Training method of prediction model and computer device

Abstract

The application relates to the field of integrated circuits, and discloses a training method of a prediction model and a computer device. The method comprises the following steps: acquiring historical measurement data of electroplating liquid concentration; based on the historical measurement data, a static prediction model is established and trained until a static test standard is met; parameters of the static prediction model are copied as initial parameters of a dynamic prediction model; real-time measurement data of the electroplating liquid concentration is acquired, and a dynamic prediction result is obtained by predicting the real-time measurement data by the dynamic prediction model; based on the real-time measurement data, the dynamic prediction model is trained, and the real-time measurement data and the dynamic prediction result are compared until a dynamic test standard is met; and the parameters of the dynamic prediction model are updated to the static prediction model.

Description

Technical Field

[0001] This application relates to the field of integrated circuits, and specifically to a method for training a prediction model and a computer device. Background Technology

[0002] In integrated circuit manufacturing workshops, electroplating equipment is used for ECP (Electrical Chemical Plating) processes. During electroplating, the equipment consumes plating solution, requiring close monitoring of the concentrations of various components within the solution. These components may include copper ions, chloride ions, sulfate ions, accelerators, leveling agents, and inhibitors. The concentration of the plating solution is typically measured periodically (e.g., every 6 hours) by a dedicated measuring device. The measured concentration is then fed back to the equipment system, which replenishes the solution as needed based on the actual measured value.

[0003] However, the equipment system can only perform simple calculations based on the amount of solution discharged and added, and the calculated values ​​will deviate significantly from the actual values, leading to serious quality problems in the wafers. Furthermore, the measurement equipment uses electrochemical methods, requiring periodic electrode replacement to ensure accuracy. Therefore, high-frequency measurements increase costs and can also result in measurement anomalies. Summary of the Invention

[0004] In view of this, embodiments of this application provide a method for training a prediction model and a computer device, which can improve the accuracy of electroplating solution concentration prediction and reduce the measurement frequency of electroplating solution concentration.

[0005] The technical solution of this application embodiment is implemented as follows:

[0006] This application provides a method for training a prediction model, comprising: acquiring historical measurement data of electroplating solution concentration; establishing and training a static prediction model based on the historical measurement data until a static test standard is met; copying the parameters of the static prediction model as the initial parameters of a dynamic prediction model; acquiring real-time measurement data of electroplating solution concentration and using the dynamic prediction model to predict the dynamic prediction result; training the dynamic prediction model based on the real-time measurement data and comparing the real-time measurement data with the dynamic prediction result until a dynamic test standard is met; and updating the parameters of the dynamic prediction model to the static prediction model.

[0007] In some embodiments of this application, the step of establishing and training a static prediction model based on the historical measurement data until the static testing criteria are met includes: dividing the historical measurement data into static training data and static test data according to a ratio; inputting the static training data into the training dataset of the static prediction model for training to establish the static prediction model; inputting the static test data into the test dataset of the static prediction model for testing and retraining the static prediction model until the static testing criteria are met; wherein, the static testing criteria include: the root mean square error on the test dataset is lower than a first threshold.

[0008] In some embodiments of this application, the real-time measurement data includes: first real-time measurement data; the first real-time measurement data is obtained by measuring according to a first cycle; the dynamic prediction result includes: a first dynamic prediction result; the first dynamic prediction result is obtained by the dynamic prediction model predicting according to the first cycle; the step of training the dynamic prediction model based on the real-time measurement data and comparing the real-time measurement data and the dynamic prediction result until a dynamic test standard is met includes: training the dynamic prediction model based on the first real-time measurement data and comparing each first real-time measurement data with the first dynamic prediction result at the same moment until a first dynamic test standard is met; wherein, the first dynamic test standard includes: the difference between each first real-time measurement data and the first dynamic prediction result at the same moment is lower than a second threshold.

[0009] In some embodiments of this application, the real-time measurement data further includes: second real-time measurement data; the second real-time measurement data is obtained by measuring according to a second period; the dynamic prediction result further includes: a second dynamic prediction result; the second dynamic prediction result is obtained by the dynamic prediction model predicting according to a third period; wherein, the second period is longer than the third period; the step of training the dynamic prediction model based on the real-time measurement data and comparing the real-time measurement data and the dynamic prediction result until the dynamic test standard is met further includes: training the dynamic prediction model based on the second real-time measurement data and / or a portion of the verified second dynamic prediction results, and comparing each second real-time measurement data with the second dynamic prediction result at the same moment until the second dynamic test standard is met; wherein, the second dynamic test standard includes: the difference between each second real-time measurement data and the second dynamic prediction result at the same moment is lower than a third threshold.

[0010] In some embodiments of this application, if the second dynamic prediction result of at least two consecutive second real-time measurement data is verified at the same time, then the second dynamic prediction result of the time period between at least two consecutive second real-time measurement data is verified.

[0011] In some embodiments of this application, the historical measurement data and the real-time measurement data include: first-level data and second-level data; wherein, the priority of training using the first-level data is higher than the priority of training using the second-level data.

[0012] In some embodiments of this application, the first-level data includes: process time, precise cathode drain flow rate, electricity used in electroplating, position of the gripper during wafer cleaning, rotation speed of the gripper during wafer cleaning, water flow rate during wafer cleaning, number of processed wafers, and previous measurement data; the second-level data includes: cathode circulation pump pressure, cathode circulation pump rotation speed, total anolyte supply flow rate, total cathode supply flow rate, wafer entry speed into the electroplating solution, and wafer entry angle into the electroplating solution.

[0013] In some embodiments of this application, a neural network algorithm is used for training; the neural network algorithm includes: KAN network.

[0014] In some embodiments of this application, the historical measurement data and the real-time measurement data are obtained from the FDC database.

[0015] This application also provides a computer device, including a memory and a processor; the memory stores a computer program that can run on the processor; when the processor executes the program, it implements the steps in the method described above.

[0016] It is understood that in this embodiment, a static prediction model is established by training based on historical measurement data; then, training continues based on real-time measurement data to continuously improve the performance of the static prediction model. In this way, a model capable of effectively predicting the concentration of electroplating solution is obtained, improving prediction accuracy; simultaneously, the measurement frequency of the measuring equipment can be reduced accordingly, lowering measurement costs. Attached Figure Description

[0017] Figure 1 Schematic diagram of the implementation flow of the training method for the prediction model provided in the embodiments of this application Figure 1 ;

[0018] Figure 2 Schematic diagram of the implementation flow of the training method for the prediction model provided in the embodiments of this application Figure 2 ;

[0019] Figure 3Schematic diagram of the implementation flow of the training method for the prediction model provided in the embodiments of this application Figure 3 ;

[0020] Figure 4 Schematic diagram of the implementation flow of the training method for the prediction model provided in the embodiments of this application Figure 4 ;

[0021] Figure 5 A schematic diagram of the architecture of the training method for the prediction model provided in the embodiments of this application;

[0022] Figure 6 A schematic diagram of the hardware entity of the computer device provided in the embodiments of this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application are further described in detail below with reference to the accompanying drawings and embodiments. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] In the following description, references to "some embodiments" refer to a subset of all possible embodiments. It is understood that "some embodiments" may be the same or different subsets of all possible embodiments and may be combined with each other without conflict. The terms "first / second / third" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first / second / third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0025] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used herein is for descriptive purposes only and is not intended to be limiting of this application.

[0026] Figure 1 This is a schematic diagram illustrating an optional implementation flow of the training method for the prediction model provided in this application embodiment. For example... Figure 1 As shown, the training method for the prediction model includes steps S101 to S106. Each step will be explained in detail.

[0027] S101. Obtain historical measurement data of electroplating solution concentration.

[0028] In this embodiment of the application, all characteristic data (characteristic data related to the concentration of electroplating solution) generated by the electroplating equipment over a historical period can be obtained, as well as the corresponding concentration data in the measuring device. That is, historical measurement data of the concentration of electroplating solution can be obtained.

[0029] For example, if an electroplating machine has two process modules and generates one measurement data point every 6 hours, then approximately 2,920 measurement data points can be collected in a year. Obtaining these 2,920 historical measurement data points from the past year allows for initial training of the electroplating machine.

[0030] S102. Based on historical measurement data, establish a static prediction model and train it until it meets the static testing standards.

[0031] In this embodiment, after obtaining historical measurement data, the historical measurement data can be cleaned to remove outliers, and then classified, summarized, and statistically analyzed to convert it into statistical quantities. The statistical quantities can include feature data X and label data Y. Feature data X includes feature data related to the concentration of the electroplating solution, while label data Y can be set as the corresponding measurement data.

[0032] In some embodiments of this application, the integral of the data between every two adjacent measurement intervals can be calculated. For example, if the first measurement data is 12 ml / L, and the next measurement data is obtained six hours later, then the integral of each feature data over six hours can be calculated. Furthermore, the calculated data can be used to form a feature vector Z, for example, feature vector Z = (X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X12, X13, X14), representing 14 dimensions of feature data X. The feature vector Z can be fed into a deep neural network for training.

[0033] In this embodiment, the dataset can be divided into a training dataset and a test dataset. The training dataset is fed into a deep neural network for training, and the trained static prediction model is tested on the test dataset. If the test results on the test dataset meet the static test criteria, it indicates that the static prediction model has reached a usable or excellent state. In this way, a model capable of predicting the concentration of electroplating solution is initially established.

[0034] S103. Copy the parameters of the static prediction model as the initial parameters of the dynamic prediction model.

[0035] In this embodiment, the parameters of the static prediction model can be used as the initial parameters of the dynamic prediction model. Then, the dynamic prediction model can be trained based on all the feature data generated in real time by the electroplating equipment and the concentration data measured in real time by the corresponding measuring equipment, thereby further improving the prediction accuracy of the model.

[0036] S104. Obtain real-time measurement data of electroplating solution concentration, and obtain dynamic prediction results by dynamic prediction model.

[0037] In this embodiment of the application, after initially establishing a model capable of predicting the concentration of the electroplating solution, the dynamic prediction model can predict the concentration of the electroplating solution and simultaneously acquire real-time measurement data of the electroplating solution concentration to continue training the model. The real-time measurement data includes all feature data generated in real time by the electroplating equipment (feature data related to the concentration of the electroplating solution), as well as the corresponding concentration data in the measuring equipment.

[0038] It should be noted that each time real-time measurement data is acquired, it can be all data within a certain time period. For example, real-time measurement data can be acquired every 6 hours, with each acquisition including all data within the 6-hour interval.

[0039] In some embodiments of this application, the prediction period of the dynamic prediction model and the measurement period in real time can be the same, for example, once every 6 hours; in this way, each prediction result has a real-time measurement data at the same moment, which is beneficial to accurately verify the prediction effect of the model.

[0040] In some embodiments of this application, the prediction period of the dynamic prediction model and the measurement period in real time may be different. For example, the prediction period of the dynamic prediction model is once every 6 hours, while the measurement period in real time is once every 12 hours. In this way, the frequency of real-time measurement can be appropriately reduced, thereby reducing the cost of measurement.

[0041] S105. Based on real-time measurement data, train the dynamic prediction model and compare the real-time measurement data with the dynamic prediction results until the dynamic test standards are met.

[0042] In this embodiment, after acquiring real-time measurement data, the data can be cleaned to remove outliers. Then, it is categorized, summarized, and statistically analyzed to obtain feature data. Similar to the training method of static prediction models, the data integral between every two adjacent measurement intervals can be calculated, and the calculated data can be used to form a feature vector. This feature vector is then fed into a deep neural network for training.

[0043] In this embodiment, real-time measurement data is used not only to train the dynamic prediction model but also to verify the dynamic prediction results. If the real-time measurement data and the dynamic prediction results meet the dynamic testing criteria after comparison, then the dynamic prediction model has been effectively trained, that is, the phased dynamic training objective has been achieved.

[0044] S106. Update the parameters of the dynamic prediction model to the static prediction model.

[0045] In this embodiment of the application, after the training results of the dynamic prediction model meet the dynamic testing criteria, that is, after the phased dynamic training objectives are achieved, the parameters of the dynamic prediction model can be updated to the static prediction model to improve the performance of the static prediction model, and the static prediction model can continuously predict the concentration of the electroplating solution.

[0046] In this embodiment, the training process of the dynamic prediction model can be carried out continuously in stages, thereby continuously improving the performance of the static prediction model, making the model increasingly closer to the real physical state, and continuously improving the model's generalization ability. Correspondingly, while the performance of the static prediction model continues to improve, the measurement frequency of the measurement equipment can be gradually reduced to lower the measurement cost.

[0047] In this embodiment, the static prediction model is applied to actual production, providing assistance in measuring and monitoring the concentration of the electroplating solution during the production process. Therefore, the static prediction model needs to maintain a certain degree of stability and cannot be updated or iterated at any time. The dynamic prediction model, on the other hand, is not applied to actual production; therefore, it can be continuously trained. If the dynamic prediction model achieves better training results, its parameters can be updated to the static prediction model, thereby improving the accuracy of the static prediction model. Conversely, if the dynamic prediction model exhibits poor training results, these results will not be updated to the static prediction model to avoid adverse effects on actual production.

[0048] It is understood that in this embodiment, a static prediction model is established based on historical measurement data for training, providing assistance in the measurement and monitoring of electroplating solution concentration during the production process. Then, based on real-time measurement data, further training is conducted to obtain a more effective dynamic prediction model. The parameters of this improved dynamic prediction model are then updated in the static prediction model, continuously improving its performance. In this way, a model capable of effectively predicting electroplating solution concentration is obtained, improving prediction accuracy. Simultaneously, the measurement frequency of the measuring equipment can be reduced accordingly, lowering measurement costs.

[0049] In some embodiments of this application, it can be achieved through Figure 2 The steps S201 to S203 shown are used to achieve the following:Figure 1 Step S102 is shown. Each step will be explained in detail.

[0050] S201. Divide the historical measurement data into static training data and static test data according to the proportion.

[0051] In this embodiment, the acquired historical measurement data can be divided into static training data and static test data according to a certain ratio. For example, 80% of the historical measurement data can be divided into static training data, and the remaining 20% ​​into static test data. The static training data is used to train the model, while the static test data is used to evaluate the model.

[0052] S202. Input the static training data into the training dataset of the static prediction model, train it, and establish the static prediction model.

[0053] S203. Input static test data into the test dataset of the static prediction model, test and retrain the static prediction model until the static test criteria are met; wherein, the static test criteria include: the root mean square error on the test dataset is lower than the first threshold.

[0054] In this embodiment, static training data is input into the training dataset for training, thus initially establishing a static prediction model. Static test data is then input into the test dataset, and the static prediction model needs to be tested on the test dataset to improve prediction accuracy.

[0055] In this embodiment of the application, if the root mean square error (RMES) on the test dataset is lower than a first threshold, it can be considered that the static prediction model has reached a usable or excellent state. For example, when the root mean square error of the model on the test dataset is lower than 0.01, it indicates that the model's performance is usable; when the root mean square error of the model on the test dataset is lower than 0.001, it indicates that the model's performance is excellent.

[0056] In some embodiments of this application, it can be achieved through Figure 3 The shown step S1051 is to achieve Figure 1 Step S105 is shown. The explanation will be based on each step.

[0057] S1051. Based on the first real-time measurement data, train the dynamic prediction model and compare each first real-time measurement data with the first dynamic prediction result at the same moment until the first dynamic test standard is met; wherein, the first dynamic test standard includes: the difference between each first real-time measurement data and the first dynamic prediction result at the same moment is lower than the second threshold.

[0058] In this embodiment, the real-time measurement data may include first real-time measurement data; wherein, the first real-time measurement data is obtained by measuring according to a first cycle. The dynamic prediction result may include a first dynamic prediction result; wherein, the first dynamic prediction result is obtained by the dynamic prediction model making predictions according to a first cycle. That is, the cycle of the first dynamic prediction result and the cycle of the first real-time measurement data are consistent, both being the first cycle, for example, once every 6 hours. In this way, each first dynamic prediction result has a first real-time measurement data at a simultaneous moment, thereby facilitating accurate verification of the model's prediction effect.

[0059] In some embodiments of this application, the second threshold may be set to 3%.

[0060] In some embodiments of this application, it can be achieved through Figure 4 The shown step S1052 is to achieve Figure 1 Step S105 is shown. The explanation will be based on each step.

[0061] S1052. Based on the second real-time measurement data and / or the verified partial second dynamic prediction results, train the dynamic prediction model and compare each second real-time measurement data with the second dynamic prediction results at the same moment until the second dynamic test standard is met; wherein, the second dynamic test standard includes: the difference between each second real-time measurement data and the second dynamic prediction results at the same moment is lower than a third threshold.

[0062] In this embodiment, the real-time measurement data may further include second real-time measurement data; wherein the second real-time measurement data is obtained by measuring according to a second cycle. The dynamic prediction result may further include a second dynamic prediction result; wherein the second dynamic prediction result is obtained by the dynamic prediction model predicting according to a third cycle. The second cycle is longer than the third cycle. That is, the cycle of the second real-time measurement data is longer than the cycle of the second dynamic prediction result; for example, the cycle of the second dynamic prediction result is once every 6 hours, while the cycle of the second real-time measurement data is once every 12 hours. This appropriately reduces the frequency of real-time measurements, thereby reducing measurement costs.

[0063] In some embodiments of this application, the third threshold may be set to 5%.

[0064] In some embodiments of this application, if the second dynamic prediction result of at least two consecutive second real-time measurement data is verified at the same time, then the second dynamic prediction result of the time period between at least two consecutive second real-time measurement data is verified.

[0065] For example, the period for the second dynamic prediction result is once every 6 hours, while the period for the second real-time measurement data is once every 12 hours. Second real-time measurement data A and B are measured at 0:00 and 12:00 respectively, and second dynamic prediction results a, b1, and b are predicted at 0:00, 6:00, and 12:00 respectively. If two consecutive second real-time measurement data A and B are consistent with two simultaneous second dynamic prediction results a and b, that is, the two second dynamic prediction results a and b are verified, then the second dynamic prediction result b1 for the time interval between the two second dynamic prediction results a and b can also be considered verified.

[0066] Furthermore, during the training of the dynamic prediction model, since no corresponding second real-time measurement data was obtained at 6 o'clock, the verified second dynamic prediction result b1 can also be added to the training dataset to supplement the training data.

[0067] In some embodiments of this application, if the difference between at least two consecutive second real-time measurement data and the second dynamic prediction result at the same moment is within 5%, the verification can be considered successful; furthermore, the second dynamic prediction result of the time period between at least two consecutive second real-time measurement data is also verified, and this data can be added to the training dataset to continue training the model in order to improve the performance of the model.

[0068] It is understandable that verifying and judging the second dynamic prediction results between at least two consecutive second real-time measurement data periods, and adding the verified second dynamic prediction results to the training dataset, can effectively supplement the training data, improve the training effect, and thus improve the prediction accuracy of the model.

[0069] In some embodiments of this application, historical measurement data and real-time measurement data include: first-level data and second-level data. The priority for training with first-level data is higher than that for training with second-level data; that is, first-level data has a stronger correlation with the electroplating solution concentration and should be prioritized as feature data for training.

[0070] In some embodiments of this application, the first-level data includes: process time, precise cathode drain flow rate, electricity used in electroplating, position of the gripper during wafer cleaning, rotation speed of the gripper during wafer cleaning, water flow rate during wafer cleaning, number of processed wafers, and the previous measurement data.

[0071] The following is an explanation of the first-level data listed:

[0072] Process time: The process time determines the amount of solute consumed (i.e., the various components in the electroplating solution). The longer the process time, the more solute is consumed.

[0073] Precise discharge flow rate of cathode: Electroplating equipment periodically discharges a portion of the electroplating solution and adds new solutes and solvents, which means the electroplating solution is periodically renewed. The precise discharge flow rate of cathode represents the volume of the electroplating solution that is periodically renewed.

[0074] Electroplating power consumption: The greater the power consumption for electroplating, the faster the electroplating process and the more solute is consumed.

[0075] Position of the chuck during wafer cleaning: The position of the chuck during wafer cleaning affects the amount of cleaning water flowing into the electroplating solution; generally speaking, the higher the chuck is, the less cleaning water flows into the electroplating solution, and the less impact it has on the electroplating solution.

[0076] The rotation speed of the gripper during wafer cleaning: During wafer cleaning, the rotation speed of the gripper also affects the amount of cleaning water flowing into the electroplating solution; generally speaking, the faster the rotation speed of the gripper, the less cleaning water flows into the electroplating solution, and the smaller the impact on the electroplating solution.

[0077] Water flow rate during wafer cleaning: The flow rate of the cleaning water during wafer cleaning also affects the amount of cleaning water flowing into the electroplating solution.

[0078] Number of wafers processed: The more wafers processed (run volume), the more solute is consumed;

[0079] Previous measurement data: The previous measurement data will be used as a reference value for this measurement data.

[0080] In some embodiments of this application, the second-level data includes: cathode circulation pump pressure, cathode circulation pump speed, total flow rate of anode supply solution, total flow rate of cathode supply solution, speed at which the wafer enters the electroplating solution, and angle at which the wafer enters the electroplating solution.

[0081] The following is an explanation of the listed second-level data:

[0082] Cathode circulation pump pressure: The higher the cathode circulation pump pressure, the faster the electroplating solution circulates, thereby increasing the evaporation rate of the electroplating solution and reducing the amount of solvent.

[0083] Cathode circulation pump speed: characterizes the pressure loss in the electroplating solution circulation and also affects the evaporation rate of the electroplating solution;

[0084] Total flow rate of anolyte supply circulation: This affects the ion exchange capacity of the anode and cathode. The larger the total flow rate of anolyte supply circulation, the faster the ions in the cathode solution are replenished.

[0085] Total cathode supply flow rate: characterizes the circulation status of the electroplating solution and affects the evaporation rate of the electroplating solution;

[0086] The speed at which the wafer enters the electroplating solution: that is, the speed at which it enters the water along the Z-axis. The faster the speed at which the wafer enters the water along the Z-axis, the more electroplating solution will splash up when the wafer enters the electroplating solution, and the greater the loss of electroplating solution.

[0087] The angle at which the wafer enters the electroplating solution affects the amount of electroplating solution splashed up when the wafer enters the solution.

[0088] It is understood that the embodiments of this application use a variety of measurement data related to the concentration of electroplating solution as feature data for training. At the same time, the measurement data are prioritized according to their impact on the concentration of electroplating solution. This can effectively improve the training effect and the accuracy of prediction.

[0089] In some embodiments of this application, a neural network algorithm is used for training. This neural network algorithm includes the KAN network.

[0090] It should be noted that the core of KAN (Kolmogorov-Arnold Networks) is the Kolmogorov-Arnold representation theorem, which asserts that complex multivariate functions can be decomposed into simpler one-dimensional functions. Therefore, KAN networks are suitable for handling multivariate models.

[0091] Understandably, training with KAN networks offers advantages such as scalability, accuracy, and interpretability, which can improve training effectiveness and increase the accuracy of predicting electroplating solution concentration. Scalability refers to the ability to decompose complex functions into simpler components, thus efficiently handling large datasets; accuracy refers to achieving higher accuracy and lower loss with fewer parameters; and interpretability refers to insights into how input features are transformed throughout the network, enhancing transparency and understanding.

[0092] In some embodiments of this application, historical measurement data and real-time measurement data are obtained from the FDC database.

[0093] It should be noted that FDC is an abbreviation for Fault Detection and Classification System, which is a data collection and monitoring system used in integrated circuit manufacturing workshops. EAP is an abbreviation for Equipment Automation Programming, which is a system used to control semiconductor manufacturing equipment for automated production.

[0094] In this embodiment, data from the electroplating equipment and measuring equipment can be transmitted to the FDC database via the EAP system, and then the data collected at the 1Hz frequency can be obtained from the FDC database. This facilitates data acquisition and reduces the difficulty of data acquisition.

[0095] Figure 5 This is a schematic diagram of an optional architecture for training a prediction model provided in an embodiment of this application.

[0096] refer to Figure 5 On one hand, the electroplating equipment collects characteristic data related to the concentration of the electroplating solution, and on the other hand, the measuring equipment transmits the corresponding concentration data to the electroplating equipment. This characteristic data and concentration data are uploaded to the FDC database via the EAP system. After statistical processing and cleaning, they are used to train the static prediction model. The prediction results of the static prediction model are then fed back to the electroplating equipment, which calibrates the prediction results.

[0097] On the other hand, the data continuously generated by the measurement equipment is also input into the KAN network to train the dynamic prediction model. The training of the dynamic prediction model can be carried out in stages and cycles. The training parameters of the dynamic prediction model are used to update the static prediction model, so that the predictive ability of the model can be continuously improved.

[0098] It should be noted that, in the embodiments of this application, if the above-described training method for the prediction model is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, or the part that contributes to the related technology, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of this application are not limited to any specific hardware, software, or firmware, or any combination of hardware, software, and firmware.

[0099] This application provides a computer device including a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the program, it implements some or all of the steps in the above-described method.

[0100] Figure 6 This application provides a hardware entity diagram of a computer device as an embodiment of the present application, such as...Figure 6 As shown, the hardware entity of the computer device 1100 includes a processor 1101 and a memory 1102, wherein the memory 1102 stores a computer program that can run on the processor 1101, and the processor 1101 executes the program to implement the steps in the method of any of the above embodiments.

[0101] The memory 1102 stores computer programs that can run on the processor. The memory 1102 is configured to store instructions and applications that can be executed by the processor 1101. It can also cache data to be processed or already processed (e.g., image data, audio data, voice communication data, and video communication data) in the processor 1101 and various modules in the computer device 1100. It can be implemented by flash memory or random access memory (RAM).

[0102] The processor 1101 executes the program to implement the training method of the prediction model described above. The processor 1101 typically controls the overall operation of the computer device 1100.

[0103] This application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements some or all of the steps in the above-described method. The computer-readable storage medium can be transient or non-transient.

[0104] This application provides a computer program including computer-readable code, wherein when the computer-readable code is executed in a computer device, a processor in the computer device performs some or all of the steps in the above-described method.

[0105] This application provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program. When the computer program is read and executed by a computer, it implements some or all of the steps in the above-described method. This computer program product can be implemented specifically through hardware, software, or a combination thereof. In some embodiments, the computer program product is specifically embodied as a computer storage medium; in other embodiments, the computer program product is specifically embodied as a software product, such as a software development kit (SDK), etc.

[0106] It should be noted that the descriptions of the various embodiments above tend to emphasize the differences between them, while their similarities or commonalities can be referred to interchangeably. The descriptions of the above embodiments of the device, storage medium, computer program, and computer program product are similar to the descriptions of the above method embodiments and have similar beneficial effects. For technical details not disclosed in the embodiments of the device, storage medium, computer program, and computer program product of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0107] The aforementioned processor can be at least one of the following: Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), Controller, Microcontroller, and Microprocessor. It is understood that other electronic devices can also implement the functions of the aforementioned processor, and this application does not specifically limit the specific implementation.

[0108] The aforementioned computer storage media / memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), magnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM), etc.; or it can be various terminals that include one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

[0109] It should be understood that the phrase "an embodiment" or "one embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "one embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above steps / processes do not imply a sequential order of execution; the execution order of each step / process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above embodiments of this application are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0110] It should be noted that, in this document, 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. Unless otherwise specified, 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 that element.

[0111] The above description is merely an embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for training a prediction model, characterized in that, include: Obtain historical measurement data of electroplating solution concentration; Based on the historical measurement data, a static prediction model is established and trained until it meets the static test criteria. The parameters of the static prediction model are copied as the initial parameters of the dynamic prediction model; Real-time measurement data of electroplating solution concentration is obtained, and dynamic prediction results are obtained by the dynamic prediction model. Based on the real-time measurement data, the dynamic prediction model is trained, and the real-time measurement data and the dynamic prediction results are compared until the dynamic test standard is met. Update the parameters of the dynamic prediction model to the static prediction model.

2. The training method for the prediction model according to claim 1, characterized in that, The step of establishing and training a static prediction model based on the historical measurement data until it meets the static testing criteria includes: The historical measurement data is divided into static training data and static test data according to a certain ratio; The static training data is input into the training dataset of the static prediction model for training, thereby establishing the static prediction model. The static test data is input into the test dataset of the static prediction model to test and retrain the static prediction model until the static test criteria are met; wherein, the static test criteria include: the root mean square error on the test dataset is lower than a first threshold.

3. The training method for the prediction model according to claim 1, characterized in that, The real-time measurement data includes: first real-time measurement data; the first real-time measurement data is obtained by measuring according to a first cycle; the dynamic prediction result includes: a first dynamic prediction result; the first dynamic prediction result is obtained by the dynamic prediction model predicting according to the first cycle; The step of training the dynamic prediction model based on the real-time measurement data and comparing the real-time measurement data with the dynamic prediction results until the dynamic testing criteria are met includes: Based on the first real-time measurement data, the dynamic prediction model is trained, and each of the first real-time measurement data is compared with the first dynamic prediction result at the same moment until the first dynamic test standard is met; wherein, the first dynamic test standard includes: the difference between each of the first real-time measurement data and the first dynamic prediction result at the same moment is lower than a second threshold.

4. The training method for the prediction model according to claim 1, characterized in that, The real-time measurement data further includes: second real-time measurement data; the second real-time measurement data is obtained by measuring according to a second cycle; the dynamic prediction result further includes: second dynamic prediction result; the second dynamic prediction result is obtained by the dynamic prediction model predicting according to a third cycle; wherein, the second cycle is longer than the third cycle; The step of training the dynamic prediction model based on the real-time measurement data and comparing the real-time measurement data with the dynamic prediction results until the dynamic testing criteria are met also includes: Based on the second real-time measurement data and / or the verified portion of the second dynamic prediction results, the dynamic prediction model is trained, and each second real-time measurement data is compared with the second dynamic prediction result at the same moment until the second dynamic test standard is met; wherein, the second dynamic test standard includes: the difference between each second real-time measurement data and the second dynamic prediction result at the same moment is lower than a third threshold.

5. The training method for the prediction model according to claim 4, characterized in that, If the second dynamic prediction result of at least two consecutive second real-time measurement data is verified at the same time, then the second dynamic prediction result of the time period between at least two consecutive second real-time measurement data is verified.

6. The training method for the prediction model according to claim 1, characterized in that, The historical measurement data and the real-time measurement data include: first-level data and second-level data; Among them, the priority of training with the first level of data is higher than the priority of training with the second level of data.

7. The training method for the prediction model according to claim 6, characterized in that, The first level of data includes: process time, precise cathode drain flow rate, electricity used in electroplating, position of the gripper during wafer cleaning, rotation speed of the gripper during wafer cleaning, water flow rate during wafer cleaning, number of processed wafers, and the previous measurement data. The second level of data includes: cathode circulation pump pressure, cathode circulation pump speed, total flow rate of anode supply solution, total flow rate of cathode supply solution, speed at which the wafer enters the electroplating solution, and angle at which the wafer enters the electroplating solution.

8. The training method for the prediction model according to claim 1, characterized in that, The training is performed using a neural network algorithm, which includes the KAN network.

9. The training method for the prediction model according to claim 1, characterized in that, The historical measurement data and the real-time measurement data are obtained from the FDC database.

10. A computer device comprising a memory and a processor; the memory storing a computer program executable on the processor; the processor executing the program to implement the steps of the method according to any one of claims 1 to 9.

Images