Chip optimization method and device of intelligent chip factory, controller and medium
By filtering and model training on the training data multiple times in the smart chip factory, the problem of not saving filtered data in the training data is solved, the generalization ability and prediction accuracy of the model are improved, and the chip quality is improved.
Patent Information
- Application Number
- CN202510278381.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-10
- Publication Date
- 2025-06-27
AI Technical Summary
The filtered data is not saved in the training data, resulting in a decrease in the data volume or a sampling deviation is introduced, making the training data unable to accurately reflect the true distribution of the statistical parent, which in turn affects the generalization ability and prediction accuracy of the model, resulting in a decrease in chip quality.
A chip optimization method for smart chip factory is proposed, by acquiring original data and correcting data, determining the first data, and filtering it multiple times to determine a plurality of second data. Based on each second data, the preset model is trained and the model with the highest prediction accuracy is selected for deployment to reduce sampling bias and outlier data interference of the training data.
Through multiple filtering and model training, the generalization ability and prediction accuracy of the preset model are improved, thereby improving chip quality, reducing sampling deviations and outlier data interference of training data.
Smart Images
Figure CN120218005A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of chip technology, and in particular, to a chip optimization method for an intelligent chip factory, a chip optimization device for an intelligent chip factory, a computer-readable storage medium, and a controller. Background Art
[0002] In the related art, the training data used to train a model usually does not save the filtered data, resulting in a reduction in the amount of data or the introduction of sampling bias, such that the training data cannot accurately reflect the true distribution of the statistical population. Due to the lack of representativeness of the training data, the generalization ability and prediction accuracy of the model are significantly affected, thereby leading to a decline in the quality of the chip. Summary of the Invention
[0003] The present application aims to at least solve one of the technical problems in the related art to some extent. For this purpose, the first object of the present application is to provide a chip optimization method for an intelligent chip factory, the method including: obtaining original data and calibration data; determining first data according to the original data and the calibration data; performing multiple data filtrations on the first data to determine a plurality of second data; training a preset model according to each second data to determine a preset model corresponding to the second data; using the process tool operation parameters and the target product parameters as inputs of the plurality of preset models respectively to output corresponding predicted product parameters; and determining a target preset model according to the plurality of predicted product parameters and the target product parameters. In the chip optimization method of the present application, by performing multiple filtrations on the first data, second data corresponding to different filtration ranges are obtained, and based on each second data, the preset model is trained respectively, so as to determine the model corresponding to each second data, and finally, the model with the highest prediction accuracy is selected for deployment. In this way, through multiple filtrations, the training data can be avoided from being interfered by outlier data to a certain extent, and more non-outlier data can be retained as much as possible, reducing the sampling bias of the training data, improving the generalization ability and prediction accuracy of the preset model, and further improving the chip quality.
[0004] The second object of the present application is to provide a chip optimization device for an intelligent chip factory.
[0005] The third object of the present application is to provide a computer-readable storage medium.
[0006] The fourth object of the present application is to provide a controller.
[0007] To achieve the above object, an embodiment of the first aspect of the present application provides a chip optimization method for an intelligent chip factory, the method comprising: obtaining original data and calibration data; determining first data according to the original data and the calibration data; performing multiple data filtering operations on the first data to determine a plurality of second data; training a preset model according to each second data to determine a preset model corresponding to the second data; using the process tool operation parameters and the target product parameters as inputs to a plurality of preset models respectively to output corresponding predicted product parameters; determining a target preset model according to the plurality of predicted product parameters and the target product parameters.
[0008] According to an embodiment of the present application, determining a target preset model according to the plurality of predicted product parameters and the target product parameters includes: determining the prediction accuracy of each predicted product parameter based on the difference between each predicted product parameter and the target product parameter; determining the target preset model according to the prediction accuracies of the plurality of predicted product parameters.
[0009] According to an embodiment of the present application, determining a target preset model according to the prediction accuracies of the plurality of predicted product parameters includes: using the preset model corresponding to the maximum prediction accuracy as the target preset model.
[0010] According to an embodiment of the present application, using the process tool operation parameters and the target product parameters as inputs to a plurality of preset models respectively to output corresponding predicted product parameters includes: inputting the process tool operation parameters into the plurality of preset models respectively, so that each preset model outputs the corresponding relationship between the process tool operation parameters and the product parameters; determining the corresponding target process tool operation parameters according to the target product parameters and the plurality of corresponding relationships; controlling the process tool based on the plurality of target process tool operation parameters respectively to determine the corresponding predicted product parameters.
[0011] According to an embodiment of the present application, performing multiple data filtering operations on the first data to determine a plurality of second data includes: determining a plurality of data filtering intervals, the range sizes of the plurality of data filtering intervals being different; performing data filtering processing on the first data based on each data filtering interval to determine a plurality of second data.
[0012] According to an embodiment of the present application, the calibration data includes metrology tool calibration data, metrology tool sensor calibration data, process tool sensor calibration data, and laser exposure alignment offset data. The original data includes online monitoring data, offline monitoring data, metrology tool sensor data, process tool sensor data, and scanned map digitized data. Determining the first data based on the original data and the calibration data includes: calibrating and determining the third data for the online monitoring data and the offline monitoring data according to the metrology tool calibration data and / or the laser exposure alignment offset data; calibrating and determining the fourth data for the metrology tool sensor data and the process tool sensor data according to the metrology tool sensor calibration data and the process tool sensor calibration data; calibrating and determining the fifth data for the scanned map digitized data according to the laser exposure alignment offset data; and determining the first data according to the combination of the third data, the fourth data, and the fifth data.
[0013] According to an embodiment of the present application, the above method further includes: obtaining the performance change trends of the metrology tool, the process tool sensor, and the metrology tool sensor; calibrating and determining the sixth data for the third data according to the performance change trend of the metrology tool; calibrating and determining the seventh data for the fourth data according to the performance change trends of the process tool sensor and the metrology tool sensor; and determining the first data according to the combination of the sixth data, the seventh data, and the fifth data.
[0014] To achieve the above object, an embodiment of the second aspect of the present application provides a chip optimization device for an intelligent chip factory. The device includes: an acquisition module for acquiring original data and calibration data; a first determination module for determining the first data according to the original data and the calibration data; a second determination module for performing multiple data filtrations on the first data to determine multiple second data; a third determination module for training a preset model according to each second data to determine a preset model corresponding to the second data; a fourth determination module for using the process tool operation parameters and the target product parameters as inputs of multiple preset models respectively to output corresponding predicted product parameters; and a fifth determination module for determining a target preset model according to the multiple predicted product parameters and the target product parameters.
[0015] To achieve the above object, an embodiment of the third aspect of the present application provides a computer-readable storage medium, on which a chip optimization program for an intelligent chip factory is stored. When the chip optimization program for the intelligent chip factory is executed by a processor, the control method for the intelligent chip factory described above is implemented.
[0016] To achieve the above object, an embodiment of the fourth aspect of the present application provides a controller, including a memory, a processor, and a chip optimization program of an intelligent chip factory stored in the memory and executable on the processor. When the processor executes the chip optimization program of the intelligent chip factory, the control method of the aforementioned intelligent chip factory is implemented.
[0017] According to the chip optimization method, device, controller, and medium of the intelligent chip factory according to the embodiments of the present application, original data and calibration data are obtained; first data is determined based on the original data and the calibration data; the first data is filtered multiple times to determine multiple second data; each second data is used to train a preset model to determine a preset model corresponding to the second data; the process tool operation parameters and the target product parameters are respectively used as inputs of the multiple preset models to output corresponding predicted product parameters; a target preset model is determined based on the multiple predicted product parameters and the target product parameters. In the chip optimization method of the present application, the first data is filtered multiple times to obtain second data corresponding to different filtering ranges, and each second data is used to train the preset model respectively, so as to determine the model corresponding to each second data, and finally the model with the highest prediction accuracy is selected for deployment. In this way, through multiple filtering, the training data can be avoided from being interfered by outlier data to a certain extent, and more non-outlier data can be retained as much as possible, reducing the sampling bias of the training data, improving the generalization ability and prediction accuracy of the preset model, and thus improving the chip quality. Description of the Drawings
[0018] Figure 1 It is a flowchart of the chip optimization method of the intelligent chip factory according to some embodiments of the present application;
[0019] Figure 2 It is a flowchart of the chip optimization method of the intelligent chip factory according to other embodiments of the present application;
[0020] Figure 3 It is a schematic block diagram of the chip optimization device of the intelligent chip factory according to some embodiments of the present application;
[0021] Figure 4 It is a schematic block diagram of the controller according to some embodiments of the present application. Detailed Embodiments
[0022] The embodiments of the present application will be described in detail below. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals denote the same or similar elements or elements with the same or similar functions from beginning to end. The embodiments described below with reference to the drawings are exemplary and are intended to explain the present application, but should not be construed as a limitation of the present application.
[0023] The chip optimization method, device, controller, and medium of the intelligent chip factory according to the embodiments of the present application will be described in detail below with reference to the accompanying drawings.
[0024] Figure 1 It is a flowchart of the chip optimization method of the intelligent chip factory according to some embodiments of the present application. Referring to Figure 1 , the chip optimization method of the intelligent chip factory according to the embodiments of the present application may include the following steps:
[0025] S110, obtain the original data and calibration data.
[0026] Specifically, the original data includes online monitoring data, offline monitoring data, measurement tool sensor data, process tool sensor data, and scanned map digitized data. Among them, the online monitoring data refers to data such as the thickness and uniformity of the insulating oxide film of the product wafer collected by the measurement tool, the offline monitoring data refers to data such as the thickness and uniformity of the insulating oxide film of the control wafer collected by the measurement tool, the measurement tool sensor data refers to data such as pressure, temperature, light intensity, and current collected by the sensors built in the measurement tool, the process tool sensor data refers to data such as chamber temperature, gas flow rate, vibration amplitude, and frequency collected by the sensors built in the process tool, and the scanned map digitized data refers to data such as the defect positions of the wafer.
[0027] The calibration data includes measurement tool calibration data, measurement tool sensor calibration data, process tool sensor calibration data, and laser exposure alignment offset data. Among them, the measurement tool calibration data can be determined by comparing the measurement data obtained by measuring the standard wafer with the standard data of the standard wafer using the measurement tool. For example, multiple oxide layer thickness data are obtained by measuring the oxide layer thickness of the standard wafer multiple times using the measurement tool, the average value of the multiple oxide layer thickness data is taken, and the difference between this average value and the standard oxide layer thickness of the standard wafer is calculated. The obtained difference is the measurement tool calibration data. Similarly, the measurement tool sensor calibration data and the process tool sensor calibration data can also be determined by the above method, and no specific limitation is made here. The laser exposure alignment offset data includes laser head offset data and crosshair offset data. The laser head offset data can be determined by comparing the current laser head position with the standard laser head position, and the crosshair offset data can be determined by comparing the current crosshair position with the standard crosshair position.
[0028] S120, determine the first data according to the original data and the calibration data.
[0029] Specifically, measurement tool calibration data can be used to correct on-line monitoring data and off-line monitoring data. Since the acquisition of some on-line monitoring data and off-line monitoring data is also related to the positions of the laser head and the crosshair, laser head offset data also needs to be used to correct on-line monitoring data and off-line monitoring data. Measurement tool sensor calibration data and process tool sensor calibration data can be used to correct measurement tool sensor data and process tool sensor data. Laser head offset data can be used to correct scanned map digitized data. The set of the corrected on-line monitoring data, off-line monitoring data, measurement tool sensor data, process tool sensor data, and scanned map digitized data is the first data.
[0030] S130. Perform multiple data filtrations on the first data to determine multiple second data.
[0031] Specifically, the first data is used to train a preset model. However, to ensure the accuracy of the preset model's calculation to a certain extent, the first data needs to be filtered to obtain the second data. For example, statistical methods or artificial intelligence calculation methods can be used to filter the first data so that the second data is not interfered by outlier data. However, if the first data is only filtered once, non-outlier data may be filtered out, affecting the accuracy of the trained preset model. Therefore, multiple data filtering ranges need to be set, and the sizes of the data filtering ranges increase gradually. The first data is filtered using multiple data filtering ranges respectively to determine the second data corresponding to each data filtering range, trying to retain more non-outlier data, thereby reducing the sampling bias of the training data.
[0032] S140. Train the preset model according to each second data to determine the preset model corresponding to the second data.
[0033] Specifically, each second data is used to train the preset model respectively to obtain the preset model corresponding to the second data. Among them, the preset model can be a preset virtual measurement model. For example, preprocess the second data, such as performing feature extraction, standardization, normalization, and encoding classification, etc.; then, divide the preprocessed second data, such as dividing the preprocessed second data into a training set, a validation set, and a test set (such as 70% training set, 15% validation set, 15% test set); then, select an optimization model, such as selecting a convolutional neural network, a recurrent neural network, and a Transformer, etc. Input the training data into the optimization model, calculate the predicted value and the loss, calculate the gradient through backpropagation, and update the parameters of the optimization model until the model converges or reaches the maximum number of iterations to obtain the preset model.
[0034] S150. Use the process tool operation parameters and the target product parameters as inputs to multiple preset models respectively, so as to output corresponding predicted product parameters.
[0035] Specifically, input the process tool operation parameters into multiple preset models respectively. The preset models will output the corresponding relationships between the process tool operation parameters and the product parameters. Based on the multiple corresponding relationships between the process tool operation parameters and the product parameters, the process tool operation parameters corresponding to the generation of the target product parameters by the process tool can be determined. Control the process tool to operate according to the obtained process tool operation parameters. After completion, a measurement tool can be used to measure the corresponding predicted product parameters.
[0036] S160. Determine the target preset model according to multiple predicted product parameters and the target product parameters.
[0037] Specifically, the difference between each predicted product parameter and the target product parameter can be calculated respectively. The prediction accuracy of the preset model can be determined according to the obtained difference value. The target preset model can be determined according to the prediction accuracy of the preset model. For example, the preset model with the highest prediction accuracy can be used as the target preset model.
[0038] In the chip optimization method of this application, the first data is filtered multiple times to obtain the second data corresponding to different filtering ranges. Each preset model is trained based on each second data respectively, so as to determine the model corresponding to each second data. Finally, the model with the highest prediction accuracy is selected for deployment. In this way, through multiple filtering, the training data can be avoided from being interfered by outlier data to a certain extent, and more non-outlier data can be retained as much as possible, reducing the sampling deviation of the training data, improving the generalization ability and prediction accuracy of the preset model, and thus improving the chip quality.
[0039] In some embodiments, determining the target preset model according to multiple predicted product parameters and the target product parameters includes: determining the prediction accuracy of each predicted product parameter based on the difference between each predicted product parameter and the target product parameter; determining the target preset model according to the prediction accuracies of multiple predicted product parameters.
[0040] In some embodiments, determining the target preset model according to the prediction accuracies of multiple predicted product parameters includes: using the preset model corresponding to the maximum prediction accuracy as the target preset model.
[0041] Specifically, the absolute value of the difference between each predicted product parameter and the target product parameter is obtained respectively. The absolute value of the obtained difference is used as the prediction accuracy of each predicted product parameter. For example, if the absolute value of the difference is smaller, it is determined that the prediction accuracy of the predicted product parameter is higher; if the absolute value of the difference is larger, it is determined that the prediction accuracy of the predicted product parameter is lower. The preset model corresponding to the maximum prediction accuracy is the target preset model.
[0042] Exemplarily, assume that multiple preset models include Preset Model 1, Preset Model 2, and Preset Model 3. The predicted product parameter 1 output by Preset Model 1 is 21 nm, the predicted product parameter 2 output by Preset Model 2 is 18 nm, and the predicted product parameter 3 output by Preset Model 3 is 22 nm. The target product parameter is 20 nm. The absolute value of the difference between the predicted product parameter 1 and the target product parameter is 1, the absolute value of the difference between the predicted product parameter 2 and the target product parameter is 2, and the absolute value of the difference between the predicted product parameter 3 and the target product parameter is 2. Among them, the absolute value of the difference between the predicted product parameter 1 and the target product parameter is the smallest, that is to say, the prediction accuracy of the predicted product parameter 1 is the highest, so Preset Model 1 is determined as the target preset model.
[0043] In this way, multiple preset models trained based on multiple second data exist independently. They are connected through data streams with relevant parameter indicators, increasing the tensor combination and retaining more opportunities for optimizing the preset models, further improving the prediction accuracy of the preset models.
[0044] In some embodiments, the process tool operation parameters and the target product parameters are respectively used as the inputs of multiple preset models to output corresponding predicted product parameters, including: inputting the process tool operation parameters into multiple preset models respectively, so that each preset model outputs the corresponding relationship between the process tool operation parameters and the product parameters; determining the corresponding target process tool operation parameters according to the target product parameters and multiple corresponding relationships; controlling the process tool based on multiple target process tool operation parameters respectively to determine the corresponding predicted product parameters.
[0045] Exemplarily, taking the copper plating tool as an example of the process tool for illustration, but not as a limitation to this application. Among them, the copper plating tool operation parameters may include the deposition rate and the deposition time, and the corresponding product parameter may be the deposition thickness. Input the current deposition rate and deposition time of the copper plating tool into Preset Model 1, Preset Model 2, and Preset Model 3 respectively. The three preset models will output Deposition Relationship 1, Deposition Relationship 2, and Deposition Relationship 3, where the deposition relationship is used to characterize the corresponding relationship between the deposition rate, the deposition time, and the deposition thickness. Based on Deposition Relationship 1, Deposition Relationship 2, and Deposition Relationship 3, the target process tool operation parameters required for the copper plating tool to produce the target deposition thickness are determined as Deposition Rate 1 and Deposition Time 1, Deposition Rate 2 and Deposition Time 2, and Deposition Rate 3 and Deposition Time 3 respectively. Control the copper plating tool to perform copper layer deposition according to Deposition Rate 1 and Deposition Time 1, Deposition Rate 2 and Deposition Time 2, and Deposition Rate 3 and Deposition Time 3 respectively. After the deposition is completed, use a measurement tool to measure and obtain Predicted Deposition Thickness 1, Predicted Deposition Thickness 2, and Predicted Deposition Thickness 3.
[0046] In some embodiments, the first data is filtered multiple times to determine multiple second data, including: determining multiple data filtering intervals with different range sizes; and performing data filtering processing on the first data based on each data filtering interval to determine multiple second data.
[0047] Exemplarily, taking the example of filtering the first data 3 times to determine 3 second data for illustration, but not as a limitation to this application. Assume that the multiple data filtering intervals include data filtering interval 1, data filtering interval 2, and data filtering interval 3, and the range sizes of the three data filtering intervals are different. For example, data filtering interval 1 is [-3, 3], data filtering interval 2 is [-4, 4], and data filtering interval 3 is [-5, 5]. Filtering the first data based on data filtering interval 1 [-3, 3] can determine second data 1, that is, taking the first data within plus or minus 3 standard deviations as second data 1. Filtering the first data based on data filtering interval 2 [-4, 4] can determine second data 2, that is, taking the first data within plus or minus 4 standard deviations as second data 2. Filtering the first data based on data filtering interval 3 [-5, 5] can determine second data 3, that is, taking the first data within plus or minus 5 standard deviations as second data 3. Thus, after filtering the first data 3 times, second data 1, second data 2, and second data 3 are obtained, and the preset models 1, 2, and 3 can be determined by using second data 1, second data 2, and second data 3 to train the preset model respectively.
[0048] It should be noted that the multiple data filtering intervals can be in an inclusion relationship or not. For example, in the case where the multiple data filtering intervals are in an inclusion relationship, data filtering interval 1 can be [-3, 3] and data filtering interval 2 can be [-4, 4]. In the case where the multiple data filtering intervals are not in an inclusion relationship, data filtering interval 1 can be [-3, 3] and data filtering interval 2 can be [-2, 5].
[0049] In some embodiments, the calibration data includes metrology tool calibration data, metrology tool sensor calibration data, process tool sensor calibration data, and laser exposure alignment offset data, and the raw data includes in-line monitoring data, off-line monitoring data, metrology tool sensor data, process tool sensor data, and scanned map digitized data. Determining the first data based on the raw data and the calibration data includes: calibrating the in-line monitoring data and the off-line monitoring data according to the metrology tool calibration data and / or the laser exposure alignment offset data to determine the third data; calibrating the metrology tool sensor data and the process tool sensor data according to the metrology tool sensor calibration data and the process tool sensor calibration data to determine the fourth data; calibrating the scanned map digitized data according to the laser exposure alignment offset data to determine the fifth data; and determining the first data based on the union of the third data, the fourth data, and the fifth data.
[0050] Specifically, the in-line monitoring data and the off-line monitoring data are acquired by a metrology tool. Some of the in-line monitoring data and the off-line monitoring data are also related to the positions of the laser head and the crosshair. Therefore, it is necessary to use the metrology tool calibration data, the laser exposure alignment offset data, or both the metrology tool calibration data and the laser exposure alignment offset data to correct the in-line monitoring data and the off-line monitoring data to obtain the third data. For example, assume that the in-line monitoring data and the off-line monitoring data measured by the metrology tool include copper deposition layer thickness data, and the copper deposition layer thickness data is 22 mm. The calibration data of the metrology tool is 2 mm, which means that the metrology tool measured 2 mm more during measurement. Then, when using the calibration data of the metrology tool to correct the copper deposition layer thickness data, the copper deposition layer thickness data needs to be modified from 22 mm to 20 mm. Another example, assume that the in-line monitoring data and the off-line monitoring data include the position of the exposure pattern, and the position coordinates of the exposure pattern are (2, 3). The laser exposure alignment offset data includes the position deviation of the laser head (0, 1), which means that the ordinate of the position of the exposure pattern is shifted 1 nm to the right. Then, when using the laser exposure alignment offset data to correct the position of the exposure pattern, the position coordinates (2, 3) of the exposure pattern need to be corrected to (2, 2).
[0051] Both the measurement tool sensor data and the process tool sensor data are collected by sensors. Therefore, it is necessary to use the measurement tool sensor calibration data and the process tool sensor calibration data to correct the measurement tool sensor data and the process tool sensor data to determine the fourth data. For example, assume that the process tool sensor data includes the chamber temperature, and the chamber temperature is 45 °C, and the process tool sensor calibration data is -1 °C, which means that the process tool sensor measures 1 °C less during detection. Therefore, it is necessary to correct the chamber temperature from 45 °C to 46 °C. Another example, assume that the measurement tool sensor data includes the light intensity, and the light intensity is 10 cd, and the measurement tool sensor calibration data is 1 cd, which means that the measurement tool sensor measures 1 cd more during detection. Therefore, it is necessary to correct the light intensity from 10 cd to 9 cd.
[0052] Since the acquisition of the scanned map digital data, such as the collection of the defect positions on the wafer, relies on the positions of the laser head and the crosshair, it is also necessary to use the laser head offset data to correct the scanned map digital data. For example, assume that the defect position on the wafer is (3, 4), and the laser exposure alignment offset data includes the position deviation of the laser head (0, -1), which means that the ordinate of the position of the exposure pattern is shifted 1 nm to the left. Then, when using the laser exposure alignment offset data to correct the defect position, it is necessary to correct the defect position coordinates (3, 4) to (3, 5).
[0053] In summary, the collection of the third data, the fourth data, and the fifth data is the first data.
[0054] In this way, by using the correction data to correct the original data to obtain the first data, the data accuracy of the first data can be improved, and further the prediction accuracy of the preset model can be improved.
[0055] In some embodiments, the above method further includes: obtaining the performance change trend of the measurement tool, the performance change trend of the process tool sensor, and the performance change trend of the measurement tool sensor; correcting the third data according to the performance change trend of the measurement tool to determine the sixth data; correcting the fourth data according to the performance change trends of the process tool sensor and the measurement tool sensor to determine the seventh data; determining the first data according to the collection of the sixth data, the seventh data, and the fifth data.
[0056] Specifically, there is a phenomenon of device aging during the use of the measurement tool, which causes the offline monitoring data and the online monitoring data obtained by the measurement tool to gradually deviate from the actual values. Therefore, it is necessary to obtain the performance change trend of the measurement tool, and further correct the third data according to the performance change trend of the measurement tool. For example, determine the correction factor according to the performance change trend of the measurement tool, and add and apply the corresponding correction factor to the third data, and use the polynomial regression model for correction to obtain the sixth data.
[0057] Among them, the actual measurement results obtained by the measurement tool can be preprocessed through the MMS (Measurement Management System) (such as denoising, standardization, interpolation, etc.); then, time series analysis methods (such as moving average, exponential smoothing, ARIMA, etc.) are used to analyze the change trend of the actual measurement results over time; finally, regression analysis is used to identify the key factors affecting the performance of the measurement tool and determine the performance change trend of the measurement tool.
[0058] During the use of process tool sensors and measurement tool sensors, there is also a phenomenon of device aging. Therefore, the process tool sensor data and measurement tool sensor data obtained by sensor detection will gradually deviate from the actual values. Therefore, it is necessary to obtain the performance change trends of the process tool sensors and measurement tool sensors, and further correct the fourth data according to the performance change trends of the process tool sensors and measurement tool sensors to obtain the seventh data. It should be noted that the method for determining the performance change trend of the sensor and the specific process of correcting the sensor data according to the performance change trend are not elaborated here.
[0059] In summary, the set of the sixth data, the seventh data, and the fifth data is the first data.
[0060] In this way, by correcting the third data and the fourth data according to the performance change trend of the measurement tool, the performance change trend of the process tool sensor, and the performance change trend of the measurement tool sensor, the data accuracy of the first data can be further improved, and thus the prediction accuracy of the preset model can be improved.
[0061] As a specific example, referring to Figure 2 , the chip optimization method of the intelligent chip factory in the embodiment of the present application may further include the following steps:
[0062] S201, obtain the original database and the correction database.
[0063] S202, use the correction data to correct the original data to obtain the first data (corrected database).
[0064] S203, perform data filtering on the first data. If the data filtering condition is satisfied, execute S205; otherwise, execute S204.
[0065] S204, the database to be filtered out.
[0066] S205, the second data 1 (filtered and corrected database).
[0067] S206, the second data 2 (hierarchically corrected original database).
[0068] In S207, use the second data 1 to train and obtain the preset model 1.
[0069] In S208, use the second data 2 to train and obtain the preset model 2.
[0070] In S209, perform AI processing and data storage.
[0071] Use the preset model 1 to predict the product parameter 1 and store it.
[0072] In S210, perform parallel AI processing and data storage.
[0073] Use the preset model 2 to predict the product parameter 2 and store it.
[0074] In S211, compare the prediction accuracy and precision.
[0075] Compare the prediction accuracy and precision of the predicted product parameter 1 and product parameter 2, and use the preset model with higher prediction accuracy and precision as the target preset model.
[0076] In S212, retrain the target preset model.
[0077] In S213, update the AI processing and data storage.
[0078] In summary, the original data, calibration data, first data, and multiple second data obtained by performing multiple data filtrations on the first data are all independently saved or marked, which can, to a certain extent, avoid data chaos and thus reduce the trouble of retraining the preset model; by using the calibration data to calibrate the original data to obtain the first data, the data precision of the first data can be improved, and by training the preset model based on the first data, the prediction accuracy of the preset model can be improved; by filtering the first data multiple times, it is possible to, to a certain extent, avoid the interference of outlier data on the training data, and also retain as many non-outlier data as possible, reduce the sampling bias of the training data, improve the generalization ability and prediction accuracy of the preset model, and thus improve the chip quality; and multiple preset models obtained by training based on multiple second data exist independently, are connected through data streams with relevant parameter indicators, increase the tensor combination, retain more opportunities for optimizing the preset model, and further improve the prediction accuracy of the preset model.
[0079] Corresponding to the above embodiments, the present application also proposes a chip optimization device for an intelligent chip factory.
[0080] Refer to Figure 3 , the chip optimization device 300 for an intelligent chip factory includes: an acquisition module 310, a first determination module 320, a second determination module 330, a third determination module 340, a fourth determination module 350, and a fifth determination module 360.
[0081] Among them, the acquisition module 310 is used to acquire the original data and the calibration data; the first determination module 320 is used to determine the first data according to the original data and the calibration data. The second determination module 330 is used to perform multiple data filtrations on the first data to determine multiple second data. The third determination module 340 is used to train a preset model according to each second data to determine a preset model corresponding to the second data. The fourth determination module 350 is used to use the process tool operation parameters and the target product parameters as the inputs of multiple preset models respectively to output corresponding predicted product parameters. The fifth determination module 360 is used to determine a target preset model according to multiple predicted product parameters and the target product parameters.
[0082] According to an embodiment of the present application, the fifth determination module 360 is specifically configured to determine the prediction accuracy of each predicted product parameter based on the difference between each predicted product parameter and the target product parameter; determine the target preset model according to the prediction accuracies of multiple predicted product parameters.
[0083] According to an embodiment of the present application, the fifth determination module 360 is further configured to use the preset model corresponding to the maximum prediction accuracy as the target preset model.
[0084] According to an embodiment of the present application, the fourth determination module 350 is specifically configured to input the process tool operation parameters into multiple preset models respectively, so that each preset model outputs the corresponding relationship between the process tool operation parameters and the product parameters; determine the corresponding target process tool operation parameters according to the target product parameters and multiple corresponding relationships; control the process tool based on multiple target process tool operation parameters respectively to determine the corresponding predicted product parameters.
[0085] According to an embodiment of the present application, the second determination module 330 is specifically configured to determine multiple data filtering intervals, and the range sizes of the multiple data filtering intervals are different; perform data filtering processing on the first data based on each data filtering interval to determine multiple second data.
[0086] According to an embodiment of the present application, the calibration data includes metrology tool calibration data, metrology tool sensor calibration data, process tool sensor calibration data, and laser exposure alignment offset data. The original data includes online monitoring data, offline monitoring data, metrology tool sensor data, process tool sensor data, and scanned map digitized data. The first determination module 320 is specifically configured to: correct and determine the third data for the online monitoring data and the offline monitoring data according to the metrology tool calibration data and / or the laser exposure alignment offset data; correct and determine the fourth data for the metrology tool sensor data and the process tool sensor data according to the metrology tool sensor calibration data and the process tool sensor calibration data; correct and determine the fifth data for the scanned map digitized data according to the laser exposure alignment offset data; and determine the first data according to the combination of the third data, the fourth data, and the fifth data.
[0087] According to an embodiment of the present application, obtain the performance change trend of the metrology tool, the performance change trend of the process tool sensor, and the performance change trend of the metrology tool sensor; correct and determine the sixth data for the third data according to the performance change trend of the metrology tool; correct and determine the seventh data for the fourth data according to the performance change trends of the process tool sensor and the metrology tool sensor; and determine the first data according to the combination of the sixth data, the seventh data, and the fifth data.
[0088] It should be noted that the above explanations of the embodiments and beneficial effects of the control method for the intelligent chip factory are also applicable to the control device of the intelligent chip factory in the embodiments of the present application. To avoid redundancy, no detailed elaboration is made here.
[0089] Corresponding to the above embodiments, the present application also proposes a computer-readable storage medium.
[0090] The computer-readable storage medium of the present application stores a chip optimization program for an intelligent chip factory. When the chip optimization program for the intelligent chip factory is executed by a processor, the foregoing control method for the intelligent chip factory is implemented.
[0091] It should be noted that the above explanations of the embodiments and beneficial effects of the control method for the intelligent chip factory are also applicable to the computer-readable storage medium in the embodiments of the present application. To avoid redundancy, no detailed elaboration is made here.
[0092] Corresponding to the above embodiments, the present application also proposes a controller.
[0093] See Figure 4As shown, the controller 400 of the present application includes a memory 410, a processor 420, and a chip optimization program of the intelligent chip factory stored on the memory 410 and operable on the processor 420. When the processor executes the chip optimization program of the intelligent chip factory, the chip optimization method of the foregoing intelligent chip factory is implemented.
[0094] It should be noted that the above explanations of the embodiments and beneficial effects of the chip optimization method of the intelligent chip factory are also applicable to the controller of the embodiments of the present application. To avoid redundancy, they will not be elaborated in detail here.
[0095] It should be noted that the logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in conjunction with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection portion (electronic device) having one or more wirings, a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or other suitable processing as necessary, and then stored in a computer memory.
[0096] It should be understood that each part of the present application can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.
[0097] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of this application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples.
[0098] In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of such features. In the description of this application, "a plurality of" means at least two, such as two, three, etc., unless otherwise specifically and clearly defined.
[0099] In this application, unless otherwise clearly specified and limited, the terms such as "install", "connect", "connection", "fix" etc. shall be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or integrated; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two elements or the interaction relationship between two elements, unless otherwise clearly limited. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.
[0100] Although the embodiments of this application have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting this application. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.
Claims
1. A chip optimization method for an intelligent chip factory, characterized in that: The method comprises: Obtaining raw data and corrected data; Determine first data according to the original data and the corrected data; Performing multiple data filtering on the first data to determine multiple second data; Training a preset model according to each of the second data to determine a preset model corresponding to the second data; Using process tool operating parameters and target product parameters as inputs of the plurality of preset models to output corresponding predicted product parameters; A target preset model is determined according to the plurality of predicted product parameters and the target product parameters.
2. The chip optimization method of the smart chip factory according to claim 1, characterized in that: Determining a target preset model according to the plurality of predicted product parameters and the target product parameters includes: Determining the prediction accuracy of each of the predicted product parameters based on the difference between each of the predicted product parameters and the target product parameters; A target preset model is determined according to the prediction accuracy of the plurality of predicted product parameters.
3. The chip optimization method of the smart chip factory according to claim 2, characterized in that: Determining a target preset model according to the prediction accuracy of the plurality of predicted product parameters includes: The preset model corresponding to the maximum prediction accuracy is used as the target preset model.
4. The chip optimization method of the smart chip factory according to claim 1, characterized in that: The process tool operation parameters and the target product parameters are respectively used as inputs of the plurality of preset models to output corresponding predicted product parameters, including: Inputting the process tool operating parameters into the plurality of preset models respectively, so that each preset model outputs a corresponding relationship between the process tool operating parameters and the product parameters; Determining corresponding target process tool operating parameters according to the target product parameters and the plurality of corresponding relationships; The process tools are controlled respectively based on the plurality of target process tool operating parameters to determine corresponding predicted product parameters.
5. The chip optimization method of the smart chip factory according to claim 1, characterized in that: Performing multiple data filtering on the first data to determine multiple second data includes: Determining a plurality of data filtering intervals, wherein the plurality of data filtering intervals have different range sizes; The first data is subjected to data filtering processing based on each data filtering interval to determine the plurality of second data.
6. The chip optimization method of the smart chip factory according to claim 1, characterized in that: The correction data includes measurement tool calibration data, measurement tool sensor calibration data, process tool sensor calibration data and laser exposure alignment offset data, the raw data includes online monitoring data, offline monitoring data, measurement tool sensor data, process tool sensor data and scan map digitization data, and determining the first data according to the raw data and the correction data includes: Correcting the online monitoring data and the offline monitoring data according to the metrology tool calibration data and / or the laser exposure alignment offset data to determine third data; Correcting the metrology tool sensor data and the process tool sensor data according to the metrology tool sensor calibration data and the process tool sensor calibration data to determine fourth data; Correcting the scanned map digitized data according to the laser exposure alignment offset data to determine fifth data; The first data is determined according to a collection of the third data, the fourth data and the fifth data.
7. The chip optimization method of the smart chip factory according to claim 6, characterized in that: The method further comprises: Obtaining the performance change trend of the measurement tool, the performance change trend of the process tool sensor, and the performance change trend of the measurement tool sensor; Correcting the third data according to the performance change trend of the measuring tool to determine sixth data; Correcting the fourth data according to the performance change trend of the process tool sensor and the performance change trend of the measurement tool sensor to determine the seventh data; The first data is determined according to a collection of the sixth data, the seventh data and the fifth data.
8. A chip optimization device for an intelligent chip factory, characterized in that: The device comprises: An acquisition module, used for acquiring original data and correction data; A first determining module, configured to determine first data according to the original data and the corrected data; A second determination module, configured to perform multiple data filtering on the first data to determine multiple second data; A third determination module, used for training a preset model according to each of the second data to determine a preset model corresponding to the second data; a fourth determination module, configured to use the process tool operation parameters and the target product parameters as inputs of the plurality of preset models respectively, so as to output corresponding predicted product parameters; The fifth determination module is used to determine the target preset model according to the multiple predicted product parameters and the target product parameters.
9. A computer-readable storage medium, characterized in that: A chip optimization program of an intelligent chip factory is stored thereon, and when the chip optimization program of the intelligent chip factory is executed by a processor, the control method of the intelligent chip factory according to claims 1-7 is implemented.
10. A controller, characterized in that: It includes a memory, a processor, and a chip optimization program of an intelligent chip factory stored in the memory and executable on the processor. When the processor executes the chip optimization program of the intelligent chip factory, the control method of the intelligent chip factory according to claims 1-7 is implemented.