Virtual measurement method, electronic equipment and readable storage medium
By combining equipment parameters and actual measurement data to correct virtual measurement data, the problem of prediction errors caused by complex inputs in virtual measurement methods is solved, thereby improving prediction accuracy and production efficiency.
Patent Information
- Application Number
- CN202511742208.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-25
- Publication Date
- 2026-03-03
AI Technical Summary
Existing virtual measurement methods often lead to unknowing prediction errors due to the presence of complex and unknown input factors, which affects the yield and production efficiency of semiconductor manufacturing.
By acquiring equipment parameter data, sampled measurement data, and measurement data from upstream process stations at the target process station, the virtual measurement data is corrected using a pre-trained virtual measurement model and Kalman filter algorithm. The actual measurement data is then combined for further correction to improve prediction accuracy.
It effectively improves the accuracy of predicted measurement data, reduces prediction errors, and increases production efficiency and product yield.
Smart Images
Figure CN121598081A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of semiconductor processing and manufacturing technology, and in particular to a virtual measurement method, electronic device, and readable storage medium. Background Technology
[0002] Metrology is crucial in semiconductor manufacturing, ensuring process accuracy and consistency throughout the entire wafer fabrication process. However, metrology equipment is expensive, and the metrology process itself is time-consuming. Therefore, reducing unnecessary metrology equipment and increasing throughput is undoubtedly of great value. For example, after the wafer passes through the polishing process station, engineers need to measure the wafer thickness. Please refer to [reference needed]. Figure 1 This is a wafer measurement distribution (map). For example... Figure 1 As shown, the thickness is measured at five key locations on the wafer. The numbers in the diagram indicate the measurement points. A redder color indicates a larger measurement value, and a bluer color indicates a smaller measurement value.
[0003] Virtual Metrology (VM) is a technology that correlates measurement results with equipment parameter data (including various sensor data generated during the operation of semiconductor equipment, such as temperature, pressure, gas flow, etc.) and equipment operating status data, thereby reducing the need for wafer metrology. Virtual metrology can significantly reduce the investment of equipment and human resources, while also reducing a large amount of measurement time, thus shortening the product production cycle and greatly improving production efficiency.
[0004] However, due to the complexity of the semiconductor manufacturing process and the large number of unknown input factors, existing virtual measurements based on equipment parameter data only cover a limited portion of these factors. Even with advanced algorithms, such solutions are difficult to predict accurately. If engineers cannot detect problems in time, it will cause serious yield losses for the company.
[0005] It should be noted that the information disclosed in the background section of this invention is intended only to enhance the understanding of the general background of this invention, and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention
[0006] The purpose of this invention is to provide a virtual measurement method, electronic device, and readable storage medium that can solve the problem that most prediction models are unaware of prediction errors caused by complex inputs by correcting virtual measurement data using actual sampled measurement data.
[0007] To achieve the above objectives, the present invention provides a virtual measurement method, comprising:
[0008] Acquire equipment parameter data, sampling measurement data, and upstream process station measurement data for the target process station. The upstream process station measurement data includes sampling measurement data for all process stations preceding the target process station.
[0009] Based on the equipment parameter data of the target process station and the pre-trained first virtual measurement model, the first virtual measurement data of the target process station is obtained.
[0010] Based on the measurement data of the preceding process station and the pre-trained second virtual measurement model, the second virtual measurement data of the target process station is obtained.
[0011] The first virtual measurement data of the target process station is corrected based on the second virtual measurement data of the target process station to obtain the third virtual measurement data of the target process station.
[0012] The third virtual measurement data of the target process station is corrected based on the sampling measurement data of the target process station to obtain the predicted measurement data of the target process station.
[0013] Optionally, the step of correcting the third virtual measurement data of the target process station based on the sampling measurement data of the target process station to obtain the predicted measurement data of the target process station includes:
[0014] Based on the sampled measurement data of the target process station and the third virtual measurement data of the target process station corrected using the Kalman filter algorithm, the predicted measurement data of the target process station is obtained.
[0015] Optionally, the step of correcting the third virtual measurement data of the target process station based on the sampled measurement data of the target process station and using a Kalman filter algorithm to obtain the predicted measurement data of the target process station includes:
[0016] The third virtual measurement data of the target process station is corrected according to the following Kalman filter formula to obtain the predicted measurement data of the target process station:
[0017]
[0018] In the formula, To predict measurement data, This is the third virtual measurement data, where K is the Kalman gain and C is the first hyperparameter. This is sampled measurement data.
[0019] Optionally, the sampling measurement data includes the sampling wafer measurement mean and the sampling wafer measurement standard deviation; the first virtual measurement data includes the first virtual measurement mean and the first virtual measurement standard deviation; the second virtual measurement data includes the second virtual measurement mean and the second virtual measurement standard deviation; the third virtual measurement data includes the third virtual measurement mean and the third virtual measurement standard deviation; and the predicted measurement data includes the predicted measurement mean and the predicted measurement standard deviation.
[0020] Optionally, the mean of the third virtual measurement can be obtained according to the following formula:
[0021]
[0022] In the formula, u p denoted as the mean of the third virtual measurement, u1 as the mean of the first virtual measurement, u2 as the mean of the second virtual measurement, and m as the second hyperparameter.
[0023] Optionally, the standard deviation of the third virtual measurement can be obtained using the following formula:
[0024]
[0025] In the formula, Let x be the standard deviation of the third virtual measurement, n be the third hyperparameter, and x be the standard deviation of the third virtual measurement. i Let u be the measurement value corresponding to the i-th training sample participating in the training of the first virtual measurement model or the second virtual measurement model, u be the mean of the measurement values of all training samples participating in the training of the first virtual measurement model and the second virtual measurement model, and k be the total number of training samples participating in the training of the first virtual measurement model and the second virtual measurement model.
[0026] Optionally, the first virtual measurement model includes a first feature selection module and a first multivariate regression prediction module. The first feature selection module is configured to filter out first feature data related to the measurement data prediction from the device parameter data. The first multivariate regression prediction module is configured to predict the measurement data based on the first feature data to obtain the first virtual measurement data.
[0027] Optionally, the second virtual measurement model includes a second feature selection module and a second multivariate regression prediction module. The second feature selection module is configured to select second feature data related to measurement data prediction from the measurement data of the upstream process station. The second multivariate regression prediction module is configured to predict the measurement data based on the second feature data to obtain the second virtual measurement data.
[0028] Optionally, the virtual measurement method provided by the present invention further includes:
[0029] A similarity comparison is performed on the third virtual measurement data, the sampled measurement data, and the pre-acquired benchmark measurement data;
[0030] If the similarity value of any two of the third virtual measurement data, the sampled measurement data, and the benchmark measurement data is less than a preset threshold, an alarm is triggered and the first virtual measurement model and the second virtual measurement model are updated.
[0031] To achieve the above objectives, the present invention also provides an electronic device, including a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, it implements the virtual measurement method described above.
[0032] To achieve the above objectives, the present invention also provides a readable storage medium storing a computer program, which, when executed by a processor, implements the virtual measurement method described above.
[0033] Compared with the prior art, the virtual measurement method, electronic device, and readable storage medium provided by the present invention have the following advantages:
[0034] The virtual measurement method provided by this invention first acquires equipment parameter data, sampled measurement data, and measurement data from upstream process stations at the target process station. The upstream process station measurement data includes sampled measurement data from all process stations preceding the target process station. Then, based on the equipment parameter data of the target process station and a pre-trained first virtual measurement model, first virtual measurement data for the target process station is acquired. Next, based on the upstream process station measurement data and a pre-trained second virtual measurement model, second virtual measurement data for the target process station is acquired. Then, the first virtual measurement data for the target process station is corrected based on the second virtual measurement data to obtain third virtual measurement data for the target process station. Finally, the third virtual measurement data for the target process station is corrected based on the sampled measurement data to obtain predicted measurement data for the target process station. Since front-end process site measurement data reflects the cumulative effects of the front-end process, this invention corrects the first virtual measurement data predicted based on equipment parameter data (including various sensor data and equipment operating status data generated during semiconductor equipment operation) by using second virtual measurement data predicted based on front-end process site measurement data. This fully utilizes data from different sources to capture more process variations, thereby effectively improving prediction accuracy. Furthermore, although the actual measurement values of the sampled wafers are only approximate reference values for other wafers, they are still very valuable. Therefore, this invention corrects the third virtual measurement data using sampled measurement data obtained from actual sampling. This allows for correction of prediction errors caused by unknown factors, effectively improving the accuracy of the final predicted measurement data. This also solves the problem that most prediction models are unaware of prediction errors caused by complex inputs.
[0035] Since the electronic device and readable storage medium provided by this invention belong to the same inventive concept as the virtual measurement method provided by this invention, the electronic device and readable storage medium provided by this invention have at least all the beneficial effects of the virtual measurement method provided by this invention. Specifically, the beneficial effects of the virtual measurement method provided by this invention can be described in the above description. Therefore, the beneficial effects of the electronic device and readable storage medium provided by this invention will not be repeated here. Attached Figure Description
[0036] Figure 1 This is a distribution map of wafer measurements.
[0037] Figure 2 This is a flowchart of a virtual measurement method provided in one embodiment of the present invention.
[0038] Figure 3 This is a schematic diagram of the first and second virtual measurement data predicted by the virtual measurement method provided in an embodiment of the present invention.
[0039] Figure 4 This is a structural block diagram of a first virtual measurement model provided in one embodiment of the present invention.
[0040] Figure 5 This is a distribution chart showing the importance of equipment parameter characteristics.
[0041] Figure 6 A heatmap showing the correlation between key equipment parameters.
[0042] Figure 7 This is a structural block diagram of a second virtual measurement model provided in one embodiment of the present invention.
[0043] Figure 8 This is a comparison chart of the third virtual measurement data, the sampled measurement data, and the baseline measurement data.
[0044] Figure 9 The flowchart illustrates the training process of a first virtual measurement model and a second virtual measurement model, as provided in one embodiment of the present invention.
[0045] Figure 10 This is a block diagram of an electronic device provided according to an embodiment of the present invention.
[0046] The reference numerals in the attached drawings are explained as follows: First feature selection module - 110; First multivariate regression prediction module - 120; Second feature selection module - 210; Second multivariate regression prediction module - 220; Processor - 310; Communication interface - 320; Memory - 330; Communication bus - 340. Detailed Implementation
[0047] The virtual measurement method, electronic device, and readable storage medium proposed in this invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Please refer to the accompanying drawings to make the objectives, features, and advantages of this invention more apparent and understandable. It should be noted that the structures, proportions, sizes, etc., illustrated in the accompanying drawings are only for illustrative purposes and to enable those skilled in the art to understand and read the invention, and are not intended to limit the implementation conditions of the invention. Any modifications to the structure, changes in proportions, or adjustments to the size, provided that the effects and objectives achieved by this invention are the same or similar, should still fall within the scope of the technical content disclosed in this invention.
[0048] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, 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 process, method, article, or apparatus. In addition, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated.
[0049] Furthermore, in the description of this specification, the reference to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., means that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions 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. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.
[0050] The core idea of this invention is to provide a virtual measurement method, electronic device, and readable storage medium. By using actual sampled measurement data to correct virtual measurement data, the problem of most prediction models making prediction errors due to complex inputs without realizing it can be solved.
[0051] It should be noted that the virtual measurement method provided by the present invention can be applied to the electronic device provided by the present invention. The electronic device can be a personal computer, a mobile terminal, etc., and the mobile terminal can be a mobile phone, a tablet computer, or other hardware device with various operating systems.
[0052] To achieve the above-mentioned goals, this invention provides a virtual measurement method, please refer to [the relevant documentation]. Figure 2 This is a flowchart of a virtual measurement method provided by an embodiment of the present invention. Figure 2 As shown, the virtual measurement method includes the following steps:
[0053] Step S100: Obtain equipment parameter data, sampling measurement data and upstream process station measurement data of the target process station. The upstream process station measurement data includes sampling measurement data of all process stations located before the target process station.
[0054] Step S200: Based on the equipment parameter data of the target process station and the pre-trained first virtual measurement model, obtain the first virtual measurement data of the target process station;
[0055] Step S300: Based on the measurement data of the preceding process station and the pre-trained second virtual measurement model, obtain the second virtual measurement data of the target process station;
[0056] Step S400: Correct the first virtual measurement data of the target process station based on the second virtual measurement data of the target process station to obtain the third virtual measurement data of the target process station;
[0057] Step S500: Correct the third virtual measurement data of the target process station based on the sampling measurement data of the target process station to obtain the predicted measurement data of the target process station.
[0058] Therefore, this invention corrects the first virtual measurement data predicted based on equipment parameter data (including various sensor data generated during semiconductor equipment operation, such as temperature, pressure, gas flow, etc.) and equipment operating status data) by using second virtual measurement data predicted based on front-end process station measurement data. This allows for full utilization of data from different sources to capture more process variations, thereby effectively improving prediction accuracy. Furthermore, while the actual measurement values of sampled wafers are approximate reference values for other wafers, they are still very valuable. Therefore, this invention corrects the third virtual measurement data using sampled measurement data obtained from actual sampling. This allows for correction of prediction errors caused by unknown factors, effectively improving the accuracy of the final predicted measurement data. This also solves the problem that most prediction models are unaware of prediction errors caused by complex inputs.
[0059] It should be noted that, as those skilled in the art will understand, equipment parameter data can be collected by the FDC (Fault Detection and Classification) system. Furthermore, it should be noted that, as those skilled in the art will understand, due to the different process flows and varying sensitivities to equipment among different products, separate training is required for different products, resulting in different parameters for the final trained models (including the first and second virtual metrology models). It should also be noted that, as those skilled in the art will understand, the virtual metrology method provided by this invention is only applicable to mature and stable mass-produced products (product output > 100 wafers / month). It is not suitable for engineering, new research and development (NTO) products, design of experiments (DOE), or wafers undergoing significant changes. In addition, the virtual metrology method provided by this invention is suitable for relatively simple and easily implemented non-pattern processes (such as CMP (Chemical Mechanical Polishing), WET (Wet Etching), and FUR (Furnace Tube Process)), but is not currently applicable to virtual metrology at process stations requiring more complex data such as photomasks, such as Etch (Dry Etching), Photo (Photolithography), and Implant (Ion Implantation). For process stations such as CMP, WET, and FUR, separate training is required to obtain a first virtual measurement model and a second virtual measurement model for predicting the measurement data of each process station. Furthermore, it should be noted that, as those skilled in the art will understand, the measurement data for the CMP process station can be the thickness data of the thin film on the wafer surface; the measurement data for the WET process station can be the etching uniformity data; and the measurement data for the FUR process station can be the oxide layer thickness data.
[0060] In some exemplary embodiments, the step of correcting the third virtual measurement data of the target process station based on the sampled measurement data of the target process station to obtain the predicted measurement data of the target process station includes:
[0061] Based on the sampled measurement data of the target process station and the third virtual measurement data of the target process station corrected using the Kalman filter algorithm, the predicted measurement data of the target process station is obtained.
[0062] Since the Kalman filter does not completely trust either predictions or measurements, it can dynamically adjust the level of trust in the predicted data (third virtual measurement data) and the actual measurement data (sampled measurement data) by correcting the third virtual measurement data of the target process station based on sampled measurement data and the Kalman filter algorithm. This further improves the accuracy of the final predicted measurement data and effectively solves the problem that most prediction models are unaware of prediction errors caused by complex inputs.
[0063] In some exemplary embodiments, the step of correcting the third virtual measurement data of the target process station based on the sampled measurement data of the target process station and using a Kalman filter algorithm to obtain the predicted measurement data of the target process station includes:
[0064] The third virtual measurement data of the target process station is corrected according to the following Kalman filter formula to obtain the predicted measurement data of the target process station:
[0065] (1)
[0066] In the formula, To predict measurement data, This is the third virtual measurement data, where K is the Kalman gain and C is the first hyperparameter. This is sampled measurement data.
[0067] It should be noted that the first hyperparameter C is obtained through machine learning, and its initial value can be set to 1. The Kalman gain K is not calculated from the covariance matrix, but rather is the weight between the actual measured value and the predicted value obtained through continuous optimization in machine learning after an initial empirical setting. The initial value of Kalman gain K can be set to 0.5. Since the standard Kalman filter algorithm requires maintaining the state covariance matrix and performing recursive calculations, its computational complexity is high. In contrast, the Kalman gain K in this invention is obtained through machine learning, thus avoiding complex real-time matrix operations and making it easier to deploy and run in industrial environments. Furthermore, by optimizing Kalman gain K and the first hyperparameter C, the degree of trust in the model prediction and the actual measurement can be adaptively adjusted. For example, when the measurement equipment is very reliable, a larger K value may be learned to place greater trust in the actual measurement value; when the model prediction is very accurate, the K value may be smaller, thus relying more on the model prediction value. This adaptability improves robustness.
[0068] It should also be noted that, as those skilled in the art will understand, in practice, the weight of actual measurements is generally quite large, unless there is a problem with the testing equipment causing serious errors in the actual measurements. When the Kalman gain K=0, the final predicted measurement data... From virtual measurement data (third virtual measurement data) The final predicted measurement data is determined by the Kalman gain K; conversely, when the Kalman gain K=1, the final predicted measurement data is determined by the Kalman gain K. Based on actual test data from sampled wafers (Z) k )Decide.
[0069] Specifically, the standard Kalman filter equation obtains the posterior estimate by fusing prior predictions and actual predictions:
[0070] (2)
[0071] In the formula, State estimation at time K, Let A be the state estimate at time (K-1), A be the state transition matrix, B be the control input matrix, and K be the state value at time (K-1). k For Kalman gain, Z k U is the observed value, C is the hyperparameter, and U is the hyperparameter. k This is the control input vector at time K, such as the acceleration of a moving object.
[0072] Based on the application of special scenarios in semiconductor testing, the present invention makes the following changes to the standard Kalman filter equation shown in the above formula (2), thereby obtaining the Kalman filter formula shown in the above formula (1).
[0073] In some exemplary embodiments, the sampled measurement data includes the sampled wafer measurement mean and the sampled wafer measurement standard deviation; the first virtual measurement data includes the first virtual measurement mean and the first virtual measurement standard deviation; the second virtual measurement data includes the second virtual measurement mean and the second virtual measurement standard deviation; the third virtual measurement data includes the third virtual measurement mean and the third virtual measurement standard deviation; and the predicted measurement data includes the predicted measurement mean and the predicted measurement standard deviation.
[0074] Because any measurement and prediction in semiconductor manufacturing involves uncertainty, using only the mean would miss this uncertainty information. This invention, by simultaneously providing the mean and standard deviation, can fully describe a probability distribution (e.g., a Gaussian distribution), thus more comprehensively reflecting the statistical characteristics of the data. Furthermore, this invention, by fully describing a prediction distribution (Gaussian distribution, such as...),... Figure 3 As shown, this is a schematic diagram of the first and second virtual measurement data predicted by the virtual measurement method provided in an embodiment of the present invention. (where y1 represents the first virtual measurement data and y2 represents the second virtual measurement data). It is not necessary to excessively pursue absolute accuracy in predicting every value. Since monitoring measurement results is essentially for discovering product quality problems caused by the manufacturing process, if the virtual measurement prediction results have some errors but are relatively close to the true values, then engineers can perform time-series control and SPC (Statistical Process Control) control together with the sampled measurement data and virtual measurement data. This avoids monitoring the entire manufacturing process, thereby lowering the monitoring threshold. Simultaneously, to ensure accurate prediction, the first virtual measurement model will conduct a more comprehensive and thorough analysis of equipment parameter data, thus solving the monitoring problem of FDC (Fault Detection and Classification).
[0075] It should be noted that, as those skilled in the art can understand, the formula (1) above can be used to correct the mean and standard deviation of the third virtual measurement respectively.
[0076] In some exemplary implementations, the mean of the third virtual measurement is obtained according to the following formula (3):
[0077] (3)
[0078] In the formula, u p denoted as the mean of the third virtual measurement, u1 as the mean of the first virtual measurement, u2 as the mean of the second virtual measurement, and m as the second hyperparameter.
[0079] Therefore, by adopting the above formula (3), the first virtual measurement mean predicted based on equipment parameter data can be fine-tuned using the second virtual measurement mean predicted based on upstream process station measurement data, thereby further improving the accuracy of the prediction. It should be noted that, as those skilled in the art will understand, the second hyperparameter m can be learned through machine learning algorithms.
[0080] In some exemplary implementations, the standard deviation of the third virtual measurement is obtained according to the following formula (4):
[0081] (4)
[0082] In the formula, Let x be the standard deviation of the third virtual measurement, n be the third hyperparameter, and x be the standard deviation of the third virtual measurement. i Let u be the measurement value corresponding to the i-th training sample participating in the training of the first virtual measurement model or the second virtual measurement model, u be the mean of the measurement values of all training samples participating in the training of the first virtual measurement model and the second virtual measurement model, and k be the total number of training samples participating in the training of the first virtual measurement model and the second virtual measurement model.
[0083] Therefore, by using the above formula (4), the training data of the first virtual measurement model and the second virtual measurement model can be combined to calculate the standard deviation, thereby ensuring the calculated standard deviation of the third virtual measurement. It can objectively reflect the overall fluctuations, thereby ensuring the corrected predicted distribution (i.e., the third virtual measurement data). This can more accurately reflect the reliability of the prediction. It should be noted that, as those skilled in the art will understand, the third hyperparameter can be learned through machine learning algorithms.
[0084] Please continue to refer to this. Figure 4 This is a structural block diagram of the first virtual measurement model provided in one embodiment of the present invention. Figure 4As shown, in some exemplary embodiments, the first virtual measurement model includes a first feature selection module 110 and a first multivariate regression prediction module 120. The first feature selection module 110 is configured to filter out first feature data related to measurement data prediction from the device parameter data, and the first multivariate regression prediction module 120 is configured to predict measurement data based on the first feature data to obtain the first virtual measurement data.
[0085] Therefore, the first feature selection module 110 can sort numerous equipment parameters by feature degree to filter out key equipment parameters (i.e., first feature data) that are relevant to the measurement data prediction from the equipment parameter data. This reduces feature dimensionality and noise, enabling engineers to trace which specific equipment parameters are abnormal when the first virtual measurement model's prediction deviates. It also avoids interference from irrelevant or redundant features, ensuring the first virtual measurement model can learn the true process rules. Furthermore, feature selection by the first feature selection module 110 can prevent overfitting, improving the generalization ability of the first virtual measurement model. The first multivariate regression prediction module 120 can establish a mapping relationship between the filtered first feature data and the measurement data, ensuring the first virtual measurement model can successfully predict the first virtual measurement data.
[0086] Specifically, the first virtual measurement model can be an XGBoost regression model based on the XGBOOST (Extreme Gradient Boosting) algorithm. Since the XGBOOST algorithm converges faster and is less prone to overfitting compared to traditional tree models, using a first virtual measurement model based on the XGBOOST algorithm can solve the problems of highly correlated variables affecting model stability when engineers select features based on experience, and the loss of some nonlinear features when directly using a traditional linear regression model. Please refer to Table 1, which is the pseudocode table of the first virtual measurement model based on the XGBOOST algorithm.
[0087] Table 1 XGBOOST Pseudocode XGBOOST pseudocode 1. Prepare data: Input (equipment parameter data) and output (measured values) 2. Initialize prediction: All samples are predicted as the average of the measurement values. 3. Repeat tree building (T rounds in total): - Calculate the current error = actual measurement value - current predicted value - Train a tree: use device parameter data to predict and reduce error - Update prediction: new prediction = old prediction + learning rate × tree output: 4. Final prediction = Initial value + Sum of adjustments to all trees (learning rate × tree output)
[0088] To facilitate understanding, the following example uses the PAD (oxide layer) oxidation process station in semiconductor furnace tube technology to illustrate how to predict measurement data using equipment parameters. There are 34 equipment parameters for the PAD (oxide layer) oxidation process station in the furnace tube process, numbered F1 to F... 34It is stated that, after expert selection and analysis using the XGBOOST algorithm (first feature selection module 110) (the analysis results of the XGBOOST algorithm will be confirmed by semiconductor experts to eliminate random errors), 13 key equipment parameters (F1~F2) related to the measurement data prediction were finally selected. 13 The degree of characteristic of each device parameter is as follows: Figure 5 As shown, the correlation heatmap among the selected key equipment parameters is as follows: Figure 6 As shown.
[0089] Based on the 13 key equipment parameters selected, a multivariate regression model of key equipment parameters and measurement data was established using the machine learning algorithm Ridge. Considering the special characteristics of this algorithm, the input feature data (key equipment parameters) were first standardized. The general multivariate regression model is shown in the following formula (5):
[0090] (5)
[0091] In the formula, y is the predicted value. It is the intercept term. , ... These are the regression coefficients of each feature. , ... This is the input feature data. Based on the actual training data, through calculation and analysis, the feature importance values and ridge regression coefficients of the top 13 key equipment parameters were obtained, as shown in Table 2.
[0092] Table 2 Importance values of equipment parameter features and ridge regression coefficients Ranking feature Importance score Ridge regression coefficient 1 <![CDATA[F1]]> 0.1041 0.053 2 <![CDATA[F2]]> 0.1008 0.2876 3 <![CDATA[F3]]> 0.0937 0.4521 4 <![CDATA[F4]]> 0.0795 0.04 5 <![CDATA[F5]]> 0.079 0.321 6 <![CDATA[F6]]> 0.0672 0.007 7 <![CDATA[F7]]> 0.0636 0.3987 8 <![CDATA[F8]]> 0.055 0.221 9 <![CDATA[F9]]> 0.0505 0.061 10 <![CDATA[F 10 ]]> 0.0481 0.1543 11 <![CDATA[F 11 ]]> 0.0466 0.1987 12 <![CDATA[F 12 ]]> 0.0375 0.1543 13 <![CDATA[F 13 ]]> 0.0217 0.132 14 intercept 0.452
[0093] Based on Table 2, the final regression model of equipment parameters and measurement data can be obtained, as shown in the following formula (6):
[0094] (6)
[0095] In the formula, y1 represents the predicted value of the measurement data.
[0096] Therefore, the first virtual measurement model (specifically the first multiple regression prediction module 120) can predict the first virtual measurement data (including the first virtual measurement mean and the first virtual measurement standard deviation) according to the above formula (6).
[0097] It should be noted that, as those skilled in the art will understand, the first virtual measurement model can be, in addition to the XGBoost regression model based on the XGBOOST (Extreme Gradient Boosting) algorithm, a machine learning model such as the Bagging (Self-Governing Aggregation) regression model, the Random Forest regression model, the ExtraTree (Extreme Random Tree) regression model, the Gradient Boosting regression model, and the CatBoost (Category Boosting) regression model.
[0098] Please continue to refer to this. Figure 7 This is a structural block diagram of the second virtual measurement model provided in one embodiment of the present invention. Figure 7 As shown, in some exemplary embodiments, the second virtual measurement model includes a second feature selection module 210 and a second multivariate regression prediction module 220. The second feature selection module 210 is configured to select second feature data related to measurement data prediction from the measurement data of the upstream process station. The second multivariate regression prediction module 220 is configured to predict the measurement data based on the second feature data to obtain the second virtual measurement data.
[0099] Therefore, the second feature selection module 210 can extract the measurement results of a few key upstream process stations that significantly affect the measurement data of the current process station (target process station) from complex upstream measurement data. This not only helps process engineers understand the correlations in the process and provides guidance for process optimization, but also reduces feature dimensionality, noise, and avoids interference from irrelevant or redundant features on the second virtual measurement model. It also avoids overfitting and helps improve the generalization ability of the second virtual measurement model. The second multivariate regression prediction module 220 can establish a mapping relationship between the selected second feature data and the measurement data, ensuring that the second virtual measurement model can successfully predict the second virtual measurement data.
[0100] It should be noted that, as those skilled in the art will understand, the second virtual measurement model can be a machine learning model such as XGBoost regression model, Bagging regression model, Random Forest regression model, ExtraTree regression model, Gradient Boosting regression model, or CatBoost regression model. For a more accurate understanding of the specific prediction principle of the second virtual measurement model, please refer to the relevant content above regarding the prediction principle of the first virtual measurement model; further explanation will not be provided here.
[0101] In some exemplary embodiments, the virtual measurement method provided by the present invention further includes:
[0102] A similarity comparison is performed on the third virtual measurement data, the sampled measurement data, and the pre-acquired benchmark measurement data;
[0103] If the similarity value of any two of the third virtual measurement data, the sampled measurement data, and the benchmark measurement data is less than a preset threshold, an alarm is triggered and the first virtual measurement model and the second virtual measurement model are updated.
[0104] Therefore, by comparing the similarity between the third virtual measurement data and the sampled measurement data, and issuing an alarm when the similarity value between the third virtual measurement data and the sampled measurement data is less than a preset threshold, the production line can promptly take temporary control measures and increase the sampled measurement data. Simultaneously, it allows engineers to intervene in a timely manner to investigate the cause of the prediction deviation, thereby reducing company losses. After finding the cause, updating the first and second virtual measurement models can continuously improve the accuracy of the virtual measurement algorithm and the system's capabilities. By comparing the similarity between the third virtual measurement data and the benchmark measurement data, deviations between the third virtual measurement data and the benchmark measurement data can be captured; by comparing the similarity between the sampled measurement data and the benchmark measurement data, deviations between the sampled measurement data and the benchmark measurement data can be captured, thus helping to capture overall process drift.
[0105] It should be noted that the specific details regarding how to calculate the similarity values between predicted measurement data and sampled measurement data, between predicted measurement data and baseline measurement data, and between sampled measurement data and baseline measurement data can be adapted to understanding from relevant content known to those skilled in the art, and will not be elaborated upon here. It should also be noted that, as those skilled in the art will understand, when the predicted distribution of any product on the same platform is severely abnormal, all other products on the same platform should be scheduled for additional testing (increasing the sample size for sampling measurement) to ensure product quality. Furthermore, it should be noted that, as those skilled in the art will understand, this invention does not limit the specific value of the preset threshold; the specific value of the preset threshold can be set according to actual needs, for example, the preset threshold can be set to 90%. Additionally, it should be noted that, as those skilled in the art will understand, the baseline measurement data can be obtained based on the actual measurement data of the wafer at the target site that is recognized by the user.
[0106] Please refer to Figure 8 This is a comparison chart of the third virtual measurement data, the sampled measurement data, and the baseline measurement data. For example... Figure 8 As shown, due to the third virtual measurement data With sampling measurement data Z k The distributions are relatively close, indicating that the predicted third virtual measurement data... It is accurate, but the third virtual measurement data The distribution of the baseline measurement data BL, which was previously accepted by users, is significantly different, and an anomaly alarm will be triggered at this time.
[0107] The following is a brief explanation of the specific training process of the first and second virtual measurement models.
[0108] Specifically, before training the first and second virtual measurement models, some data and scenario implementation preparations are needed. Taking the furnace tube process site as an example, the furnace tube data collection work must be done first. Before building the model, engineers need to perform correlation analysis between electrical parameters (CP (wafer electrical test) / WAT (wafer acceptance test)) and the measurement data of the furnace tube process site to analyze the impact of the measurement data size on the electrical parameters, and to check whether the current furnace tube process site measurement data specification range is reasonable, so as to facilitate effective SPC monitoring of the predicted data in the later stage.
[0109] Further, please refer to Figure 9 This is a flowchart illustrating the training process of the first and second virtual measurement models provided in one embodiment of the present invention. Figure 9 As shown, since equipment parameter data has a significant impact on measurement results, the training of the first virtual measurement model based on equipment parameter data is called master data training. Other external data will also affect the measurement results. For example, when the measurement data of the front end changes significantly, it will also affect the measurement results of the back end. Therefore, the training of the second virtual measurement model based on the measurement data of the front end is called external influence data training.
[0110] Specifically, the collected equipment parameter data and the historical measurement data of a corresponding process station can be used as training data for the first virtual measurement model. The training process of the first virtual measurement model can be written as y1=F(X1), where X1 is the equipment parameter data of the process station corresponding to the manufacturing of a certain product, which is an M×N1 matrix, where M represents the number of wafers (sample size), and N1 represents the number of equipment parameters that can be collected at the process station (e.g., N1=34). Furthermore, the relationship between the measurement values and the equipment parameter data can be trained using correlation algorithms such as the XGBOOST algorithm and multinomial regression. The trained first virtual measurement model will predict a predicted value (the first virtual measurement mean u1) and a standard deviation (the first virtual measurement standard deviation). As can be seen, the first virtual measurement model essentially predicts a Gaussian distribution. Due to data errors, predicting a distribution can represent the characteristics of the data. Similarly, the collected upstream measurement data and the corresponding historical measurement data of the process station can be used as training data for the second virtual measurement model. The training process of the second virtual measurement model can be written as y2=F(X2), where X2 is the upstream measurement data of the process station corresponding to the manufacturing of the product, which is an M×N2 matrix, and N2 represents the number of process stations preceding this process station. Furthermore, the relationship between the measurement values and the upstream measurement data can be trained using correlation algorithms such as the XGBOOST algorithm and multinomial regression. The trained second virtual measurement model will predict a predicted value (the mean u2 of the second virtual measurement) and a standard deviation (the standard deviation of the second virtual measurement). It can be seen that the second virtual measurement model also predicts a Gaussian distribution.
[0111] It should be noted that, as those skilled in the art will understand, in practical applications, only normal sampling measurement data needs to be continuously fed into the sample pool for continuous model optimization and training. Even if the prediction data is very accurate, it will not be used as a training sample. It only needs to be sent into the monitoring database in real time and labeled with virtual prediction tags to facilitate subsequent data monitoring and root cause analysis.
[0112] Based on the same inventive concept, the present invention also provides an electronic device, please refer to... Figure 10 This is a block diagram of an electronic device provided in one embodiment of the present invention. Figure 10 As shown, the electronic device includes a processor 310 and a memory 330. The memory 330 stores a computer program, which, when executed by the processor 310, implements the virtual measurement method described above. Since the electronic device and the virtual measurement method provided by this invention belong to the same inventive concept, the electronic device provided by this invention possesses at least all the beneficial effects of the virtual measurement method provided by this invention. Therefore, the beneficial effects of the electronic device provided by this invention can be referred to the relevant descriptions of the beneficial effects of the virtual measurement method provided by this invention above, and will not be repeated here.
[0113] Please continue to refer to this. Figure 10 ,like Figure 10As shown, the electronic device also includes a communication interface 320 and a communication bus 340, wherein the processor 310, the communication interface 320, and the memory 330 communicate with each other through the communication bus 340. The communication bus 340 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus 340 can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is used to represent it in the figure, but this does not indicate that there is only one bus or one type of bus. The communication interface 320 is used for communication between the aforementioned electronic device and other devices.
[0114] It should be noted that the processor 310 referred to in this invention can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor 310 is the control center of the electronic device, connecting various parts of the entire electronic device through various interfaces and lines.
[0115] It should also be noted that the memory 330 can be used to store the computer program, and the processor 310 implements various functions of the electronic device by running or executing the computer program stored in the memory 330 and calling the data stored in the memory 330. The memory 330 may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable memory (PROM), electrically programmable memory (EPROM), electrically erasable programmable memory (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, random access memory is available in a variety of forms, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous random access memory (SDRAM), dual data rate synchronous random access memory (DDRSDRAM), enhanced synchronous random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), memory bus direct random access memory (RDRAM), direct memory bus dynamic random access memory (DRDRAM), and memory bus dynamic random access memory (RDRAM), etc.
[0116] The present invention also provides a readable storage medium storing a computer program, which, when executed by a processor, can implement the virtual measurement method described above. Since the readable storage medium and the virtual measurement method provided by the present invention belong to the same inventive concept, the readable storage medium provided by the present invention possesses at least all the beneficial effects of the virtual measurement method provided by the present invention. Therefore, the beneficial effects of the readable storage medium provided by the present invention can be referred to the relevant descriptions of the beneficial effects of the virtual measurement method provided by the present invention above, and will not be repeated here.
[0117] The readable storage medium provided by this invention can be any combination of one or more computer-readable media. The readable medium can be a computer-readable signal medium or a computer-readable storage medium. Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples (not exhaustive) of computer-readable storage media include: electrical connections having one or more wires, portable computer hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, apparatus, or device.
[0118] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of transmitting, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, etc., or any suitable combination thereof.
[0119] In summary, compared with the prior art, the virtual measurement method, electronic device, and readable storage medium provided by the present invention have the following beneficial effects:
[0120] Since front-end process site measurement data reflects the cumulative effects of the front-end process, this invention corrects the first virtual measurement data predicted based on equipment parameter data (including various sensor data and equipment operating status data generated during semiconductor equipment operation) by using second virtual measurement data predicted based on front-end process site measurement data. This fully utilizes data from different sources to capture more process variations, thereby effectively improving prediction accuracy. Furthermore, although the actual measurement values of the sampled wafers are only approximate reference values for other wafers, they are still very valuable. Therefore, this invention corrects the third virtual measurement data using sampled measurement data obtained from actual sampling. This allows for correction of prediction errors caused by unknown factors, effectively improving the accuracy of the final predicted measurement data. This also solves the problem that most prediction models are unaware of prediction errors caused by complex inputs.
[0121] It should be noted that computer program code for performing the operations of this invention can be written in one or more programming languages or a combination thereof. These programming languages include object-oriented programming languages such as Java, Smalltalk, and C++, as well as conventional procedural programming languages such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0122] It should be noted that the above description is merely a description of preferred embodiments of the present invention and is not intended to limit the scope of the present invention in any way. Any changes or modifications made by those skilled in the art based on the above disclosure are within the protection scope of the present invention. Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the present invention and its equivalents, the present invention also intends to include these modifications and variations.
Claims
1. A virtual measurement method, characterized in that, include: Acquire equipment parameter data, sampling measurement data, and upstream process station measurement data for the target process station. The upstream process station measurement data includes sampling measurement data for all process stations preceding the target process station. Based on the equipment parameter data of the target process station and the pre-trained first virtual measurement model, the first virtual measurement data of the target process station is obtained. Based on the measurement data of the preceding process station and the pre-trained second virtual measurement model, the second virtual measurement data of the target process station is obtained. The first virtual measurement data of the target process station is corrected based on the second virtual measurement data of the target process station to obtain the third virtual measurement data of the target process station. The third virtual measurement data of the target process station is corrected based on the sampling measurement data of the target process station to obtain the predicted measurement data of the target process station.
2. The virtual measurement method according to claim 1, characterized in that, The step of correcting the third virtual measurement data of the target process station based on the sampling measurement data of the target process station to obtain the predicted measurement data of the target process station includes: Based on the sampled measurement data of the target process station and the third virtual measurement data of the target process station corrected using the Kalman filter algorithm, the predicted measurement data of the target process station is obtained.
3. The virtual measurement method according to claim 2, characterized in that, The step of correcting the third virtual measurement data of the target process station based on the sampled measurement data of the target process station and using a Kalman filter algorithm to obtain the predicted measurement data of the target process station includes: The third virtual measurement data of the target process station is corrected according to the following Kalman filter formula to obtain the predicted measurement data of the target process station: In the formula, To predict measurement data, This is the third virtual measurement data, where K is the Kalman gain and C is the first hyperparameter. This is sampled measurement data.
4. The virtual measurement method according to claim 1, characterized in that, The sampling measurement data includes the sampling wafer measurement mean and the sampling wafer measurement standard deviation; the first virtual measurement data includes the first virtual measurement mean and the first virtual measurement standard deviation; the second virtual measurement data includes the second virtual measurement mean and the second virtual measurement standard deviation; the third virtual measurement data includes the third virtual measurement mean and the third virtual measurement standard deviation; and the predicted measurement data includes the predicted measurement mean and the predicted measurement standard deviation.
5. The virtual measurement method according to claim 4, characterized in that, The mean of the third virtual measurement is obtained using the following formula: In the formula, u p denoted as the mean of the third virtual measurement, u1 as the mean of the first virtual measurement, u2 as the mean of the second virtual measurement, and m as the second hyperparameter.
6. The virtual measurement method according to claim 4, characterized in that, The standard deviation of the third virtual measurement is obtained using the following formula: In the formula, Let x be the standard deviation of the third virtual measurement, n be the third hyperparameter, and x be the standard deviation of the third virtual measurement. i Let u be the measurement value corresponding to the i-th training sample participating in the training of the first virtual measurement model or the second virtual measurement model, u be the mean of the measurement values of all training samples participating in the training of the first virtual measurement model and the second virtual measurement model, and k be the total number of training samples participating in the training of the first virtual measurement model and the second virtual measurement model.
7. The virtual measurement method according to claim 1, characterized in that, The first virtual measurement model includes a first feature selection module and a first multivariate regression prediction module. The first feature selection module is configured to filter out first feature data related to measurement data prediction from the device parameter data. The first multivariate regression prediction module is configured to predict measurement data based on the first feature data to obtain the first virtual measurement data. The second virtual measurement model includes a second feature selection module and a second multivariate regression prediction module. The second feature selection module is configured to select second feature data related to measurement data prediction from the measurement data of the upstream process station. The second multivariate regression prediction module is configured to predict the measurement data based on the second feature data to obtain the second virtual measurement data.
8. The virtual measurement method according to claim 1, characterized in that, The method further includes: A similarity comparison is performed on the third virtual measurement data, the sampled measurement data, and the pre-acquired benchmark measurement data; If the similarity value of any two of the third virtual measurement data, the sampled measurement data, and the benchmark measurement data is less than a preset threshold, an alarm is triggered and the first virtual measurement model and the second virtual measurement model are updated.
9. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, which, when executed by the processor, implements the virtual measurement method according to any one of claims 1 to 8.
10. A readable storage medium, characterized in that, The readable storage medium stores a computer program, which, when executed by a processor, implements the virtual measurement method according to any one of claims 1 to 8.