Semiconductor device production data processing method and system

Abstract

The invention discloses a semiconductor device production data processing method and system, and the method comprises the steps: obtaining production process data generated in the production process of a semiconductor device, and the production process data comprises at least one of enterprise resource planning (ERP) data, production execution system (MES) data, and quality assurance system (QA) data, supplier production order online information WIP data; converting unstructured data in the production process data into structured data by adopting a heterogeneous data source off-line synchronous DATAX method; and carrying out data processing on all the structured data by adopting a change data capture (CDC) method, and synchronizing a data processing result to a distributed storage system database. According to the scheme, different formats and different types of data generated in the semiconductor production process can be integrated, and subsequent data analysis is facilitated.

Description

Technical Field

[0001] The present invention relates to the technical field of semiconductor device production, and in particular to a semiconductor device production data processing method and system. Background Art

[0002] The production process of semiconductor devices is long and complex. At the same time, the production of most semiconductor devices cannot be completed by the same manufacturer, resulting in non-standardization and inconsistency in data management at each production stage.

[0003] With the rise of multiple semiconductor manufacturing plants, the phenomenon of data silos has become more and more prominent. Data from different sources have significant differences in format, naming rules, and even semantic understanding of the same field. This lack of uniformity at the data level increases the difficulty of data integration and analysis. Summary of the invention

[0004] The purpose of the present invention is at least to provide a method for processing semiconductor device production data, which can integrate data of different formats and types generated in the semiconductor production process to facilitate subsequent data analysis.

[0005] In a first aspect, the present invention provides a method for processing semiconductor device production data, comprising: acquiring production process data generated during the production process of semiconductor devices, the production process data comprising at least one of the following: enterprise resource planning ERP data, production execution system MES data, quality assurance system QA data, supplier production order online information WIP data; using a heterogeneous data source offline synchronization DATAX method to convert unstructured data in the production process data into structured data; using a change data capture CDC method to process all structured data, and synchronizing the data processing results to a distributed storage system database; the data processing performed on all structured data comprises at least one of the following: data extraction, data conversion, and deduplication processing.

[0006] The DATAX method of offline synchronization of heterogeneous data sources is used to convert unstructured data in the production process data into structured data; all structured data are processed through the CDC method of change data capture, and the processed production process data is synchronized to the distributed storage system database. For the production process data generated in the production process with high real-time requirements, the CDC method of change data capture is used to synchronize to the distributed storage system database; for the data with low real-time requirements, the DATAX method of offline synchronization of heterogeneous data sources is used to reduce the pressure of data processing. In the process of data processing, the accuracy of subsequent analysis results can be improved by deduplication of production process data and other operations.

[0007] Optionally, the adopting the change data capture CDC method to perform data processing on all structured data includes: adopting the CDC method to perform data processing on the structured data in the production process data; and adopting the CDC method to perform data processing on the converted structured data.

[0008] Optionally, the semiconductor device production data processing method further includes: acquiring historical production data from the distributed storage system database; and / or acquiring current production data from the distributed storage system database.

[0009] Optionally, after acquiring the historical production data, the method further includes: establishing a prediction model based on the historical production data; and predicting the production process of the current semiconductor device based on the prediction model to obtain prediction data.

[0010] Optionally, the semiconductor device production data processing method also includes: determining whether the prediction model needs to be corrected based on the root mean square error of the prediction data and the current production data; in response to the root mean square error being greater than a preset credibility threshold, determining to correct the prediction model.

[0011] Optionally, the semiconductor device production data processing method further includes: acquiring historical yields and historical delivery times based on the historical production data; and estimating the yield and delivery time of the current generation process based on the historical yields and the historical delivery times.

[0012] Optionally, the semiconductor device production data processing method further includes: acquiring the status of the current semiconductor device production process based on the QA data and WIP data in the current production data.

[0013] Optionally, after the state of the production process of the current semiconductor device is acquired, if it is detected that the state of the production process of the current semiconductor device is abnormal, an alarm signal is output.

[0014] Optionally, the current production data is displayed in any of the following ways: dynamic charts, dynamic graphics.

[0015] In a second aspect, the present invention provides a semiconductor device production data processing system, comprising: an acquisition unit, used to acquire production process data generated during the production process of semiconductor devices, the production process data including at least one of the following: enterprise resource planning ERP data, production execution system MES data, quality assurance system QA data, supplier production order online information WIP data; a conversion unit, used to adopt a heterogeneous data source offline synchronization DATAX method to convert unstructured data in the production process data into structured data; a processing unit, used to adopt a change data capture CDC method to perform data processing on all structured data, and synchronize the data processing results to a distributed storage system database; the data processing performed on all structured data includes at least one of the following: data extraction, data conversion, and deduplication processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of a semiconductor device production data processing method in an embodiment of the present invention;

[0017] Figure 2 is a flow chart of using Doris database to perform yield analysis in an embodiment of the present invention;

[0018] Figure 3 is a flow chart of using Doris database to perform production forecasting in an embodiment of the present invention;

[0019] Figure 4 is a flow chart of a model training method in an embodiment of the present invention;

[0020] Figure 5 It is a structural schematic diagram of a semiconductor device generating data processing system in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] Currently, data in production is mixed in different systems. There are structured data such as Mysql, Postgresql and other databases, as well as some semi-structured data, such as Fab factory web page data, interface data and email data. In the original analysis, it is necessary to connect to multiple databases through different tools. At the same time, there may be multiple ways to express the same attribute in different databases. In addition, the existence of some dirty data has caused troubles in subsequent analysis. At the same time, it also puts pressure on the access to these databases, resulting in slow analysis speed and low real-time performance.

[0022] In an embodiment of the present invention, a DATAX method for offline synchronization of heterogeneous data sources is used to convert unstructured data in production process data into structured data; data processing is performed on all structured data through a CDC method for change data capture, and the processed production process data is synchronized to a distributed storage system database. For data with high real-time requirements in the production process data generated during the production process, a CDC method for change data capture is used to synchronize the data to a distributed storage system database; for data with low real-time requirements, a DATAX method for offline synchronization of heterogeneous data sources is used to reduce data processing pressure. In the process of data processing, the accuracy of subsequent analysis results can be improved by performing operations such as deduplication processing on the production process data.

[0023] In order to make the above-mentioned objects, features and beneficial effects of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0024] The embodiment of the present invention provides a method for processing semiconductor device production data, referring to Figure 1 , the following is a detailed description through specific steps.

[0025] Step 101, obtaining production process data generated during the production process of semiconductor devices.

[0026] In an embodiment of the present invention, the production process data generated during the semiconductor device production process may include one or more of the following: enterprise resource planning (ERP) data, manufacturing execution system (MES) data, quality assurance system (QA) data, and supplier production order online information (WIP) data.

[0027] In a specific implementation, the above-mentioned production process data may be production process data within a period of time (also referred to as historical production data), as well as current production data generated during the ongoing semiconductor device production process. The above-mentioned historical production data may include production process data of multiple different batches of semiconductor devices. The production process data for a certain batch of semiconductor devices may include all data of the entire production cycle of the batch of semiconductor devices.

[0028] Step 102: using a heterogeneous data source offline synchronization method to convert unstructured data in the production process data into structured data.

[0029] In specific implementations, the production process data of semiconductor devices can be distributed on different platforms and databases, some of which may be structured data, and other data may be unstructured data. For example, WIP data comes from emails, web pages, data interfaces, etc., which are unstructured data.

[0030] The above-mentioned ERP data, MES data and QA data are data with high real-time requirements in the production process. WIP data is mainly used to record the production status of the production line and has lower real-time requirements.

[0031] Specifically, EPR data may include purchase orders, production orders, inventory data, etc. MES data may include execution data of each work order, environmental data, raw material inventory data, etc. QA data may include the production status of each outsourced manufacturer for outsourced orders, such as providing a MAP diagram of the wafer, etc. WIP data may include the current online status of orders provided by the outsourced manufacturer at a certain moment, including the online quantity and online status of the wafer, etc. It may also include the current online quantity of each test process during the FT test process, as well as data on good and bad products, etc.

[0032] For unstructured data of email type, since each Fab factory sends emails to the designated mailbox from time to time, it cannot be processed in real time. DATAX currently does not support parsing EML files. In the embodiment of the present invention, by extending the plug-in of DATAX to support reading EML files, the Fab factory sends data to the designated mailbox by email, and the mailbox system automatically saves the email content of the special recipient to the designated directory (the mark of the present invention is "_WIP", that is, the recipient's mailbox contains this content), and the save format is EML format. DATAX will extract the EML file content regularly and parse it into structured data. At the same time, some emails will carry excel attachment data. The EML plug-in of DATAX will first extract the excel attachment data and save it to the local temporary directory and parse this excel data into structured data.

[0033] For unstructured data of web page type, since web page data is not synchronized with Fab machine data in real time, it is impossible to process web page data in real time. In the embodiment of the present invention, by extending the plug-in of DATAX, a Python script is started regularly through DATAX to parse the website data and convert it into structured data.

[0034] For other data interfaces, each Fab factory mainly provides a fixed URL address and a user name, password (or other authentication methods) to read data. In the implementation of the present invention, by extending the plug-in of DATAX, it supports reading data through a Uniform Resource Locator (URL) and converting it into structured data.

[0035] In the embodiment of the present invention, after obtaining the production process data, the DATAX method can be used to convert the unstructured data in the production process data into structured data. Structured data is strictly organized into rows and columns and has a fixed type, which is more convenient for analysis and query. Therefore, by converting all unstructured data into structured data, the training speed of the subsequent prediction model and the query speed of related reports can be improved.

[0036] Step 103: Using a change data capture method, all structured data is processed, and the processed production process data is synchronized to a distributed storage system database.

[0037] In specific implementations, the distributed storage system database can distribute data on different nodes, which can reduce the pressure on a single node during data query. At the same time, even if some nodes fail, the system can still continue to run, thereby improving the robustness of the entire database.

[0038] In the embodiment of the present invention, a data change capture (CDC) method may be used to process the structured data in the production process data; and the CDC method may be used to process the structured data converted in step 102.

[0039] That is to say, all the structured data mentioned above include: the structured data in the production process data acquired in step 101 , and the structured data converted in step 102 .

[0040] In a specific implementation, processing all structured data may include: data extraction, data conversion, loading processing of all structured data, as well as deduplication processing, default value processing, etc. of all structured data.

[0041] In data extraction, by extending the DATAX plug-in, it supports reading EML (email), website, and data interface unstructured data into structured data, and synchronizes all structured data to the distributed database through the CDC technical solution. This solution only relies on two tools, DATAX and CDC, to reduce system dependence and improve system robustness. At the same time, all data processed are structured data, which reduces the difficulty of subsequent data conversion.

[0042] In the data conversion process, the CDC is extended to support the conversion of foreign key data into real data, and the conversion of different types of time data into data of the same time zone and the same type. For example, the time returned by some interface data is UTC time (Coordinated Universal Time), and some are stored in the form of serial numbers in Excel (storing all seconds from January 1, 1970 to the current time). The present invention can convert them all into Beijing time for unified processing during data conversion.

[0043] In actual applications, it can be seen that the existing DATAX supports data formats such as TxtFile and HDFS for unstructured data, but does not support formats such as EML, URL, and HTML.

[0044] In the embodiment of the present invention, based on DATAX supporting custom plug-in extensions, the Reader interface in DATAX is implemented in Java language to implement EMLReader, URLReader, and HTMLReader plug-ins, and the custom plug-ins are packaged through assembly and registered into DATAX through plugin.json to realize custom reading of data in the required format.

[0045] In addition, existing CDC methods support synchronization of relational databases such as Mysql, Oracle, and Postgresql. However, a large amount of data is stored in the database in the form of foreign keys, which makes foreign key processing complicated during subsequent data analysis and model training.

[0046] In the embodiment of the present invention, a Foreign plug-in is added through the Java language, and the plug-in supports automatically splicing the foreign key data into one field in a specified manner for format conversion during CDC synchronization, thereby realizing the processing of the foreign key data.

[0047] In the specific implementation, for the structured data with high real-time requirements in the production process data generated in the production process, the change data capture CDC method is used to synchronize it to the distributed storage system database; for the unstructured data with low real-time requirements, the heterogeneous data source offline synchronization DATAX method is first used to convert it into structured data, and then the CDC method is used to synchronize the converted structured data to the distributed storage system database, which can reduce the data processing pressure.

[0048] In specific implementations, when using the data change capture method to process structured data, there may be deviations. For example, data at 23:59:59 (i.e. 23:59:59) on January 1 will be considered as 0:00 on January 2, resulting in the data at 23:59:59 being calculated as the data for the next day.

[0049] In order to avoid the above-mentioned errors, the offline synchronization method of heterogeneous data sources can be combined to synchronize and correct the data of the previous day, thereby ensuring the accuracy of the data of the previous day.

[0050] For data deduplication, the system can define deduplication methods for each field, such as taking the latest value, taking the maximum value, taking the minimum value, summing, etc. At the same time, if an error occurs during the deduplication process, the latest value will be taken as the deduplication value for subsequent tasks, and the alarm system will be used to notify relevant personnel that there is data that has failed to be deduplicated for data processing.

[0051] For data default value processing, the system defines a list of important fields. If the default value is an important field, the system will notify manual processing and continue to execute subsequent tasks after the data is completed. If the default value is a non-important field, the system will give priority to obtaining whether the data exists in the system in the previous N days (currently N is set to 3). If it exists, the data of the most recent date is used as the default value. If it does not exist, the default value is given according to the field type, and the alarm system is used to remind relevant personnel of the existence of default values.

[0052] After the production process data is processed, the processed production process data can be synchronized to the distributed storage system (Doris) database.

[0053] In the specific implementation, after completing data synchronization, when there is a demand for production process data in the future, the production process data can be read from the Doris database by accessing the Doris database. Due to the high performance and real-time characteristics of the Doris database, higher data throughput can be achieved and the performance of data analysis can be improved.

[0054] In the embodiment of the present invention, after the processed production process data is synchronized to the Doris database, statistical analysis can be performed on the processed production process data (ie, the data processing result).

[0055] In a specific implementation, the data processing results can be read from the Doris database to obtain data related to the yield rate and the delivery time. Based on the obtained data related to the yield rate and the delivery time, the yield rate and the delivery time in the subsequent semiconductor device production process can be predicted.

[0056] During the production process of semiconductor devices, the yield of semiconductor devices may not reach 100% due to various reasons. The historical production data of semiconductor devices is obtained from the Doris database. From the historical production data, we can know the total number of initial semiconductor devices (gross die) and the number of good products (good die) produced in the chip design process, as well as the first inspection yield, batch yield, project yield, Fab factory annual yield and other yield data. Furthermore, based on the historical production data, the yield of subsequent semiconductor device production can be predicted.

[0057] Reference Figure 2, a flow chart of using the Doris database to perform yield analysis in an embodiment of the present invention is given.

[0058] Step 201, obtaining the stored data processing results from the Doris database.

[0059] Step 202, determining the yield type to be obtained.

[0060] In a specific implementation, the yield type may include first inspection yield, batch yield, project yield, Fab plant annual yield, etc.

[0061] Step 203, based on the yield type, set the yield judgment threshold and obtain the yield judgment result.

[0062] Step 204: When the yield determination result indicates that the yield is low, an alarm signal is output.

[0063] By outputting an alarm signal, production personnel can be reminded of abnormal yield conditions, so that they can analyze the causes of the abnormal yield.

[0064] Specifically, the first inspection yield can be used to evaluate the initial quality of the production process. The first inspection yield is the quotient of the number of semiconductor devices that pass the first inspection and the total number of first inspections. If the first inspection yield is low, it is necessary to analyze the corresponding reasons and propose improvement plans.

[0065] Batch yield can be used to evaluate the production quality of a batch. Batch yield is the quotient of the number of qualified semiconductor devices in a batch and the total number of semiconductor devices produced in the batch. If the batch yield of a batch is lower than the average yield of the same type of semiconductor devices over a period of time, an early warning message can be output to remind production personnel that the yield of the batch of semiconductor devices is low. This makes it easier for subsequent production personnel to analyze the cause of the low yield of the batch of semiconductor devices.

[0066] Project yield can be used to characterize the yield of a project. A project may include multiple batches of semiconductor device production. Project yield can be: Among them, w k is the weight of the batch yield of the kth batch of the project, A k is the batch yield of the kth batch of the project, w k It is the quotient of the total number of qualified semiconductor devices in the kth batch and the total number of semiconductor devices produced in the kth batch.

[0067] The annual yield rate of a Fab factory can be used to characterize the production quality of the Fab factory. The annual yield rate of a Fab factory can be: the quotient of the total number of all qualified semiconductor devices of the Fab factory in one year and the total number of semiconductor devices produced in one year.

[0068] In practical applications, other types / names of yields can also be included, all of which can be obtained from historical production data. For example, the monthly yield of the Fab factory can also be included, which is the quotient of the total number of all qualified semiconductor devices in the Fab factory in a month and the total number of semiconductor devices produced in a month.

[0069] In actual implementation, the production cycle of semiconductor devices is relatively long and may involve multiple Fab manufacturers. Therefore, in the production process of semiconductor devices, no matter which link has problems, it will affect the subsequent production, delivery and other time.

[0070] In the embodiment of the present invention, the order placement time of the production order, the completion time of the production order and the WIP data in the production process can be obtained from the historical production data. Based on the delivery time of the semiconductor device and the time of each major node in the production process of the semiconductor device, accurate time data is provided for subsequent production prediction.

[0071] Reference Figure 3 , a flowchart of using the Doris database for production forecasting in an embodiment of the present invention is given.

[0072] Step 301, obtaining the stored data processing results from the Doris database.

[0073] Step 302, obtaining the delivery time based on the data processing result.

[0074] Step 303: predict the yield and delivery time in the current production process based on the historical yield and historical delivery time.

[0075] Specifically, delivery time may include first delivery time, batch delivery time, and project delivery time, where:

[0076] The first delivery time can be the time difference between the first order receipt time and the first order placement time. The first delivery time can be used to evaluate the time of the production process and to evaluate the scheduling of subsequent production orders.

[0077] Batch delivery time can be: Among them, p1, p2, ..., p n Characterize the delivery time of the batch production process, q1, q2, ..., q n Characterizes the order placement time of the batch production process. Since the semiconductor device production process includes multiple steps, including wafer tape-out, CP test, FT test, etc., in the embodiment of the present invention, the delivery time of a batch is defined as: from the placement of the wafer order to the completion of the product FT test.

[0078] Project delivery time can be: Among them, w1, w2, ..., wn The duration weight that represents the delivery time of a batch of the project, B1, B2, ..., B n Characterizes the delivery time of the batch. The project delivery time can truly reflect the delivery cycle of the project.

[0079] In the embodiment of the present invention, the yield in the current production process can be predicted based on the project yield and the monthly yield of the Fab plant.

[0080] Specifically, the statistical yields of n projects are X1, X2, ..., Xn, and the monthly yields of the Fab factory in the past m months are Y1, Y2, ..., Ym. The predicted yield in the current production process is: A1X1+A2X2+...+AnXn+B1Y1+B2Y2+...+BmYm.

[0081] By obtaining historical yields and historical delivery times, the yield and delivery time in the current production process can be predicted.

[0082] In the specific implementation, the current production data can be obtained from the Doris database, and the current production data can be analyzed in real time to monitor the current production process in real time. When an abnormality is detected in the current production process, an early warning is triggered to remind the production personnel that the current production process is abnormal.

[0083] Specifically, the WIP data may be obtained from the current production data, and the WIP data may be monitored to determine whether there is any abnormality in the production process.

[0084] In specific applications, it can be seen that WIP data plays an important role in each stage of the semiconductor device production process. WIP data is provided by various Fab manufacturers. After obtaining the WIP data of the current production process, the WIP data can be compared and analyzed with the predicted data. If it is detected that the data of a batch is continuously abnormal, it is determined that the production process of the batch is abnormal.

[0085] The WIP data includes a status field, which is used to identify the job status, such as PENDDING, HOLD, etc. If the status field in the WIP data includes PENDDING, HOLD, etc., but no data warning is received in the QA system, an early warning can be triggered to remind the production personnel, thereby ensuring the smooth progress of the production process.

[0086] QA data is used in the semiconductor testing phase, and the measured data is regularly reported to the system. The system can perform multi-dimensional analysis on historical measured data and current measured data to ensure the quality and production efficiency of semiconductor devices. At the same time, when there are abnormalities in the QA data, an early warning can also be triggered to remind production personnel.

[0087] Therefore, by real-time monitoring of the data of the current production process, the changing trends of various data in the current production process can be fully reflected, thereby ensuring the smooth progress of the production process.

[0088] In an embodiment of the present invention, if an abnormal situation occurs in the current production process, the abnormal situation can also be notified to relevant operations and sales personnel, which helps the operations and sales personnel to timely check inventory information and delivery information to ensure normal operations and sales.

[0089] In the embodiment of the present invention, XGBoost is an ensemble learning algorithm based on gradient boosting, which is an efficient implementation of gradient boosting decision tree. XGBoost builds a strong learner by integrating multiple weak learners, thereby achieving high-precision prediction of data.

[0090] In specific implementation, for each link in the semiconductor device production process, XGBoost-based models can be built for different products in wafer production, CP production, and FT testing.

[0091] Specifically, historical production data can be obtained from the Doris database, and a prediction model can be trained based on the XGBoost method and historical production data.

[0092] Before model training, parameters that may affect the Good Die and production delivery time of semiconductor devices during the production process can be obtained from historical production data, including: semiconductor device process, packaging form, package size, wafer size, pad composition, wafer thickness, process, pad thickness, pad size, wafer removal method, adhesive type, etc.

[0093] In a specific implementation, the output of the initial model can be: Among them, f k Represented as a regression tree, K is the number of regression trees; the above formula represents a given output X i , output the sum of K regression trees.

[0094] Reference Figure 4 , a flowchart of a model training method in an embodiment of the present invention is given.

[0095] Step 401, obtaining production order data when placing an order.

[0096] Step 402, obtaining predicted production process data based on the production order data.

[0097] Step 403, obtaining actual production process data when closing the order.

[0098] Step 404, comparing the predicted production process data with the actual production process data.

[0099] Step 405: If the difference between the predicted production process data and the actual production process data is less than a preset threshold, the model is saved.

[0100] Step 406: If the difference between the predicted production process data and the actual production process data is greater than a preset threshold, the model training is performed again.

[0101] Step 407, modify the existing model.

[0102] In the embodiment of the present invention, the existing trained model (hereinafter referred to as the existing model) can also be continuously learned and optimized.

[0103] In specific implementation, the existing model can be automatically corrected. After obtaining the latest real data in the production process, the real data in the production process can be compared with the current predicted data of the existing model. The root mean square error between the real data and the current predicted data can be calculated to comprehensively evaluate and quantify the current credibility of the existing model.

[0104] If the current credibility of the existing model is less than the preset credibility threshold, it can be determined that the prediction performance of the existing model is poor, so the existing model can be corrected. Correcting the existing model can refer to retraining the existing model, or training the existing model based on incremental data. The above incremental data can refer to data that, after being input into the existing model, the current credibility of the existing model is less than the preset credibility threshold.

[0105] In specific implementations, the existing model can also be modified by combining manual feedback. Feedback and suggestions from production personnel are received through the Application Programming Interface (API). After receiving feedback and suggestions from production personnel, historical data can be evaluated to determine whether to modify the existing model.

[0106] In the specific implementation, the model in the model training process can be saved as an operator. When revising an existing model, the model closest in time is preferred for correction training, and the revised model is verified at the same time; if the credibility of the revised model is less than the preset credibility threshold, the model with the second closest time is used for correction training, and the revised model is verified at the same time.

[0107] By continuously optimizing and correcting existing models, the production process can be controlled and adjusted in a timely manner, thereby improving production efficiency and reducing risks.

[0108] The value of the above credibility threshold can be set based on the actual application scenario. For example, the credibility threshold is set to 10%. If the root mean square error between the real data and the current predicted data is greater than 10%, the existing model is corrected; conversely, if the root mean square error between the real data and the current predicted data is less than 10%, the existing model may not be corrected.

[0109] In the embodiment of the present invention, the production data can also be displayed in the form of real-time reports based on the Doris database, so that production personnel can obtain production data in a timely manner and make corresponding decisions and adjustments.

[0110] In some embodiments, the evolution trend of production data can be intuitively displayed through animation or timeline. Alternatively, production personnel can intuitively observe the change process of production data through dynamic icons and images.

[0111] Reference Figure 5 , a semiconductor device production data processing system 50 in an embodiment of the present invention is provided, comprising:

[0112] The acquisition unit 501 is used to acquire production process data generated during the production process of semiconductor devices, wherein the production process data includes at least one of the following: enterprise resource planning ERP data, production execution system MES data, quality assurance system QA data, and supplier production order online information WIP data;

[0113] A conversion unit 502, configured to convert the unstructured data in the production process data into structured data by adopting a DATAX method for offline synchronization of heterogeneous data sources;

[0114] The processing unit 503 is used to adopt the change data capture CDC method and the heterogeneous data source offline synchronization DATAX method to process the production process data and synchronize the processed production process data to the distributed storage system database; the data processing of the production process data includes at least one of the following: data extraction, data conversion, and deduplication processing.

[0115] In a specific implementation, the specific execution process of the above-mentioned acquisition unit 501, conversion unit 502 and processing unit 503 can correspond to step 101 to step 103, which will not be repeated here.

[0116] In a specific implementation, the processing unit 503 may include a data synchronization device, a distributed storage system storage device, a data analysis device, a data prediction device, an alarm device, a reporting device, etc. The specific implementation functions of the processing unit 503 can refer to the description of step 103 in the above embodiment, which will not be repeated here.

[0117] In a specific implementation, each module / unit included in each device or product described in the above embodiments may be a software module / unit or a hardware module / unit, or may be partly a software module / unit and partly a hardware module / unit.

[0118] For example, for each device or product applied to or integrated in a chip, each module / unit contained therein may be implemented in the form of hardware such as circuits, or at least some of the modules / units may be implemented in the form of software programs, which run on a processor integrated inside the chip, and the remaining (if any) modules / units may be implemented in the form of hardware such as circuits; for each device or product applied to or integrated in a chip module, each module / unit contained therein may be implemented in the form of hardware such as circuits, and different modules / units may be located in the same component (such as a chip, circuit module, etc.) or different components of the chip module, or at least some of the modules / units may be implemented in the form of software programs. The element can be implemented in the form of a software program, which runs on a processor integrated inside the chip module, and the remaining (if any) modules / units can be implemented in the form of hardware such as circuits; for various devices and products applied to or integrated in the terminal, the various modules / units contained therein can be implemented in the form of hardware such as circuits, and different modules / units can be located in the same component (for example, chip, circuit module, etc.) or in different components in the terminal, or, at least some modules / units can be implemented in the form of a software program, which runs on a processor integrated inside the terminal, and the remaining (if any) modules / units can be implemented in the form of hardware such as circuits.

[0119] An embodiment of the present invention also provides a computer-readable storage medium, which is a non-volatile storage medium or a non-transient storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the semiconductor device production data processing method provided in any of the above embodiments are executed.

[0120] A person skilled in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by instructing the relevant hardware through a program, and the program may be stored in a computer-readable storage medium, which may include: ROM, RAM, disk or CD, etc.

[0121] Although the present invention is disclosed as above, the present invention is not limited thereto. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the scope defined by the claims.

Claims

1. A method for processing semiconductor device production data, characterized in that: include: Acquire production process data generated during the production of semiconductor devices, wherein the production process data includes at least one of the following: enterprise resource planning ERP data, production execution system MES data, quality assurance system QA data, and supplier production order online information WIP data; Using a DATAX method for offline synchronization of heterogeneous data sources to convert unstructured data in the production process data into structured data; A change data capture (CDC) method is used to process all structured data, and the data processing results are synchronized to a distributed storage system database; the data processing performed on all structured data includes at least one of the following: data extraction, data conversion, and deduplication processing.

2. The semiconductor device production data processing method according to claim 1, characterized in that: The change data capture (CDC) method is used to process all structured data, including: The CDC method is used to process the structured data in the production process data; and the CDC method is used to process the converted structured data.

3. The semiconductor device production data processing method according to claim 1, characterized in that: It also includes: acquiring historical production data from the distributed storage system database; and / or, The current production data is obtained from the distributed storage system database.

4. The semiconductor device production data processing method according to claim 3, characterized in that: After obtaining the historical production data, the method further includes: Establishing a prediction model based on the historical production data; Based on the prediction model, the production process of the current semiconductor device is predicted to obtain prediction data.

5. The semiconductor device production data processing method according to claim 4, characterized in that: Also includes: Based on the root mean square error of the predicted data and the current production data, determining whether the prediction model needs to be corrected; In response to the root mean square error being greater than a preset credibility threshold, it is determined to correct the prediction model.

6. The semiconductor device production data processing method according to claim 3, characterized in that: Also includes: Based on the historical production data, obtain historical yields and historical delivery times; Based on the historical yield and the historical delivery time, the yield and delivery time of the current production process are estimated.

7. The semiconductor device production data processing method according to claim 3, characterized in that: Also includes: Based on the QA data and WIP data in the current production data, the status of the current production process of the semiconductor device is obtained.

8. The method for processing semiconductor device production data according to claim 7, wherein: After the state of the production process of the current semiconductor device is obtained, the method further includes: It is detected that the state of the current semiconductor device production process is abnormal, and an alarm signal is output.

9. The semiconductor device production data processing method according to claim 7, characterized in that: It also includes: displaying the current production data in any of the following ways: dynamic charts, dynamic graphics.

10. A semiconductor device production data processing system, characterized in that: include: An acquisition unit is used to acquire production process data generated during the production process of semiconductor devices, wherein the production process data includes at least one of the following: enterprise resource planning ERP data, production execution system MES data, quality assurance system QA data, and supplier production order online information WIP data; A conversion unit, used for converting unstructured data in the production process data into structured data by adopting a DATAX method for offline synchronization of heterogeneous data sources; A processing unit is used to process all structured data using a change data capture (CDC) method and synchronize the data processing results to a distributed storage system database; the data processing performed on all structured data includes at least one of the following: data extraction, data conversion, and deduplication processing.

Images