devices, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the methods shown in the above embodiments are implemented.

[0194] This application provides a computer program product including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method of any embodiment of this application.

[0195] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and all such modifications and variations fall within the scope defined by the appended claims.

[0196] It is understood that before using the technical solutions disclosed in the various embodiments of this disclosure, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this disclosure in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained.

[0197] For example, upon receiving a user's active request, a prompt message is sent to the user to explicitly inform them that the requested operation will require the acquisition and use of the user's personal information. This allows the user to independently choose whether to provide personal information to the software or hardware, such as the electronic device, application, server, or storage medium performing the operations of this disclosed technical solution, based on the prompt message.

[0198] As an optional but non-limiting implementation, in response to a user's active request, sending a prompt message to the user can be done via a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control allowing the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0199] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.

[0200] It is understood that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and related provisions.

[0201] It is understood that in the specific implementation of this application, data such as user information, location information, and navigation data are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0202] The apparatus, module, or unit described in the above embodiments can be implemented by a computer chip or entity, or by a product having a certain function. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0203] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0204] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0205] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0206] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0207] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0208] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0209] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0210] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

[0211] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for training a semiconductor process parameter optimization model, characterized in that, The method includes: Collect structured sample data, each of which contains process parameters recorded during the production of semiconductor devices, as well as actual performance indicators measured on semiconductor devices produced based on the process parameters. Using the structured sample data, multiple basic learners are trained respectively; the basic learners are used to output predicted performance indicators based on process parameters. Based on the actual performance metrics and the predicted performance metrics, the prediction accuracy of each of the base learners is calculated, and a classification model is generated based on the prediction accuracy; the classification model is used to determine the base learners for processing parameters. Based on the classification model and each of the basic learners, a process parameter optimization model is formed; the process parameter optimization model is used to output predicted performance indicators according to the process parameters. The process includes calculating the prediction accuracy of each of the basic learners for the process parameters and generating a classification model based on the prediction accuracy, including: The predicted performance index is obtained by using each of the basic learners to predict all the process parameters; and the difference between the predicted performance index and the actual performance index of each process parameter is calculated. Training data is generated based on the process parameters and the difference values, and a preset classification algorithm is trained to obtain a classification model; the number of categories in the classification model is determined based on the number of base learners. The classes in the classification model correspond one-to-one with the base learners; The process involves generating training data based on the process parameters and the difference values, training a preset classification algorithm to obtain a classification model, including: The process parameters are input into the classification model for classification, and the class corresponding to each process parameter is determined. Based on the difference value of the process parameters contained in each class when predicted using the base learner corresponding to the class, a first total difference value for the class is calculated; and the first total difference values ​​of all classes are summarized to obtain a second total difference value. Based on the second total difference value, the process parameters of the classification model are optimized to minimize the second total difference value of the classification model.

2. The method according to claim 1, characterized in that, Based on the actual performance metrics and the predicted performance metrics, the prediction accuracy of each of the basic learners is calculated, and a classification model is generated based on the prediction accuracy, including: The process parameters are classified based on a preset classification algorithm. The prediction accuracy of each of the aforementioned process parameters when using the respective basic learners is statistically analyzed. The classification model is generated based on the mapping relationship between the base learner with the highest prediction accuracy in each class and the class.

3. The method according to claim 2, characterized in that, Based on a preset classification algorithm for the process parameters, the process parameters are classified, including: The process parameters are classified according to a preset numerical range.

4. The method according to claim 2, characterized in that, Based on a preset classification algorithm for the process parameters, the process parameters are classified to generate a classification model, including: The process parameters are clustered to obtain at least one cluster; each cluster contains multiple process parameters. Based on the process parameters contained in each cluster, the cluster center of each cluster is determined; the cluster center is used to generate the classification model.

5. The method according to any one of claims 1-4, characterized in that, The method further includes: The process parameters generated using a genetic algorithm are input into a trained process parameter optimization model to generate a predicted performance index; wherein, the process parameters are obtained by optimizing the process parameters of the previous iteration using a genetic algorithm based on the predicted performance index and a preset target performance index; Repeat the above process until the predicted performance index matches the target performance index, then output the process parameters.

6. A semiconductor process parameter optimization model training device, characterized in that, The device includes: The acquisition module is used to collect structured sample data. Each piece of structured sample data contains process parameters recorded during the production of semiconductor devices, as well as actual performance indicators measured on semiconductor devices produced based on the process parameters. The training module is used to train multiple basic learners using the structured sample data; the basic learners are used to output predicted performance indicators based on process parameters; the prediction accuracy of each basic learner for the process parameters is calculated, and a classification model is generated based on the prediction accuracy; the classification model is used to process the basic learners for the process parameters; based on the classification model and each basic learner, a process parameter optimization model is formed; the process parameter optimization model is used to output predicted performance indicators based on the process parameters. The process includes calculating the prediction accuracy of each of the basic learners for the process parameters and generating a classification model based on the prediction accuracy, including: The predicted performance index is obtained by using each of the basic learners to predict all the process parameters; and the difference between the predicted performance index and the actual performance index of each process parameter is calculated. Training data is generated based on the process parameters and the difference values, and a preset classification algorithm is trained to obtain a classification model; the number of categories in the classification model is determined based on the number of base learners. The classes in the classification model correspond one-to-one with the base learners; The process involves generating training data based on the process parameters and the difference values, training a preset classification algorithm to obtain a classification model, including: The process parameters are input into the classification model for classification, and the class corresponding to each process parameter is determined. Based on the difference value of the process parameters contained in each class when predicted using the base learner corresponding to the class, a first total difference value for the class is calculated; and the first total difference values ​​of all classes are summarized to obtain a second total difference value. Based on the second total difference value, the process parameters of the classification model are optimized to minimize the second total difference value of the classification model.

7. A computer device, characterized in that, include: A memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, the processor executing the computer instructions to perform the method of any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to perform the method of any one of claims 1 to 5.

9. A computer program product, characterized in that, Includes computer instructions for causing a computer to perform the method of any one of claims 1 to 5.

Images