Methods, systems, electronic devices and storage media for determining epitaxial film thickness
By identifying deposition steps in silicon epitaxial production and using a multiple linear regression model, the global slope coefficient and cavity intercept term are employed to predict film thickness, thus solving the problem of low reliability in epitaxial film thickness prediction and improving the stability and coverage of film thickness.
Patent Information
- Application Number
- CN202611400444.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-09
- Publication Date
- 2026-11-17
AI Technical Summary
In the silicon epitaxial production process, the prediction results of epitaxial film thickness are unreliable, and there is a lack of basis for judging the film thickness of unmeasured wafers, making it difficult to grasp the film thickness status of all wafers in a timely manner during the production process.
By identifying the deposition steps in the epitaxial process, process data of the epitaxial equipment is obtained, and a pre-constructed multiple linear regression model is used to predict film thickness. The global slope coefficient and the intercept term corresponding to the cavity are used to compensate for the differences in temperature field, flow field and hardware state, thereby achieving film thickness prediction.
This improves the reliability of film thickness prediction for unmeasured wafers, reduces the systematic bias of prediction results due to cavity differences, and ensures the stability and coverage of film thickness status.
Smart Images

Figure CN123116218A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of silicon epitaxial growth technology, specifically to a method, system, electronic device, and storage medium for determining epitaxial film thickness. Background Technology
[0002] In 12-inch silicon epitaxial growth, epitaxial film thickness is a crucial parameter affecting product quality and subsequent process windows. The production process typically relies on physical metrology equipment to sample and measure the film thickness of a subset of wafers, then uses the measurement results to determine the epitaxial process status. However, limitations in metrology equipment capacity, production cycle time, and measurement timeframes prevent film thickness measurement from covering every wafer. Unmeasured wafers lack corresponding film thickness assessment criteria, making it difficult to promptly grasp the film thickness status of all wafers during production.
[0003] Epitaxial wafer fabrication equipment generates process data related to its operational status. Different cavities performing the same or similar processes are also affected by factors such as temperature field, flow field, hardware status, and maintenance cycle. This data is typically distributed between the equipment's process data system and the actual film thickness measurement system. These systems may differ in recording formats, time bases, wafer identification, or batch identification, and there may be baseline offsets between different cavities, leading to low reliability of epitaxial film thickness prediction results. Summary of the Invention
[0004] This invention provides a method, system, electronic device, and storage medium for determining epitaxial film thickness, aiming to solve the problem of low reliability in predicting the epitaxial film thickness of wafers using process data from epitaxial equipment in silicon epitaxial production processes of related technologies.
[0005] In a first aspect, a method for determining epitaxial film thickness is provided, comprising the following steps: Identify the deposition step in the epitaxial process and acquire process data generated during the deposition step when the epitaxial equipment processes the wafer to be predicted, wherein the epitaxial equipment includes a reaction chamber; Based on the process data generated during the deposition step, the process characteristics of the wafer to be predicted are determined; The epitaxial film thickness of the wafer to be predicted is predicted by using a pre-constructed multiple linear regression model to analyze the process characteristics of the wafer to be predicted. The multiple linear regression model includes a global slope coefficient and an intercept term. The global slope coefficients of the multiple linear regression models corresponding to different reaction chambers are the same, and the intercept term corresponds to the reaction chamber.
[0006] Optionally, the process characteristics of the wafer to be predicted are determined based on the process data generated during the deposition step, including: Feature extraction is performed on the process data generated during the deposition step to obtain film thickness-related process parameters; The process characteristics of the wafer to be predicted are determined by statistical aggregation of the film thickness-related process parameters.
[0007] Optionally, the steps for constructing the multiple linear regression model include: Acquire process data and actual film thickness measurement data of the epitaxial equipment for wafer processing; Identify the deposition steps in the epitaxial process and generate corresponding wafer process features based on the process data within the deposition steps; The process characteristics are matched with the actual film thickness measurement data to obtain a matching sample; The multiple linear regression model is established based on the matched samples.
[0008] Optionally, the process characteristics are matched with the actual film thickness measurement data to obtain a matching sample, including: Obtain the cavity identifier, batch identifier, wafer identifier, slot information, and process time corresponding to the process characteristics; Obtain the batch identifier, wafer identifier, slot information, and measurement time corresponding to the actual film thickness measurement data; Based on batch identifier, wafer identifier, and slot information, the process characteristics are initially matched with the actual film thickness measurement data to obtain the initial matching result; The initial matching result is verified based on the cavity identifier and the process time. Wafers that pass the verification and simultaneously possess complete process characteristics and reliable actual film thickness measurement data are identified as matching samples. The complete process characteristics are process characteristics without missing data, and the reliable actual film thickness measurement data are actual film thickness measurement data that have not been judged as measurement anomalies.
[0009] Optionally, the actual film thickness measurement data is the average film thickness of the wafer measured at multiple points.
[0010] Optionally, establishing the multiple linear regression model based on the matched samples includes: Within the same process menu or the same process family, the global slope coefficient and the intercept term corresponding to each of the reaction chambers are trained based on matching samples from multiple reaction chambers. The same process family refers to a set of process menus where the ratio of main hydrogen to slit hydrogen is the same.
[0011] Optionally, the method further includes: The wafer with complete process characteristics but without actual film thickness measurement data is identified as the wafer to be predicted. The complete process characteristics are process characteristics without missing data.
[0012] Optionally, the method further includes: When the newly added reaction chamber of the epitaxial device operates the same process menu or meets the first slope reuse condition, the global slope coefficient in the multiple linear regression model is fixed. The first slope reuse condition is that the average absolute error obtained by verifying the model with shared global slope coefficients is less than a preset error threshold. Under the condition of fixing the global slope coefficient, the intercept term corresponding to the newly added reaction chamber is learned using the matching samples corresponding to the newly added reaction chamber; Based on the global slope coefficient and the intercept term corresponding to the newly added reaction cavity, the epitaxial film thickness of the wafer to be predicted in the newly added reaction cavity is predicted to obtain the predicted film thickness of the newly added reaction cavity.
[0013] Optionally, the method further includes: The predicted film thickness of the newly added reaction chamber was verified using validation samples. If the preset error index corresponding to the predicted film thickness is less than or equal to the preset error threshold, the newly added reaction chamber will be included in the prediction range.
[0014] Optionally, the method further includes: Obtain the process information and target film thickness of the new product. The process information includes at least one of the process menu, core gas settings, power ratio, and flow field settings. The process information of the newly added product is compared with the process information of the existing product to determine whether the newly added product meets the second slope reuse condition. The second slope reuse condition includes that the process information of the newly added product and the existing product meet a preset similarity condition. If the newly added product meets the second slope reuse condition, the global slope coefficient in the multiple linear regression model is reused. Multiple wafer samples with complete process characteristics and reliable actual film thickness measurement data are obtained from the new product, and an intercept term corresponding to the new product and its reaction chamber is determined based on each wafer sample; the complete process characteristics are process characteristics without missing data, and the reliable actual film thickness measurement data are actual film thickness measurement data that have not been judged as measurement abnormalities. Based on the global slope coefficient and the intercept term corresponding to the new product, the epitaxial film thickness of the wafer to be predicted in the new product is predicted.
[0015] Secondly, an epitaxial film thickness determination system is also provided, comprising: A deposition step identification module is used to identify deposition steps in an epitaxial process and acquire process data generated during the deposition steps when the epitaxial equipment processes the wafer to be predicted. The epitaxial equipment includes a reaction chamber. The process feature determination module is used to determine the process features of the wafer to be predicted based on the process data generated during the deposition step. The film thickness prediction module is used to predict the epitaxial film thickness of the wafer to be predicted by using a pre-constructed multiple linear regression model to analyze the process characteristics of the wafer to be predicted, thereby obtaining the predicted film thickness of the wafer to be predicted. The multiple linear regression model includes a global slope coefficient and an intercept term, and the global slope coefficient of the multiple linear regression model corresponding to different reaction chambers is the same, while the intercept term corresponds to the reaction chamber.
[0016] Thirdly, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the method described in any of the preceding claims.
[0017] Fourthly, a computer-readable storage medium is also provided, on which a computer program is stored, said computer program being loaded by a processor to perform the steps of any of the methods described above.
[0018] Beneficial effects: This application enables the generation of process features from process data during epitaxial processing of the wafer to be predicted, even when the actual film thickness measurement of the wafer has not been performed. These process features are then input into a pre-constructed multiple linear regression model to obtain the predicted film thickness. This addresses the issue that sampling measurements cannot cover every wafer, allowing for the estimation of the epitaxial film thickness. Furthermore, since the multiple linear regression model uses a global slope coefficient shared by multiple cavities and an intercept term corresponding to each cavity, the global slope coefficient characterizes the relatively stable process influence relationship between multiple cavities, and the intercept term compensates for the film thickness baseline shift caused by differences in temperature field, flow field, hardware condition, or maintenance cycle in different cavities. Therefore, while utilizing the commonalities of multi-cavity data to improve model stability, it reduces the systematic bias caused by cavity differences in the film thickness prediction results, thereby improving the reliability of predicting the epitaxial film thickness of the wafer to be predicted. Attached Figure Description
[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0020] Figure 1This is a schematic flowchart of an epitaxial film thickness determination method provided by an exemplary embodiment of this disclosure; Figure 2 This is a schematic diagram of a wafer-level matching process between process features and actual film thickness measurement data provided by an exemplary embodiment of this disclosure; Figure 3 This is a schematic diagram of a multiple linear regression model structure provided by an exemplary embodiment of this disclosure; Figure 4 This is a schematic diagram of the horizontal expansion of a multiple linear regression model corresponding to a newly added reaction chamber provided by an exemplary embodiment of this disclosure; Figure 5 This is a schematic diagram of a model reuse and independent modeling judgment process for a new product provided by an exemplary embodiment of this disclosure; Figure 6 This is a schematic diagram of the functional modules of an epitaxial film thickness determination system provided by an exemplary embodiment of this disclosure; Figure 7 This is a schematic diagram of the structure of an electronic device provided by an exemplary embodiment of this disclosure. Detailed Implementation
[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0022] In the description of this application, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.
[0023] "A and / or B" includes the following three combinations: A only, B only, and a combination of A and B.
[0024] The use of "applies to" or "configured to" in this application implies open and inclusive language, which does not exclude the applicability to or configuration to devices performing additional tasks or steps. Additionally, the use of "based on" implies openness and inclusivity, because processes, steps, calculations, or other actions "based on" one or more of the stated conditions or values may in practice be based on additional conditions or values beyond those stated.
[0025] In this application, the term "exemplary" is used to mean "serving as an example, illustration, or description." Any embodiment described as "exemplary" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes are not described in detail to avoid obscuring the description of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application. In this application, FDC refers to Fault Detection and Classification, SPC refers to Statistical Process Control, and TCS refers to Trichlorosilane.
[0026] Firstly, this embodiment provides a method for determining epitaxial film thickness; please refer to [link to relevant documentation]. Figure 1 This includes the following steps: Step S100: Identify the deposition step in the epitaxial process and obtain the process data generated in the deposition step during the processing of the wafer to be predicted by the epitaxial equipment. The epitaxial equipment includes a reaction chamber.
[0027] Specifically, during the epitaxial processing of the wafer to be predicted, the epitaxial equipment identifies the deposition steps in the epitaxial process and acquires the process data generated within the deposition steps from the data acquisition system of the epitaxial equipment. This process data refers to the process operation data generated during wafer processing within the reaction chamber. The processing record corresponding to the wafer to be predicted can be located based on wafer identification, batch identification, slot information, reaction chamber identification, and processing time. Subsequently, data generated within the deposition steps is extracted from this processing record, avoiding the mixing of data from heating, wafer transfer, purging, or other non-deposition stages into the film thickness prediction process. The purpose of identifying the deposition steps in the epitaxial process is to pinpoint the actual process stage where epitaxial film deposition occurs from the wafer processing process data, so as to generate process characteristics based only on the process data directly related to film thickness formation within that stage.
[0028] Step S200: Determine the process characteristics of the wafer to be predicted based on the process data generated during the deposition step.
[0029] Specifically, the process data within the deposition step undergoes preprocessing and characterization. This process data is typically a data sequence acquired continuously or periodically over time. First, obviously outlier data points are removed, and the units from different data sources or acquisition channels are standardized. Then, process parameters related to the epitaxial film thickness are extracted from the process data within the deposition step, and these parameters are converted into wafer-level process features suitable for input into a multiple linear regression model. Each process feature corresponds one-to-one with the wafer to be predicted, characterizing the actual process state of the wafer during deposition.
[0030] Step S300: The epitaxial film thickness of the wafer to be predicted is predicted using a pre-constructed multiple linear regression model based on the process characteristics of the wafer to be predicted. The predicted film thickness of the wafer is obtained by using this model. The multiple linear regression model includes a global slope coefficient and an intercept term. The global slope coefficient is the same for different reaction chambers, and the intercept term corresponds to the reaction chamber.
[0031] Specifically, the process characteristics of the wafer to be predicted are input into a pre-constructed multiple linear regression model. This model calls upon a global slope coefficient shared by multiple reaction chambers to characterize the common influence of process characteristics on epitaxial film thickness under the same or similar process conditions. Simultaneously, based on the reaction chamber to which the wafer belongs, the intercept term corresponding to that chamber is invoked. This intercept term is used to compensate for the baseline shift in film thickness caused by differences in temperature field, flow field, hardware condition, or maintenance condition between different reaction chambers. Based on the process characteristics, global slope coefficient, and corresponding intercept term, the multiple linear regression model outputs the predicted film thickness of the wafer to be predicted.
[0032] This embodiment acquires process data generated by the wafer to be predicted during the deposition step and uses a multiple linear regression model containing a shared global slope coefficient and an intercept term corresponding to the reaction chamber to predict the film thickness. This allows for the prediction of the film thickness without actual film thickness measurement of the wafer to be predicted, thus addressing the problem that sampling measurement cannot cover all wafers. At the same time, by learning the common process influence laws among multiple reaction chambers through the shared global slope coefficient and compensating for the fixed baseline differences of each reaction chamber through the intercept term, the reliability and adaptability of the virtual measurement results of epitaxial film thickness can be improved.
[0033] In some embodiments, step S200, determining the process characteristics of the wafer to be predicted based on process data generated during the deposition step, includes: Step S210: Extract features from the process data generated during the deposition step to obtain film thickness-related process parameters.
[0034] Specifically, process parameters related to epitaxial film thickness variations are screened from the process data generated during the deposition step. Film thickness-related process parameters may include deposition time, reaction chamber pressure, temperature, reactant gas flow rate, reactant gas pressure, and other process parameters related to the epitaxial deposition state. The purpose of feature extraction is to select data items that reflect the film thickness formation process from the large amount of process data recorded by the epitaxial equipment, reducing the interference of data with weak correlation to film thickness prediction on the input of the multiple linear regression model.
[0035] Step S220: Statistically aggregate the film thickness-related process parameters to determine the process characteristics of the wafer to be predicted.
[0036] Specifically, film thickness-related process parameters are aggregated according to the wafer to be predicted. For process parameters that change over time, the mean of the data sequence within the deposition step can be statistically analyzed to obtain the statistical characteristic value corresponding to the wafer to be predicted. For parameters representing the deposition persistence state or process setting state, their stable values or statistical values within the deposition step can be extracted. Through this processing, each wafer to be predicted forms a set of process characteristics with fixed dimensions, which can then be input into a multiple linear regression model for film thickness prediction.
[0037] This embodiment first extracts film thickness-related process parameters and then statistically aggregates the extracted process parameters. This allows the continuously changing process data within the deposition step to be converted into wafer-level multiple linear regression model input data, providing a stable and unified data foundation for epitaxial film thickness prediction and reducing the impact of non-deposition stage data or irrelevant process parameters on the prediction results.
[0038] In some embodiments, the step of constructing the multiple linear regression model in step S300 includes: Step M100: Obtain process data and actual film thickness measurement data of the epitaxial equipment processing the wafer sample.
[0039] Specifically, process data generated during the epitaxial processing of historical or training wafer samples is acquired, along with corresponding actual film thickness measurement data. The process data can originate from the epitaxial equipment or its process data system. The actual film thickness measurement data comes from film thickness measurement equipment or its measurement data system. By simultaneously acquiring both process data and actual film thickness measurement data, input features and target outputs can be provided for training the multiple linear regression model.
[0040] Step M200: Identify the deposition steps in the epitaxial process and generate the corresponding wafer process features based on the process data within the deposition steps.
[0041] Specifically, the deposition steps in the epitaxial process are identified from the processing records of wafer samples, and the process data generated within these deposition steps is extracted. Subsequently, the process data within the deposition steps is cleaned, units are standardized, features are extracted, and statistically aggregated to form the process features of the corresponding wafer sample. The process features of each wafer sample are used to characterize the process state of that wafer sample during the deposition stage.
[0042] Step M300: Match the process characteristics with the actual film thickness measurement data to obtain a matching sample.
[0043] Specifically, a wafer-level correspondence is established between the process characteristics of wafer samples and their corresponding actual film thickness measurement data. When the process characteristics and actual film thickness measurement data match, information such as batch identifier, wafer identifier, slot information, reaction chamber identifier, process processing time, and measurement time can be used to determine the process characteristics and actual film thickness measurement data of the same wafer. After successful matching, wafers with complete data and reliable measurements are further screened as matching samples for training the multiple linear regression model.
[0044] Step M400: Establish a multiple linear regression model based on the matched samples.
[0045] Specifically, the process features in the matching samples are used as input to the multiple linear regression model, and the actual film thickness measurement data in the matching samples are used as the output target of the multiple linear regression model. During training, the model learns the global slope coefficients shared by multiple reaction chambers and learns corresponding intercept terms for different reaction chambers. After training, the multiple linear regression model can output the corresponding predicted film thickness based on the input wafer-level process features and the reaction chamber to which the wafer belongs.
[0046] This embodiment reduces training sample mismatch caused by inconsistencies in the recording of process data between different equipment process data systems by matching process data and actual film thickness measurement data at the wafer level before establishing a multiple linear regression model. At the same time, training a multiple linear regression model based on the matched samples, which includes a global slope coefficient and a reaction chamber intercept term, can take into account both the common process rules of multiple reaction chambers and the fixed film thickness offset between reaction chambers.
[0047] In some embodiments, step M300 involves matching process characteristics with actual film thickness measurement data to obtain a matching sample, including: Step M310: Obtain the cavity identifier, batch identifier, wafer identifier, slot information, and process time corresponding to the process characteristics.
[0048] Specifically, when generating the process characteristics of a wafer sample, the corresponding identity information and process time information are simultaneously read. The cavity identifier indicates the reaction cavity in which the wafer sample is processed. The batch identifier, wafer identifier, and slot information are used to locate the wafer sample's identity within the production batch. The process time indicates the time interval or point in time during which the wafer sample is processed in the epitaxial equipment.
[0049] Step M320: Obtain the batch identifier, wafer identifier, slot information, and measurement time corresponding to the actual film thickness measurement data.
[0050] Specifically, the identification information and measurement time information associated with the measurement results are read from the actual film thickness measurement data. Batch identifier, wafer identifier, and slot information are used to locate the measurement results to the corresponding wafer. Measurement time is used to indicate the time when the actual film thickness measurement data was generated, so as to verify its chronological relationship with the process processing time or time window.
[0051] Step M330: Based on batch identifier, wafer identifier and slot information, perform initial matching between process characteristics and actual film thickness measurement data to obtain initial matching results.
[0052] Specifically, wafer matching is performed based on batch identifier, wafer identifier, and slot information. If the batch identifier, wafer identifier, and slot information corresponding to the process characteristics match the batch identifier, wafer identifier, and slot information corresponding to the actual film thickness measurement data, the two are considered candidate correspondences, forming an initial matching result. This initial matching result is used to preliminarily determine whether the process characteristics and the actual film thickness measurement data belong to the same wafer.
[0053] Step M340: Verify the initial matching results based on the cavity identification and process time.
[0054] Specifically, the initial matching results are verified using cavity identification and process processing time. For candidate correspondences of the same wafer, it is determined whether the process characteristics originate from the reaction cavity where the wafer is actually processed, and whether the measurement time of the actual film thickness measurement data meets the preset time relationship with the process processing time. For example, the measurement time should be later than the corresponding process processing time, and the time interval between the two should be within a reasonable range. This verification can reduce the probability of incorrect matching caused by batch, slot, or wafer identification anomalies.
[0055] Step M350: Wafers that pass the verification and simultaneously have complete process characteristics and reliable actual film thickness measurement data are identified as matching samples. Complete process characteristics are process characteristics without missing data, and reliable actual film thickness measurement data are actual film thickness measurement data that have not been judged as measurement abnormalities.
[0056] Specifically, the wafers that pass the verification process undergo further data validity assessment. If the wafer's process features are complete and its actual film thickness measurement data is not marked as an anomaly, then the wafer is identified as a matching sample. If the process features are missing, or the actual film thickness measurement data is determined to be an anomaly, then the wafer is not used as a training sample for the multiple linear regression model. The resulting matching samples thus possess both reliable input features and credible output labels.
[0057] This embodiment improves the matching accuracy between process features and actual film thickness measurement data by initial matching of identity information, cavity and time verification, and screening of data integrity and measurement reliability. It also reduces the number of mismatched and abnormal samples entering the training process of the multiple linear regression model, thereby improving the training quality and prediction reliability of the multiple linear regression model.
[0058] Please see Figure 2 , Figure 2The diagram illustrates a wafer-level matching process between process features and actual film thickness measurement data. Specifically, it includes: first, acquiring process data from the epitaxial equipment side, and extracting data segments corresponding to the wafer deposition process based on the deposition steps in the epitaxial process menu; then, performing feature extraction and statistical aggregation on these data segments to obtain wafer-level process features, generating a process feature record containing reaction chamber identifier, batch identifier, wafer identifier, slot information, and process processing time. Alternatively, actual film thickness measurement data can be obtained from the measurement side, and the corresponding batch identifier, wafer identifier, slot information, and measurement time can be extracted. Afterwards, initial matching is performed based on the batch identifier, wafer identifier, and slot information, and the initial matching result is verified using the reaction chamber identifier and process processing time. If the verification passes and the corresponding wafer possesses both complete process characteristics and reliable actual film thickness measurement data, then the wafer is identified as a matching sample. If the corresponding wafer has complete process characteristics but lacks actual film thickness measurement data, then the wafer is retained as a wafer to be predicted. If the process characteristics of the corresponding wafer are missing or the actual film thickness measurement data is determined to be abnormal, then the wafer is not used as a matching sample. Through the process shown in the figure, a wafer-level correspondence can be established even when the equipment process data and measurement data come from different epitaxial film thickness determination systems, thereby reducing the impact of training sample mismatch on the reliability of the multiple linear regression model.
[0059] In some embodiments, the actual film thickness measurement data in step M100 is the average film thickness of the wafer measured at multiple points.
[0060] Specifically, when obtaining actual film thickness measurement data for a wafer sample, the average value of multi-point film thickness test results can be used as the actual film thickness measurement data for that wafer sample. Multi-point testing can cover multiple locations on the wafer surface to reflect the overall film thickness level of the wafer. The average film thickness from the multi-point tests is used as the target value during the training of the multiple linear regression model, establishing a correspondence between the features learned by the multiple linear regression model and the overall film thickness of the wafer.
[0061] This embodiment uses the average film thickness from multiple tests on the wafer as the actual film thickness measurement data, which can reduce the impact of differences in the location of single-point measurements on the training target value of the multiple linear regression model, making the output of the multiple linear regression model more suitable for characterizing the overall film thickness level of the wafer, thereby improving the stability of the epitaxial film thickness prediction results.
[0062] In some embodiments, step M400, which establishes a multiple linear regression model based on matched samples, includes: Step M410: Within the same process menu or the same process family, train the global slope coefficient and the intercept term corresponding to each reaction chamber based on the matching samples of multiple reaction chambers; Among them, the process menu set with the same ratio of main hydrogen to slit hydrogen is the same for the same process family.
[0063] Specifically, the matching samples are first grouped according to the process menu or process family. For matching samples under the same process menu, or matching samples belonging to the same process family with the same ratio of main hydrogen to slit hydrogen, the matching samples from multiple reaction chambers are used together for training the multiple linear regression model. During the training of the multiple linear regression model, each reaction chamber shares the same global slope coefficient to characterize the common influence of process characteristics on film thickness; at the same time, intercept terms are trained separately for different reaction chambers to characterize the differences in film thickness baselines between different reaction chambers.
[0064] This embodiment improves the stability of slope coefficient training by sharing the global slope coefficient within the same process menu or process family and training the intercept term for each reaction chamber separately. It also compensates for the fixed offset of different reaction chambers by using the intercept term, avoiding insufficient samples due to independent modeling of each reaction chamber and avoiding systematic bias caused by multiple reaction chambers sharing the same multiple linear regression model.
[0065] Please see Figure 3 , Figure 3 The diagram illustrates a multivariate linear regression model. Matching samples from multiple reaction chambers within the same process menu or process family are acquired, and the process characteristics of each matching sample are used as input to the multivariate linear regression model. Actual film thickness measurements are used as the output target of the model. During training, the multivariate linear regression models for different reaction chambers use the same global slope coefficient to represent the combined influence of process characteristics such as deposition time, reaction chamber pressure, temperature, reactant gas flow rate, and reactant gas pressure on epitaxial film thickness. Simultaneously, different reaction chambers are configured with corresponding intercept terms to represent the differences in film thickness baselines caused by variations in temperature field, flow field, hardware condition, or maintenance status.
[0066] The multiple linear regression model corresponding to a certain cavity can be expressed as: , in, The first shared by multiple cavities A global slope coefficient, For the first Each type of Fault Detection and Classification (FDC) feature or Statistical Process Control (SPC) feature, This is the intercept specific to this cavity.
[0067] When predicting film thickness, the corresponding intercept term is called according to the reaction chamber to which the wafer to be predicted belongs, and the predicted film thickness is obtained by combining the shared global slope coefficient and the process characteristics of the wafer to be predicted. Through the multiple linear regression model structure shown in the figure, the slope training stability can be improved by using multiple reaction chamber samples, while the fixed offset between reaction chambers can be compensated by the intercept term.
[0068] For example, suppose the epitaxial device includes cavity A, cavity B, and cavity C. Training samples are constructed for each of these three cavities as inputs to their respective multiple linear regression models. The shared global slopes corresponding to process characteristics such as deposition time, reaction chamber pressure, cavity bottom temperature, reactant gas TCS flow rate, and reactant gas TCS pressure for these three cavities are respectively represented as: , , , , The specific intercepts for cavities A, B, and C are respectively expressed as: , and .
[0069] For example, the multiple linear regression model corresponding to cavity A can be expressed as: , in, Indicates the first shared by multiple cavities The global slope coefficient of the process feature, at this time, .
[0070] Similarly, the only difference between the multiple linear regression models for cavities B and C and those for cavity A is the specific intercept. It should be noted that while multiple cavities share the process influence rules, different cavities absorb fixed-stage line offsets through independent specific intercepts.
[0071] In some embodiments, the above-described method for determining epitaxial film thickness further includes step S400, which is as follows: Step S400: The wafer with complete process characteristics but without actual film thickness measurement data is identified as the wafer to be predicted. Complete process characteristics are process characteristics without missing data.
[0072] Specifically, the data status of wafers that have completed epitaxial processing is assessed. If a wafer has complete process characteristics but lacks corresponding actual film thickness measurement data, it indicates that the wafer has the input data required for prediction by the multiple linear regression model, but lacks actual measurement results. This wafer is identified as the wafer to be predicted, and the pre-built multiple linear regression model is used to predict its epitaxial film thickness. For wafers with missing process characteristics, they may not be identified as wafers to be predicted, or they may be marked as wafers for which data needs to be supplemented.
[0073] This embodiment identifies wafers with complete process characteristics but no actual film thickness measurement data as wafers to be predicted, thereby clarifying the applicable objects of virtual measurement and enabling wafers with complete process data that have not been sampled to obtain predicted film thickness, thus improving wafer-level film thickness status coverage.
[0074] In some embodiments, the above-described method for determining epitaxial film thickness further includes step S500, specifically steps S510 to S530.
[0075] Step S510: When the newly added reaction chamber of the epitaxial equipment operates with the same process menu or meets the first slope reuse condition, fix the global slope coefficient in the multiple linear regression model. The first slope reuse condition is that the average absolute error obtained by the verification using the multiple linear regression model with shared global slope coefficients is less than the preset error threshold.
[0076] Specifically, when a new reaction chamber in an epitaxial device needs to be included in the epitaxial film thickness prediction range, the first step is to determine whether the new reaction chamber operates on the same process menu as the mature reaction chamber, or whether it meets the first slope reuse condition. If the mean absolute error obtained by validating the new reaction chamber using a multiple linear regression model with shared global slope coefficients is less than a preset error threshold, then the new reaction chamber is considered to be able to reuse the global slope coefficients from the mature multiple linear regression model. In this case, the global slope coefficients in the mature multiple linear regression model are fixed, and the global slope coefficients are not retrained.
[0077] Step S520: Under the condition of fixing the global slope coefficient, learn the intercept term corresponding to the newly added reaction chamber using the matching samples corresponding to the newly added reaction chamber.
[0078] Specifically, matching samples are obtained for the newly added reaction chamber. These matching samples are wafer samples fabricated in the new reaction chamber that simultaneously possess complete process characteristics and reliable actual film thickness measurement data. With a fixed global slope coefficient, the intercept term corresponding to the new reaction chamber is learned using these matching samples. This intercept term is used to compensate for the difference in film thickness baseline between the new reaction chamber and the mature reaction chamber.
[0079] Step S530: Based on the global slope coefficient and the intercept term corresponding to the newly added reaction cavity, predict the epitaxial film thickness of the wafer to be predicted in the newly added reaction cavity to obtain the predicted film thickness of the newly added reaction cavity.
[0080] Specifically, for the wafer to be predicted in the newly added reaction chamber, the process data generated during the deposition step is acquired, and corresponding process features are generated. Subsequently, these process features are input into a multiple linear regression model containing a fixed global slope coefficient and an intercept term corresponding to the newly added reaction chamber, to obtain the predicted film thickness of the wafer. Thus, the newly added reaction chamber can enter the film thickness prediction process without retraining all model parameters.
[0081] This embodiment fixes the global slope coefficients in a mature multiple linear regression model when the new reaction chamber meets the slope reuse condition, and learns the corresponding intercept term using only the matching samples of the new reaction chamber. This reduces the number of actual film thickness measurement samples required to import the multiple linear regression model corresponding to the new reaction chamber, shortens the model spanning period, and compensates for the baseline offset of the new reaction chamber through the intercept term.
[0082] In some embodiments, the above-described method for determining epitaxial film thickness further includes step S600, specifically steps S610 to S620.
[0083] Step S610: Verify the error of the predicted film thickness of the newly added reaction chamber using the verification sample.
[0084] Specifically, validation samples were selected from the newly added reaction chamber. These validation samples possess complete process characteristics and reliable actual film thickness measurements. A multiple linear regression model, including a fixed global slope coefficient and an intercept term corresponding to the newly added reaction chamber, was used to predict the film thickness of the validation samples. Subsequently, the predicted film thickness of the validation samples was compared with their actual film thickness measurements, and a preset error index was calculated to evaluate the accuracy of the prediction results for the newly added reaction chamber.
[0085] Step S620: If the preset error index corresponding to the predicted film thickness is less than or equal to the preset error threshold, the newly added reaction chamber is included in the prediction range.
[0086] Specifically, it is determined whether the preset error index is less than or equal to the preset error threshold, where the preset error index refers to the deviation between the predicted film thickness and the actual film thickness. If the determination result is yes, it means that the prediction error of the newly added reaction chamber under the current fixed global slope coefficient and corresponding intercept term meets the requirements. The newly added reaction chamber is included in the virtual measurement prediction range, so that the wafer to be predicted, processed in the newly added reaction chamber and having complete process characteristics, can obtain the predicted film thickness through a multiple linear regression model. If the determination result is no, validation samples or training samples can be added to relearn the intercept term, or a model can be independently built for the newly added reaction chamber.
[0087] This embodiment verifies the predicted film thickness of the newly added reaction chamber by verifying the error, and only includes it in the prediction range after the error meets the threshold requirement. This reduces the risk of prediction deviation caused by inappropriate reuse of mature multiple linear regression models and improves the cross-sectional reliability of the multiple linear regression model corresponding to the newly added reaction chamber.
[0088] Please see Figure 4 , Figure 4 The diagram illustrates a horizontal flow chart of a multiple linear regression model corresponding to a newly added reaction chamber. When a new reaction chamber is present in the epitaxial equipment, it is first determined whether the new reaction chamber operates on the same process menu or whether it meets the first slope reuse condition. If the condition is met, the global slope coefficients in the mature multiple linear regression model are fixed, and matching samples corresponding to the new reaction chamber are obtained. Matching samples can be wafer samples processed in the new reaction chamber that simultaneously possess complete process characteristics and reliable actual film thickness measurement data. Subsequently, under the condition of fixed global slope coefficients, the intercept term corresponding to the new reaction chamber is learned using the matching samples. Based on the fixed global slope coefficients and the intercept term corresponding to the new reaction chamber, the epitaxial film thickness of the wafer to be predicted in the new reaction chamber is predicted. Validation samples can also be used to verify the error of the predicted film thickness of the new reaction chamber. If the error verification is successful, the new reaction chamber is included in the prediction range; if the error verification fails, matching samples for the new reaction chamber are supplemented and the intercept term is relearned, or a separate multiple linear regression model is established for the new reaction chamber. The process shown in the figure allows for the rapid expansion of a new reaction chamber without retraining all model parameters.
[0089] In some embodiments, the above-described method for determining epitaxial film thickness further includes step S700, specifically steps S710 to S750.
[0090] Step S710: Obtain the process information and target film thickness of the new product. The process information includes at least one of the following: process menu, core gas setting, power ratio, and flow field setting.
[0091] Specifically, when a new product is introduced into the epitaxial production process, the corresponding process information and target film thickness for that new product are obtained. The process information can originate from process menus, production recipes, or engineering configuration data. For example, the process information may include at least one of the following: target film thickness, process menu, core gas settings, power ratio, and flow field settings. The process menu refers to the process recipe or procedure called when the epitaxial equipment performs the epitaxial growth of a specific product, specifying process control conditions such as deposition steps, temperature, pressure, gas flow rate, and power. The core gas settings refer to the settings of the reactive gas and carrier gas that have a major impact on the film thickness or deposition rate during the epitaxial deposition process, such as the type, flow rate, or proportion of gases like trichlorosilane. The power ratio refers to the power distribution ratio between different heating zones, power channels, or temperature control areas in the epitaxial equipment, used to influence the temperature field distribution within the reaction chamber. The flow field settings refer to the process settings related to the gas flow state within the reaction chamber, such as the main hydrogen, slit hydrogen, and inner / outer ring gas flow ratio, used to influence the distribution of reactive gases on the wafer surface. The target film thickness characterizes the desired epitaxial layer thickness for the new product. By acquiring process information and target film thickness, we can obtain the basic information needed to determine whether a new product can reuse the mature multiple linear regression model corresponding to an existing product.
[0092] Step S720: Compare the process information of the new product with the process information of the existing product to determine whether the new product meets the second slope reuse condition. The second slope reuse condition includes that the process information of the new product and the existing product meet the preset similarity condition.
[0093] Specifically, the target film thickness, process menu, core gas settings, power ratio, flow field settings, or other process information of the new product are compared with the corresponding process information of the existing product. If the two meet the preset similarity conditions, it indicates that the new product and the existing product are similar in the main process conditions affecting film thickness changes. Based on this, it is determined that the new product meets the second slope reuse condition. If the two differ significantly, it can be determined that the new product does not meet the second slope reuse condition and is transferred to the independent modeling process. The preset similarity condition means that the process menu of the new product and the existing product are the same or similar, and at least one of the core gas settings, power ratio, and flow field settings is the same or the deviation is within the preset allowable range, so that the slope of the film thickness influence factor can be reused. It is easy to understand that in the actual modeling process, models need to be built independently for different types of core gases, and the allowable deviation ranges for different types of core gases may be different.
[0094] Step S730: If the new product meets the second slope reuse condition, reuse the global slope coefficient in the multiple linear regression model.
[0095] Specifically, if a new product is determined to meet the second slope reuse condition, the global slope coefficient in the mature multiple linear regression model corresponding to the existing product is fixed. This global slope coefficient is used to represent the influence of process characteristics on film thickness. When reusing this global slope coefficient, the global slope coefficient is not retrained; instead, the validated stable slope relationship in the mature multiple linear regression model corresponding to the existing product is used.
[0096] Step S740: Obtain multiple wafer samples from the new product that simultaneously possess complete process characteristics and reliable actual film thickness measurement data, and determine the intercept term corresponding to the new product and its reaction chamber based on each wafer sample; complete process characteristics are process characteristics without missing data, and reliable actual film thickness measurement data are actual film thickness measurement data that have not been judged as measurement anomalies.
[0097] Specifically, process data from multiple wafer samples in the new product are collected, and corresponding process features are generated. Simultaneously, actual film thickness measurement data for these wafer samples are acquired. Wafer samples possessing both complete process features and reliable actual film thickness measurement data are selected and used as sample data for the new product. Under a fixed global slope coefficient, the intercept term corresponding to the new product and its reaction chamber is determined based on these wafer samples. This intercept term is used to compensate for the baseline difference in film thickness of the new product within the corresponding reaction chamber. For example, the intercept term can be determined by the average of the difference between the reliable actual film thickness measurement data and the fixed slope calculation result, or by a robust statistical value. Specifically, the process involves: substituting the process characteristics of the wafer sample into the fixed slope portion to calculate its film thickness prediction component without considering the newly added intercept correction term; subtracting the film thickness prediction component from the reliable actual film thickness measurement data of the wafer sample to obtain the corresponding residual value; calculating the residual values for multiple wafer samples under the same new product and the same reaction chamber, and averaging the residual values, or calculating robust statistical values such as the median, truncated mean, or weighted average after removing abnormal residuals; using the obtained statistical value as the intercept term or intercept correction term corresponding to the new product and its reaction chamber to compensate for the film thickness baseline offset that the fixed slope portion cannot reflect. It is worth noting that the above intercept term estimation process belongs to the parameter update method in the model application stage or incremental adaptation stage and does not affect the modeling method of jointly training the global slope coefficient and the existing intercept term in the basic model.
[0098] Step S750: Based on the global slope coefficient and the intercept term corresponding to the new product, predict the epitaxial film thickness of the wafer to be predicted in the new product.
[0099] Specifically, for the wafer to be predicted in the new product, process data during the deposition step of the wafer's processing is obtained, and the corresponding process characteristics are determined. Then, these process characteristics are input into a multiple linear regression model that reuses the global slope coefficient and includes the intercept term corresponding to the new product, to obtain the predicted film thickness of the wafer to be predicted in the new product. If the prediction error of the new product does not meet the requirements, the reuse of the global slope coefficient can be stopped, and an independent multiple linear regression model can be established for the new product.
[0100] This embodiment determines whether the slope reuse condition is met by comparing the process information of the new product with that of the existing product. If the slope reuse condition is met, the global slope coefficient in the mature multiple linear regression model is reused. The corresponding intercept term is determined only based on the new product sample, which can reduce the actual film thickness measurement sample requirements and the training cost of the multiple linear regression model during the introduction of the new product. At the same time, by determining the intercept term for the new product and its reaction chamber, the difference in film thickness baseline under the combination of new product and reaction chamber can be compensated, thereby improving the online efficiency and predictive adaptability of the virtual measurement model for the new product.
[0101] Please see Figure 5 , Figure 5 The diagram illustrates a process for reusing and independently modeling a new product. During the introduction of a new product into the epitaxial production process, the process information and target film thickness of the new product are acquired. This information is then compared with the process information of existing products to determine if the new product meets the second slope reuse condition. If the process information of the new product and existing products meets a preset similarity condition, the global slope coefficient from a mature multiple linear regression model is reused, and multiple wafer samples with complete process characteristics and reliable actual film thickness measurement data are obtained for the new product. Subsequently, under a fixed global slope coefficient, the intercept term corresponding to the new product and its reaction chamber is determined based on the multiple wafer samples, and the epitaxial film thickness of the wafers to be predicted in the new product is predicted based on this intercept term. Error verification can also be performed on the predicted film thickness of the new product. If the error verification is successful, the new product is included in the prediction range; if the new product does not meet the second slope reuse condition, or the error verification fails, a separate multiple linear regression model is established for the new product. The process shown in the diagram allows for the reuse of the slope relationship of a mature multiple linear regression model when the process of a new product is similar to that of an existing product. When the reuse conditions are not met, the model can be switched to independent modeling, thus balancing model import efficiency and prediction reliability.
[0102] Secondly, this embodiment provides an epitaxial film thickness determination system; please refer to [link to relevant documentation]. Figure 6 The epitaxial film thickness determination system 60 includes: The deposition step identification module 601 is used to identify the deposition steps in the epitaxial process and to acquire the process data generated in the deposition steps during the processing of the wafer to be predicted by the epitaxial equipment. The epitaxial equipment includes a reaction chamber. The process feature determination module 602 is used to determine the process features of the wafer to be predicted based on the process data generated during the deposition step. The film thickness prediction module 603 is used to predict the epitaxial film thickness of the wafer to be predicted by using a pre-constructed multiple linear regression model to obtain the predicted film thickness of the wafer to be predicted. The multiple linear regression model includes a global slope coefficient and an intercept term. The global slope coefficient of the multiple linear regression model corresponding to different reaction chambers is the same, and the intercept term corresponds to the reaction chamber.
[0103] Thirdly, this embodiment provides an electronic device, please refer to [link / reference]. Figure 7 It includes a memory and a processor, the memory stores a computer program, and the computer program, when executed by the processor, implements any of the above methods.
[0104] Fourthly, this embodiment provides a computer-readable storage medium having a computer program stored thereon, the computer program being loaded by a processor to perform the steps in any of the methods described above.
[0105] In the embodiments of this application, the storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.
[0106] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.
[0107] The foregoing has provided a detailed description of an epitaxial film thickness determination method, system, electronic device, and storage medium provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. An epitaxial film thickness determination method characterized by, Includes the following steps: Identify the deposition step in the epitaxial process and acquire the process data generated in the deposition step during the processing of the wafer to be predicted by the epitaxial equipment, wherein the epitaxial equipment includes a reaction chamber; Based on the process data generated during the deposition step, the process characteristics of the wafer to be predicted are determined; The epitaxial film thickness of the wafer to be predicted is predicted by using a pre-constructed multiple linear regression model to analyze the process characteristics of the wafer to be predicted. The multiple linear regression model includes a global slope coefficient and an intercept term. The global slope coefficients of the multiple linear regression models corresponding to different reaction chambers are the same, and the intercept term corresponds to the reaction chamber.
2. The method of claim 1, wherein Based on the process data generated during the deposition step, the process characteristics of the wafer to be predicted are determined, including: Feature extraction is performed on the process data generated during the deposition step to obtain film thickness-related process parameters; The process characteristics of the wafer to be predicted are determined by statistical aggregation of the film thickness-related process parameters.
3. The method of claim 1, wherein The steps for constructing the multiple linear regression model include: Acquire process data and actual film thickness measurement data of the epitaxial equipment for wafer processing; Identify the deposition steps in the epitaxial process and generate corresponding wafer process features based on the process data within the deposition steps; The process characteristics are matched with the actual film thickness measurement data to obtain a matching sample; The multiple linear regression model is established based on the matched samples.
4. The method of claim 3, wherein The process characteristics are matched with the actual film thickness measurement data to obtain a matching sample, including: Obtain the cavity identifier, batch identifier, wafer identifier, slot information, and process time corresponding to the process characteristics; Obtain the batch identifier, wafer identifier, slot information, and measurement time corresponding to the actual film thickness measurement data; Based on batch identifier, wafer identifier, and slot information, the process characteristics are initially matched with the actual film thickness measurement data to obtain the initial matching result; The initial matching result is verified based on the cavity identifier and the process time. Wafers that pass the verification and simultaneously possess complete process characteristics and reliable actual film thickness measurement data are identified as matching samples. The complete process characteristics are process characteristics without missing data, and the reliable actual film thickness measurement data are actual film thickness measurement data that have not been judged as measurement anomalies.
5. The method of claim 3, wherein The actual film thickness measurement data is the average film thickness of the wafer measured at multiple points.
6. The method of claim 3, wherein The multiple linear regression model is established based on the matched samples, including: Within the same process menu or the same process family, the global slope coefficient and the intercept term corresponding to each of the reaction chambers are trained based on matching samples from multiple reaction chambers. The same process family refers to a set of process menus where the ratio of main hydrogen to slit hydrogen is the same.
7. The method for determining epitaxial film thickness according to claim 2, characterized in that, The method further includes: The wafer with complete process characteristics but without actual film thickness measurement data is identified as the wafer to be predicted. The complete process characteristics are process characteristics without missing data.
8. The method for determining epitaxial film thickness according to claim 1, characterized in that, The method further includes: When the newly added reaction chamber of the epitaxial device operates the same process menu or meets the first slope reuse condition, the global slope coefficient in the multiple linear regression model is fixed. The first slope reuse condition is that the average absolute error obtained by verifying the model with shared global slope coefficients is less than a preset error threshold. Under the condition of fixing the global slope coefficient, the intercept term corresponding to the newly added reaction chamber is learned using the matching samples corresponding to the newly added reaction chamber; Based on the global slope coefficient and the intercept term corresponding to the newly added reaction cavity, the epitaxial film thickness of the wafer to be predicted in the newly added reaction cavity is predicted to obtain the predicted film thickness of the newly added reaction cavity.
9. The method for determining epitaxial film thickness according to claim 8, characterized in that, The method further includes: The predicted film thickness of the newly added reaction chamber was verified using validation samples. If the preset error index corresponding to the predicted film thickness is less than or equal to the preset error threshold, the newly added reaction chamber will be included in the prediction range.
10. The method for determining epitaxial film thickness according to claim 1, characterized in that, The method further includes: Obtain the process information and target film thickness of the new product. The process information includes at least one of the process menu, core gas settings, power ratio, and flow field settings. The process information of the newly added product is compared with the process information of the existing product to determine whether the newly added product meets the second slope reuse condition. The second slope reuse condition includes that the process information of the newly added product and the existing product meet a preset similarity condition. If the newly added product meets the second slope reuse condition, the global slope coefficient in the multiple linear regression model is reused. Multiple wafer samples with complete process characteristics and reliable actual film thickness measurement data are obtained from the new product, and an intercept term corresponding to the new product and its reaction chamber is determined based on each wafer sample; the complete process characteristics are process characteristics without missing data, and the reliable actual film thickness measurement data are actual film thickness measurement data that have not been judged as measurement abnormalities. Based on the global slope coefficient and the intercept term corresponding to the new product, the epitaxial film thickness of the wafer to be predicted in the new product is predicted.
11. A system for determining epitaxial film thickness, characterized in that, include: A deposition step identification module is used to identify deposition steps in an epitaxial process and acquire process data generated during the deposition steps when the epitaxial equipment processes the wafer to be predicted. The epitaxial equipment includes a reaction chamber. The process feature determination module is used to determine the process features of the wafer to be predicted based on the process data generated during the deposition step. The film thickness prediction module is used to predict the epitaxial film thickness of the wafer to be predicted by using a pre-constructed multiple linear regression model to analyze the process characteristics of the wafer to be predicted, thereby obtaining the predicted film thickness of the wafer to be predicted. The multiple linear regression model includes a global slope coefficient and an intercept term, and the global slope coefficient of the multiple linear regression model corresponding to different reaction chambers is the same, while the intercept term corresponds to the reaction chamber.
12. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the method as described in any one of claims 1 to 10.
13. A computer-readable storage medium, characterized in that, It contains a computer program that is loaded by a processor to perform the steps of the method according to any one of claims 1 to 10.
