A data analysis method and related device

Abstract

An embodiment of the present application provides a data analysis method and related device, including: determining a current site among each site of the WIP data; obtaining a current wafer production cycle of the current site and a current site anomaly judgment model, where the current wafer production cycle is each wafer production cycle of the current site in the WIP data, and the current site anomaly judgment model is a site anomaly judgment model of the current site that has been pre-trained; using the current site anomaly judgment model, when it is determined that the current wafer production cycle includes an abnormal wafer production cycle, obtaining the abnormal wafer production cycle and determining that the predicted status information of the current site is abnormal. The data analysis method and related device provided by the embodiments of the present application can improve the accuracy and real-time performance of integrated circuit production data analysis.

Description

Technical Field

[0001] Embodiments of the present application relate to the field of semiconductor technology, and in particular, to a data analysis method and related device. Background Art

[0002] With the development of complex semiconductor technology, the research and development of integrated circuits have gradually deepened, and the demand for integrated circuits has also increased.

[0003] In the process of integrated circuit design and processing, the last process is wafer processing by a foundry company to complete the final packaging and testing in integrated circuit design. During the entire wafer processing process, the foundry is responsible for the entire production process. For integrated circuit design manufacturers, it is necessary to timely detect abnormal wafer processing status to remind the foundry to adjust in time to ensure the normal production of integrated circuits.

[0004] Currently, in the prior art, integrated circuit design manufacturers will analyze the WIP data (Wafer in process, referring to wafers that have been on the production line but have not been shipped) transmitted by the foundry by manually drawing charts.

[0005] However, a large number of wafer processes will generate a huge amount of WIP data. The method of manually drawing charts to analyze WIP data usually requires engineers to make judgments based on the actual situation and experience, resulting in missed judgments, misjudgments, etc., and the timeliness cannot be guaranteed.

[0006] Therefore, how to improve the accuracy and timeliness of integrated circuit production data analysis has become an urgent technical problem to be solved. Summary of the Invention

[0007] The technical problem solved by the embodiments of the present application is to improve the accuracy and timeliness of integrated circuit production data analysis.

[0008] To solve the above problems, the embodiments of the present application provide a data analysis method and related device, including:

[0009] In a first aspect, the embodiments of the present application provide a data analysis method applicable to the analysis of WIP data, including:

[0010] Determine the current station among the stations of the WIP data;

[0011] Obtain the current wafer production cycle of the current station and the current station abnormal judgment model, where the current wafer production cycle is the wafer production cycles of the current station in the WIP data, and the current station abnormal judgment model is the station abnormal judgment model of the current station that has been pre-trained;

[0012] Using the current site anomaly determination model, when it is determined that the current wafer production cycle includes an abnormal wafer production cycle, the abnormal wafer production cycle is obtained, and the predicted status information of the current site is determined to be abnormal.

[0013] In a second aspect, an embodiment of the present application provides a data analysis device, which includes:

[0014] A current site determination module, adapted to determine the current site among the respective sites of the WIP data;

[0015] An acquisition module, adapted to acquire the current wafer production cycle of the current site and the current site anomaly determination model, where the current wafer production cycle is each wafer production cycle of the current site in the WIP data, and the current site anomaly determination model is the site anomaly determination model of the current site that has been pre-trained;

[0016] A current site status acquisition module, adapted to use the current site anomaly determination model, when it is determined that the current wafer production cycle includes an abnormal wafer production cycle, obtain the abnormal wafer production cycle, and determine that the predicted status information of the current site is abnormal.

[0017] In a third aspect, an embodiment of the present application further provides a storage medium, which stores a program suitable for data analysis to implement the data analysis method as described in the first aspect.

[0018] In a fourth aspect, an embodiment of the present application further provides an electronic device, which includes at least one memory and at least one processor; the memory stores a program, and the processor calls the program to execute the data analysis method as described in the first aspect.

[0019] Compared with the prior art, the technical solution of the embodiment of the present application has the following advantages:

[0020] In the data analysis method provided by the embodiments of the present application, by using the pre-trained current site anomaly judgment model and according to the current wafer production cycle of the current site in the WIP data, the predicted status information of the current site can be obtained. In this way, on the one hand, the data analysis method provided by the embodiments of the present application uses the current site anomaly judgment model to predict the status information of the current site without relying on manual labor. And the current site anomaly judgment model is pre-trained and has high prediction accuracy. Therefore, the predicted status information of the current site can be obtained accurately and reliably, improving the accuracy of integrated circuit production data analysis. On the other hand, the analysis speed of the current site anomaly judgment model is very fast. Even for a large amount of WIP data, the current site anomaly judgment model can analyze it in real time to obtain the analysis result. Thus, the status information of the current site can be obtained quickly and fed back to the wafer foundry company in a timely manner. It can be seen that the data analysis method provided by the embodiments of the present application can use the pre-trained current site anomaly judgment model to obtain the predicted status information of the current site, thereby improving the accuracy and real-time performance of WIP data analysis in integrated circuit production. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the provided drawings.

[0022] Figure 1 is a flowchart of the data analysis method provided by the embodiments of the present application.

[0023] Figure 2 is a schematic diagram of the preprocessed WIP data.

[0024] Figure 3 is a schematic diagram of the data of the wafer production cycle in the WIP data.

[0025] Figure 4 is a schematic diagram of the data of the broken wafer information in the WIP data.

[0026] Figure 5 is a schematic diagram of the data of the batching information in the WIP data.

[0027] Figure 6 is Figure 1 a flowchart of obtaining the current wafer production cycle of the current site in the data analysis method shown.

[0028] Figure 7It is a schematic diagram of the training process of the site anomaly judgment model.

[0029] Figure 8 It is an implementation process of the data analysis method provided by the embodiments of the present application.

[0030] Figure 9 It is another schematic diagram of the process of the data analysis method provided by the embodiments of the present application.

[0031] Figure 10 It is a schematic diagram of some site information included in the historical WIP data.

[0032] Figure 11 It is another schematic diagram of the process of the data analysis method provided by the embodiments of the present application.

[0033] Figure 12 It is for Figure 4 The data schematic diagram after processing the fragment information shown.

[0034] Figure 13 It is another schematic diagram of the process of the data analysis method provided by the embodiments of the present application.

[0035] Figure 14 It is for Figure 5 The data schematic diagram after processing the batch information shown

[0036] Figure 15 It is an optional block diagram of the data analysis device provided by the embodiments of the present application.

[0037] Figure 16 It is another optional block diagram of the data analysis device provided by the embodiments of the present application.

[0038] Figure 17 It is another optional block diagram of the data analysis device provided by the embodiments of the present application.

[0039] Figure 18 It is another optional block diagram of the data analysis device provided by the embodiments of the present application. Detailed implementation manners

[0040] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0041] Figure 1 It is a schematic diagram of the process of the data analysis method provided by the embodiments of the present application.

[0042] As shown in the figure, the data analysis method provided by the embodiments of the present application is applicable to the analysis of WIP data, and may include the following steps:

[0043] Step S10, determine the current station among the stations of the WIP data.

[0044] Before performing data analysis, first obtain the WIP data. After obtaining the WIP data, since the WIP data is chip production data including multiple stations, in order to obtain the predicted status information of the stations, first determine the station where the analysis is currently being performed.

[0045] It is easy to understand that the WIP data is data provided by the foundry manufacturer. Since the original WIP data sent by the foundry manufacturer contains data information of each manufacturer, and there is a lot of information that is meaningless for data analysis to obtain the production status of the wafers. Therefore, in order to better implement data analysis and obtain more accurate results, in some embodiments, before performing data analysis, the original WIP data can be processed to obtain the WIP data. Specifically, the steps of obtaining the WIP data may include:

[0046] Obtain the original WIP data, where the original WIP data includes all production information of the wafer;

[0047] Perform preprocessing of format adjustment and noise reduction on the original WIP data;

[0048] Extract information required for data analysis from the preprocessed original WIP data to obtain the WIP data.

[0049] Among them, the format adjustment preprocessing may include adjusting the file byte size. Adjusting the file byte size can make the data meet the input standard of the used model, which is convenient for subsequent use. This is because the input data file sizes of the models established by different algorithms are different, so it needs to be processed in advance.

[0050] The noise reduction preprocessing may include filtering sensitive information contained in the WIP data or data of wafers with special requirements during the production process to ensure the accuracy of subsequent analysis results.

[0051] The noise data for noise reduction preprocessing includes: exclusive marker data of each manufacturer, WIP data with special marks, etc., which will interfere with data analysis.

[0052] Specifically, the specific content of the preprocessed WIP data can be referred to Figure 2 , Figure 2 is a schematic diagram of the preprocessed WIP data.

[0053] As Figure 2As shown, the pre - processed WIP data contains useful data for subsequent analysis of integrated circuit chips, mainly including:

[0054] Wafer lot; main name of the lot; process step at the station, number of chips (Wafer) in the wafer lot; time of entering the production line; whether the station has changed; wafer production cycle (Cycle time); sub - lot; sub - lot station; sub - lot time and fragmentation information.

[0055] The main name of the lot is for the main wafer lot corresponding to each sub - lot wafer lot in the case of sub - lotting. For example, Figure 2 in the figure, at station D, there are three lots D1, D2, and D3. Taking lot D in the figure as an example, sub - lotting means dividing the wafer with 25 chips as one lot into three lots for processing. It can be seen that Figure 2 in the figure, the main wafer lot D has 25 chips, lot D1 includes 10 chips, lot D2 includes 5 chips, and lot D3 includes 10 chips, completing the chip processing of the main wafer lot D.

[0056] In the fragmentation information, the main marked station where fragmentation occurs is station 4.0 as shown in the figure, fragmentation time: 2020 / 6 / 18, fragmented wafer label: #1.

[0057] In practice, the amount of information contained in the WIP data is very large. For the convenience of explanation, only part of the pre - processed WIP data is shown in the figure.

[0058] After pre - processing, extract the information required for data analysis from the pre - processed original WIP data to obtain the data required for data analysis, thereby obtaining the WIP data. Specifically, a pre - written python script can be used to extract the information required for data analysis.

[0059] In this way, the data included in the processed WIP data, that is, the production situation data in the production process of integrated circuits (wafers), mainly can include batch information, station information, wafer production cycle (Cycle Time) of each wafer lot 、 fragmentation information, sub - lot and mother - lot sub - lotting information, etc., which are used to understand the production process and production status of wafers, and can analyze and understand the situation of the stations of the corresponding foundry manufacturers, shipping time and other information. Then, feedback the situation of the station equipment to the foundry manufacturer in a timely manner. The foundry manufacturer can perform maintenance on the equipment according to the feedback information to reduce unnecessary resource waste.

[0060] Among them, for the convenience of understanding, the specific form of the WIP data can refer to Figures 3 - 5 , Figure 3It is a schematic diagram of the wafer production cycle in the WIP data. Figure 4 It is a schematic diagram of the broken wafer information in the WIP data. Figure 5 It is a schematic diagram of the batch information in the WIP data. Specifically:

[0061] In this embodiment, the obtained WIP data should at least include Figure 3 the content to obtain the status information of the subsequent stations.

[0062] It is easy to understand that when the WIP data obtained after processing the historical WIP data includes the content in the above figures, the content of each part can be displayed in multiple WIP data files or in one file.

[0063] Of course, as shown in the overall WIP data after preprocessing Figure 2 the WIP data includes multiple stations. In order to predict the status information of each station, it is necessary to first determine the station currently being analyzed, that is, the current station.

[0064] It is easy to understand that the WIP data can include real-time WIP data or stored WIP data that has been downloaded and stored but not processed and used. When the WIP data is real-time WIP data, the real-time status information of the predicted station can be obtained through the data analysis method provided by the embodiments of the present application.

[0065] Step S11, obtain the current wafer production cycle of the current station and the current station anomaly judgment model, where the current wafer production cycle is the production cycles of each wafer at the current station in the WIP data, and the current station anomaly judgment model is the station anomaly judgment model of the current station that has been pre-trained.

[0066] After determining the current station, obtain the current wafer production cycle of the current station and the current station anomaly judgment model.

[0067] For the wafer production cycle, it corresponds to the time spent for each batch of wafers to be processed and completed at each station. Therefore, the wafer production cycle can reflect the status of the batch of wafers during production at its corresponding station. When the wafer production cycle of a batch of wafers at a certain station (specifically, it can be the current station) is obtained, it can be determined whether the current wafer production cycle of the batch of wafers at the current station is in a normal state based on data analysis.

[0068] After determining the current station, based on the current station, screen out the wafer production cycle of the current station from the WIP data, that is, the time when the wafer is processed at the current station, to obtain the current wafer production cycle.

[0069] Specifically, in order to obtain the current wafer production cycle of the current station, an embodiment of the present application further provides a data analysis method. Please refer to Figure 6 , Figure 6 which Figure 1 is a schematic flowchart of obtaining the current wafer production cycle of the current station by the data analysis method shown.

[0070] As shown in the figure, in order to obtain the current wafer production cycle of the current station, the following steps may specifically be included:

[0071] Step S00, obtain the wafer production cycles of each wafer at each of the stations in the WIP data.

[0072] Based on the foregoing introduction, it can be known that the WIP data includes information of each station, not only including information on the wafer production cycle. Therefore, in order to ensure the convenience of data analysis, all the wafer production cycles of each station included in the WIP data may be summarized, that is, the corresponding wafer production cycles under each station in the WIP data are obtained. Specifically, it may be as Figure 3 shown.

[0073] Step S01, obtain the corresponding wafer production cycles according to the current station to obtain the current wafer production cycle of the current station.

[0074] After obtaining the wafer production cycles of each wafer at each of the stations, further determine the wafer production cycles included under the current station according to the selected current station, that is, obtain the current wafer production cycle.

[0075] In this way, the current wafer production cycle can be obtained very conveniently. When the current station is changed, the new current wafer production cycle can be quickly and accurately obtained according to the changed current station, which is convenient for subsequent processing and improves the processing efficiency.

[0076] It is easy to understand that for the station anomaly judgment model, since the specific conditions of each station are different, the station anomaly judgment models will also be different. Therefore, it is necessary to select the current station anomaly judgment model based on the current station from the station anomaly judgment models corresponding to each station that have been trained.

[0077] It is easy to understand that in order to ensure the accuracy of data analysis, the current station anomaly judgment model is a pre-trained model when performing data analysis.

[0078] In order to implement the training of the station anomaly judgment model, an embodiment of the present application provides a training process of the station anomaly judgment model. Specifically, it may be referred to Figure 7 , Figure 7 which is a schematic flowchart of the training process of the station anomaly judgment model.

[0079] As shown in the figure, the training process of the station anomaly judgment model may include the following steps:

[0080] Step S21: Obtain each historical wafer production cycle of the current station in each piece of historical WIP data, and the reference state of each historical wafer production cycle of the current station corresponding to each piece of the historical WIP data.

[0081] It is easy to understand that in order to train the station anomaly judgment model, a large amount of historical wafer production cycles of the current station in the WIP data, as well as the true state information, that is, the reference state information, of each historical wafer production cycle of the current station corresponding to the generation time of the WIP data are required. For this purpose, the historical WIP data can be used to judge the current station. By determining whether each historical wafer production cycle of each group of historical WIP data corresponds to the corresponding reference state, it can be judged whether the historical wafer production cycle is normal or abnormal.

[0082] Among them, the historical WIP data is the already obtained WIP data that does not need to be analyzed currently. Of course, the historical WIP data contains each historical wafer production cycle of the current station.

[0083] Step S22: Corresponding to each piece of the historical WIP data, use the current station anomaly judgment model to obtain the training state information of each historical wafer production cycle of the current station according to each historical wafer production cycle.

[0084] Input the historical wafer production cycles of the current station in the historical WIP data into the established current station anomaly judgment model. The model will make an initial judgment on the input historical data according to the established initial criteria, and obtain the training state information of each initial historical wafer production cycle. Of course, for the current station, each historical wafer production cycle of each group of historical WIP data will obtain a corresponding training state information.

[0085] In step S23, judge whether the loss between the training state information and the reference state information of each historical wafer production cycle corresponding to each piece of the historical WIP data meets the loss threshold. If not, execute step S24; if so, execute step S25.

[0086] Based on the comparison between the obtained training state information and the reference state information, obtain the loss between the training state information and the reference state information. If the loss meets the loss threshold, then the training of the current station anomaly judgment model is completed and the accuracy meets the requirements, and step S25 is executed. Otherwise, the prediction accuracy of the current station anomaly judgment model does not meet the requirements and further training is required.

[0087] In a specific real-time manner, the loss satisfying the loss threshold can be that the proportion of the number of historical WIP data in which the training status information and the reference status information are the same reaches a certain value, for example: 99%.

[0088] In step S24, adjust the parameters of the current site anomaly judgment model and execute step S22.

[0089] When the loss of the training status information and the reference status information corresponding to each of the historical WIP data does not satisfy the loss threshold, adjust the parameters of the current site anomaly judgment model, obtain the training status information again, and continue model training.

[0090] In step S25, obtain the trained current site anomaly judgment model.

[0091] When the loss of the training status information and the reference status information corresponding to each of the historical WIP data satisfies the loss threshold, obtain the trained current site anomaly judgment model.

[0092] It can be seen that through the training of the current site anomaly judgment model, the current site anomaly judgment model learns the ability to predict the status information of the site, which can ensure the accuracy of prediction using the current site anomaly judgment model.

[0093] In a specific embodiment, the site anomaly judgment model can be a K-means algorithm model to improve the reliability and stability of the site anomaly judgment model.

[0094] Of course, in other embodiments, other algorithm models can also be used.

[0095] Step S12, use the current site anomaly judgment model to determine whether the current wafer production cycle includes an abnormal wafer production cycle. If so, execute step S13; if not, execute step S14.

[0096] Use the current site anomaly judgment model to judge each current wafer production cycle of the current site to determine whether there is an abnormal wafer production cycle. If there is, then execute step S13; otherwise, execute step S14.

[0097] Step S13, obtain the abnormal wafer production cycle and determine that the predicted status information of the current site is abnormal.

[0098] Step S14, determine that the predicted status information of the current site is normal. Use the current site anomaly judgment model to judge the current wafer production cycle. When there is a wafer production cycle in the current wafer production cycle that is inconsistent with the time value of a normal or regular wafer production cycle, record the site information corresponding to the wafer production cycle with anomalies in the current wafer production cycle.

[0099] In this way, for the data analysis method provided by the embodiments of the present application, on the one hand, the predicted status information of the current site is predicted by using the current site anomaly judgment model, without relying on manual labor, and the current site anomaly judgment model is pre-trained and has high prediction accuracy. Therefore, the predicted status information of the current site can be obtained accurately and reliably, improving the accuracy of integrated circuit production data analysis; on the other hand, the analysis speed of the current site anomaly judgment model is very fast. Even for a large amount of WIP data, the current site anomaly judgment model can be used to perform real-time analysis to obtain the analysis results, so that the status information of the current site can be obtained quickly and fed back to the foundry company in a timely manner. It can be seen that the data analysis method provided by the embodiments of the present application can use the pre-trained current site anomaly judgment model to obtain the predicted status information of the current site, thereby improving the accuracy and real-time performance of WIP data analysis in integrated circuit production.

[0100] In other embodiments, since there are a large number of sites included in the WIP data, in order to understand the predicted status information of each site, the wafer production cycles of all sites in the WIP data can also be predicted for the status information respectively.

[0101] Specifically, please continue to refer to Figure 1 。

[0102] As shown in the figure, the data analysis method provided by the embodiments of the present application may further include:

[0103] Step S15, determine whether the predicted status information of all the sites in the WIP data is obtained. If so, execute step S16. If not, execute step S10 again.

[0104] When the predicted status information of all the sites in the WIP data is obtained, it means that the predicted status information of all the sites in the WIP data has been predicted. Then execute step S16 to count the sites with anomalies among all the sites and uniformly feedback them to the foundry manufacturer, which is convenient for the foundry manufacturer to perform one-time maintenance and improve the efficiency of equipment maintenance; when the predicted status information of all the sites in the WIP data is not obtained, it means that the status information of some sites in the WIP data has not been predicted, so step S10 can be executed again.

[0105] It is easy to understand that at this time, step S10 determines the current station among the respective stations of the WIP data, and the determined current station is the station where the prediction status information has not been obtained.

[0106] Step S16 obtains all the abnormal stations in the WIP data.

[0107] In this way, the abnormal status information of all stations included in the input WIP data can be quickly obtained, and manual operation is not required, improving the efficiency of data analysis.

[0108] To facilitate understanding of the data analysis method provided by the embodiments of the present application, the following uses the K-means algorithm model to implement the acquisition of prediction station information. Specifically, please refer to Figure 8 , Figure 8 which is an implementation process of the data analysis method provided by the embodiments of the present application.

[0109] As shown in the figure, the WIP data contains a large number of wafer lots, the respective station information corresponding to each wafer lot during production, and the wafer production cycle when the wafer lot is produced at the station.

[0110] Since the amount of information contained in the WIP data is very large, for the convenience of display, Figure 8 only a part of the information in the WIP data is exemplarily shown in

[0111] for explanation. Among them, lot is the wafer lot, and the batch labels E, F, G, H, I, J, K, L, E, F, G, H shown in the figure correspond to a batch of wafers, which is convenient for the auxiliary identification of subsequent wafer production data information.

[0112] As shown in the figure, 1.0, 2.0 represent the labels of the stations required for producing the above-mentioned respective wafer lots.

[0113] The station cycletime2, 3, 4, 2... represents the wafer production cycle of the corresponding station described herein. The unit of the production cycle can be days or hours, specifically based on the unit provided by the manufacturer. It can be seen that in the WIP data, each wafer lot corresponds to a station and the wafer production cycle under that station.

[0114] When performing the data analysis method:

[0115] First, determine the current station of each station according to the information in the WIP data. According to Figure 8 shown, that is, according to the information in the WIP data, a station can be randomly selected, such as station 1.0 as the current station. It can be seen that Figure 8The lots included in Site 1.0 are: E, F, G, H, I, J, K, L, and the cycle time of each lot corresponding to the site: 2, 3, 4, 2, 2, 3, 5, 6.

[0116] Collect the above information as the information included in the current Site 1.0, which is the implementation of the content described in steps S10 and S11.

[0117] Then, obtain the current site anomaly judgment model of the current site. Continuing to take Site 1.0 as the current site as an example, the current site anomaly judgment model is the K-means algorithm model, and it is the trained current site anomaly judgment model of Site 1.0, which is the implementation of the content described in part of step S11.

[0118] Use the K-means algorithm model of Site 1.0 to judge the start of the current wafer production cycle of the input current site. If an abnormal wafer production cycle in the current site is obtained, then record the prediction status information of Site 1.0 as abnormal, which is the implementation of step S12 ( Figure 8 not shown in the figure).

[0119] Finally, replace the current site, for example, replace it with step S20, and make predictions again until the prediction status information of all sites in the WIP data is obtained.

[0120] In addition to obtaining the prediction status information of the site by using the wafer production cycle, in order to obtain more information to understand the production status of the wafer and conduct control, in one implementation, the data analysis method provided by the embodiments of the present application can also estimate the shipping time of each wafer lot. Please refer to Figure 9 , Figure 9 which is another flow schematic diagram of the data analysis method provided by the embodiments of the present application.

[0121] As shown in the figure, the data analysis method provided by the embodiments of the present application may further include the following steps:

[0122] Step S30, obtain the current lot of each lot in the WIP data.

[0123] As mentioned above, the obtained WIP data includes a large amount of site information and lot information. In order to understand the shipping time of the wafer, it is necessary to determine the specific lot of the wafer, that is, the current lot, to obtain the wafer lot information to be obtained.

[0124] Step S31, obtain the time when the current lot enters the production line and the lot sites, where the lot sites are each of the sites required for processing the current lot.

[0125] The time of entering the production line represents the time when the current batch starts production; the batch stations are the various stations that the current batch needs to pass through during processing. For details, please refer to Figure 2 the data information included in the lot shown and the data information included in the time of entering the production line.

[0126] After determining the current batch, relevant information can be obtained according to the current batch.

[0127] Step S32, obtain the average wafer production cycle of the batch stations, where the average wafer production cycle is the wafer production cycle obtained in advance.

[0128] After obtaining the batch stations corresponding to the current batch, further obtain the average wafer production cycle of each station included in the batch stations.

[0129] In a specific embodiment, the obtaining step of the average wafer production cycle may be:

[0130] Obtain the historical wafer production cycles of each station in the historical WIP data;

[0131] According to each station, obtain the respective historical wafer production cycles corresponding to the same station, and obtain the historical wafer production cycles of each station;

[0132] Obtain the average value of the historical wafer production cycles of each station to obtain the average wafer production cycle corresponding to the station and the average wafer production cycles of each station.

[0133] For easy understanding, refer to Figure 10 for illustration. Figure 10 It is a schematic diagram of partial station information included in the historical WIP data.

[0134] As Figure 10 shown, the wafer production cycles corresponding to each batch of station 1.0 are 4, 2, 2, 3, 5, 2, and the wafer production cycles corresponding to each batch of station 2.0 are 3, 2, 4.

[0135] The average value of the historical wafer production cycles of station 1.0 is (4 + 2 + 2 + 3 + 5 + 2) / 6 = 3. Similarly, the average value of the historical wafer production cycles of station 2.0 is (3 + 2 + 4) / 3 = 3.

[0136] The above is only for illustrating the calculation of the average wafer production cycle, and the specific values are based on the data included in the actual WIP data.

[0137] Of course, since there may be abnormal data in the wafer production cycle in the historical WIP data, in order to improve the accuracy, the abnormal wafer production cycles in the historical WIP data can be removed first, and then the average wafer production cycle can be obtained according to the foregoing method.

[0138] Step S33: Obtain the shipping time of the current batch according to the entry time into the production line and the average wafer production cycle of the batch site, and replace the current batch until the shipping times of all batches are obtained.

[0139] Among them, the calculation of the shipping time is carried out according to the following formula, and the formula is expressed as:

[0140]

[0141] Among them, t ship represents the estimated shipping time, t in represents the entry time into the production line, n represents the number of sites under the current batch, and ct k represents the average wafer production cycle of each site.

[0142] Of course, according to the average wafer production cycle of all sites under the input current batch and the entry time of this batch, the shipping time of the current batch can be obtained. After calculating the shipping time of the current batch, the next batch that needs to estimate the shipping time can be used as the new current batch and then calculated until the shipping times of all batches are completed.

[0143] In this way, it is convenient for the manufacturer or the integrated circuit chip design manufacturer to understand the production line status of wafer production, understand the production capacity of the manufacturer, reasonably arrange the subsequent production capacity such as testing and packaging, timely adjust the shipping plan of the end product, and facilitate the management of the integrated circuit design manufacturer and the manufacturer.

[0144] In another embodiment, in order to obtain more site information, facilitate understanding of the wafer fragmentation situation, quickly find the reason, feedback to the foundry manufacturer, and repair the corresponding production equipment, the data analysis method provided by the embodiment of the present application can also analyze the fragmentation information according to the WIP data. Please refer to Figure 11 , Figure 11 which is another flow diagram of the data analysis method provided by the embodiment of the present application.

[0145] As shown in the figure, the data analysis method provided by the embodiment of the present application may further include:

[0146] Step S40: Obtain the fragmentation information in the WIP data, where the fragmentation information includes the fragmentation batch, fragmentation site, fragmentation time, and fragmented wafer number.

[0147] Based on the foregoing analysis, it can be seen that after processing the original WIP data, WIP data containing fragment information can be obtained, as Figure 4 shown.

[0148] Step S41: According to the fragment information, obtain the site where the fragments are generated, and obtain the fragment site and the site fragment information corresponding to the fragment site.

[0149] Based on the foregoing fragment information, further processing is performed to obtain the site where the fragments are generated, so as to obtain the fragment site and the site fragment information corresponding to the fragment site. For details, please refer to Figure 12 , Figure 12 which Figure 4 is a schematic diagram of the processed data of the fragment information shown in

[0150] Specifically, the above process can be implemented by a python script to improve convenience. Of course, it can also be implemented by other means.

[0151] Thus, based on the site fragment information, the high-incidence sites of the fragments can be further understood, and then the fragment equipment can be determined and inspected and repaired to reduce the fragment rate in the subsequent wafer production process.

[0152] In addition, during the wafer production process, there is also a process in which the same batch is divided into multiple batches for production during the production process. In order to fully understand the sub-batch and batch splitting situations of each site in the wafer production, the data analysis method provided in the embodiments of the present application can also analyze the batch splitting information in the WIP data. Please refer to Figure 13 , Figure 13 which

[0153] As shown in the figure, the data analysis method provided in the embodiments of the present application may further include:

[0154] Step S50: Obtain the batch splitting information in the WIP data, where the batch splitting information includes batch splitting batches, batch splitting sites, batch splitting times, and batch splitting wafer numbers.

[0155] The batch splitting information in the WIP data is as Figure 5 shown.

[0156] Step S51: According to the batch splitting information, obtain the sites of the batch splitting to obtain the batch splitting sites and the site batch splitting information corresponding to the batch splitting sites.

[0157] Through statistical analysis of the batch splitting information in the WIP data, the site information where the batch splitting occurs is obtained. For details, please refer to Figure 14 , Figure 14 which Figure 5Schematic diagram of data after batch information processing as shown

[0158] The batch information refers to the situation of parent and child batches. Recording the batch site, batch time, batch number of the parent and child batches, and the number of the batch wafers can facilitate tracing the cause of anomalies when integrated circuit chips are abnormal in the future, save unnecessary inspection time, quickly locate the site where problems may occur, and conduct inspection and repair

[0159] The above describes multiple embodiment solutions provided by the embodiments of the present application. The optional methods described in each embodiment solution can be combined and cross-referenced with each other without conflict, so as to extend multiple possible embodiment solutions, all of which can be considered as the embodiment solutions disclosed and made public by the embodiments of the present application

[0160] To solve the foregoing problems, the embodiments of the present application further provide a data analysis device, which can be considered as a functional module required to implement the data analysis method provided by the embodiments of the present application. The device content described below can be correspondingly referred to the method content described above

[0161] As an optional implementation, please refer to Figure 15 , Figure 15 which is an optional block diagram of the data analysis device provided by the embodiments of the present application

[0162] As shown in Figure 15 , the data analysis device provided by the embodiments of the present application may include

[0163] A current site determination module 901, adapted to determine the current site among the respective sites of the WIP data

[0164] According to the foregoing content, it can be known that the WIP data contains a large amount of information such as wafer batches and sites. Therefore, it is necessary to determine the current site in the WIP data and determine the current data to be analyzed

[0165] An acquisition module 902, adapted to acquire the current wafer production cycle of the current site and the current site anomaly judgment model, wherein the current wafer production cycle is the respective wafer production cycles of the current site in the WIP data, and the current site anomaly judgment model is the site anomaly judgment model of the current site that has been pre-trained

[0166] A current site status acquisition module 903, adapted to use the current site anomaly judgment model to obtain the abnormal wafer production cycle when it is determined that the current wafer production cycle includes an abnormal wafer production cycle, and determine that the predicted status information of the current site is abnormal

[0167] In this way, for the data analysis method provided by the embodiments of the present application, on the one hand, the current site anomaly judgment model is used to predict the status information of the current site without relying on manual labor. Moreover, the current site anomaly judgment model is pre-trained and has high prediction accuracy. Therefore, the predicted status information of the current site can be accurately and reliably obtained, improving the accuracy of integrated circuit production data analysis. On the other hand, the analysis speed of the current site anomaly judgment model is very fast. Even for a large amount of WIP data, the current site anomaly judgment model can be used to perform real-time analysis and obtain the analysis results, so that the status information of the current site can be quickly obtained and timely feedback can be provided to the foundry company. It can be seen that the data analysis method provided by the embodiments of the present application can use the pre-trained current site anomaly judgment model to obtain the predicted status information of the current site, thereby improving the accuracy and real-time performance of WIP data analysis for integrated circuit production.

[0168] In some embodiments, the obtaining module 902 is further adapted to: obtain each wafer production cycle of each site in the WIP data; obtain the corresponding wafer production cycles according to the current site, and obtain the current wafer production cycle of the current site.

[0169] In actual production work, since there are a large number of wafer production cycles obtained from foundry manufacturers, when performing site anomaly judgment, the site that is actually needed or arbitrarily determined can be used as the current site. Then, the wafer production cycles included in the selected current site are determined, and finally, the wafer production cycle to be analyzed is determined as the current wafer production cycle among the obtained wafer production cycles. In this way, the wafer production cycles included in all sites are centrally summarized to ensure the integrity of the input data of the current site anomaly judgment model and obtain a more comprehensive analysis result.

[0170] After obtaining the anomaly status information of the current site, since the WIP data contains a large amount of site information, in order to comprehensively grasp the anomaly status information of each site, in one embodiment, the data analysis device can also perform anomaly judgment on all sites. For details, please continue to refer to Figure 15 .

[0171] As shown in the figure, the data analysis device may further include:

[0172] An all-site status information obtaining module 904, which is adapted to obtain all abnormal sites in the WIP data after obtaining the predicted status information of all sites in the WIP data according to the current site anomaly judgment module.

[0173] In this way, the abnormal status information of all stations included in the input WIP data can be obtained, and the stations with abnormalities among all stations are counted and uniformly fed back to the foundry manufacturer, facilitating a single overhaul by the foundry manufacturer and improving the efficiency of equipment overhaul.

[0174] The information contained in the WIP data is extensive and messy. In one implementation, to facilitate subsequent data analysis, the WIP data can also be processed. Please continue to refer to Figure 15 。

[0175] As shown in the figure, the data analysis device further includes a WIP data acquisition module 900, which is adapted to:

[0176] Acquire the original WIP data, where the original WIP data includes all production information of the wafers;

[0177] Perform preprocessing on the original WIP data, including format adjustment and noise reduction;

[0178] Extract the information required for data analysis from the preprocessed original WIP data to obtain the WIP data.

[0179] In this way, the processed WIP data includes target data that can be used for integrated circuit design, such as information about each station, wafer lot information, fragment information, and batch information, etc., which can be used to judge or analyze production problems of integrated circuit chips.

[0180] To ensure the accuracy of the current station abnormal judgment model, the current station abnormal judgment model is trained by a model training module, and the model training module is adapted to:

[0181] Acquire each historical wafer production cycle of the current station in each historical WIP data, and the reference status of each historical wafer production cycle of the current station corresponding to each historical WIP data;

[0182] For each historical WIP data, use the current station abnormal judgment model to obtain the training status information of each historical wafer production cycle of the current station according to each historical wafer production cycle;

[0183] When the loss between the training status information and the reference status information of each historical wafer production cycle corresponding to each historical WIP data meets the loss threshold, the trained current station abnormal judgment model is obtained.

[0184] To facilitate the understanding of model training, here we continue to use the K-means algorithm as an example. For instance, the K-means algorithm uses the wafer production cycle 5 as the initial centroid, that is, the wafer production cycle as the judgment criterion to judge all the wafer production cycles in the historical WIP data. The initial output result shows that the wafer lot J under station 1.0 corresponding to the wafer production cycle 3 is an abnormal station. However, in reality, the status of the wafer lot J under station 1.0 is a normal station. Therefore, the output result is incorrect. At this time, the K-means algorithm will reselect a centroid, that is, the wafer production cycle, as the new judgment criterion until the output result is consistent with the result recorded in the historical WIP data, and a trained current station anomaly judgment model is obtained.

[0185] Of course, in its implementation, it can also be to adjust the model by manually adjusting the input parameters of the model to achieve the correct output result, or use other algorithms such as neural networks, machine learning and other algorithms.

[0186] By continuously training the model using historical WIP data, a model with reliable output results is obtained, ensuring the accuracy and reliability of the output results when the subsequent model is used.

[0187] In order to obtain more analysis of integrated circuit production data and get a more comprehensive analysis result, in one implementation, the shipping time of each wafer lot can also be estimated. Please refer to Figure 16 , Figure 16 which is another optional block diagram of the data analysis device provided by the embodiments of the present application.

[0188] The current batch acquisition module 100 is adapted to acquire the current batches of each batch in the WIP data;

[0189] The status information acquisition module 101 of the current batch is adapted to acquire the time when the current batch enters the production line and the batch stations, where the batch stations are each of the stations required for processing the current batch;

[0190] The average wafer production cycle acquisition module 102 is adapted to acquire the average wafer production cycle of the batch stations, and the average wafer production cycle is the wafer production cycle obtained in advance;

[0191] The shipping time estimation module 103 is adapted to acquire the shipping time of the current batch according to the time when the current batch enters the production line and the average wafer production cycle of the batch stations, and replace the current batch until the shipping times of each batch are obtained.

[0192] The shipping time of the current batch can be obtained based on the average wafer production cycle of all sites in the current batch of input and the time when the batch enters the production line. After calculating the shipping time of the current batch, the next batch for which the shipping time needs to be estimated is taken as the new current batch and calculated again until the shipping times of all batches are completed.

[0193] In this way, it is convenient for manufacturers or integrated circuit chip aggregators to understand more information about integrated circuits, provide useful information for integrated circuit chip design manufacturers to determine the next plan, and facilitate the management of integrated circuit design manufacturers and manufacturers.

[0194] In order to estimate the shipping time, it is necessary to calculate the average wafer production cycle, which is obtained by a pre-acquisition module. The pre-acquisition module is adapted to:

[0195] Obtain the historical wafer production cycles of each site in the historical WIP data;

[0196] According to each site, obtain the respective historical wafer production cycles corresponding to the same site to obtain the historical wafer production cycles of each site;

[0197] Obtain the average value of the historical wafer production cycles of each site to obtain the average wafer production cycle corresponding to the site and the average wafer production cycles of each site.

[0198] In order to obtain more analysis of integrated circuit production data and obtain a more comprehensive analysis result, in one embodiment, the shipping time of each wafer batch can also be estimated. Please refer to Figure 17 , Figure 17 which is another optional block diagram of the data analysis device provided by the embodiments of the present application.

[0199] As shown in the figure, it includes:

[0200] A fragment information acquisition module 110, adapted to acquire the fragment information in the WIP data, where the fragment information includes a fragment batch, a fragment site, a fragment time, and a fragment wafer number;

[0201] A fragment site status information acquisition module 111, adapted to obtain the site where the fragment is generated according to the fragment information to obtain the fragment site and the site fragment information corresponding to the fragment site.

[0202] The recording of fragment information can assist in finding the cause of the problem and determining the problem device when an abnormality occurs in an integrated circuit chip subsequently.

[0203] In order to record more data information that may occur in the abnormal integrated circuit chip, in one implementation, statistical analysis can also be performed on the batch information. For details, please refer to Figure 18 , Figure 18 which is another optional block diagram of the data analysis device provided by the embodiments of the present application.

[0204] As shown in the figure, the data analysis device may further include:

[0205] A batch information acquisition module 120, adapted to acquire the batch information in the WIP data, where the batch information includes batch number, batch site, batch time, and batch wafer number;

[0206] A site batch status information acquisition module 121, adapted to obtain the batches of sites according to the batch information, so as to obtain the batch sites and the site batch information corresponding to the batch sites.

[0207] In this way, by analyzing the wafer production cycle in the WIP data, the corresponding site and batch information of the abnormal wafer production cycle can be obtained. It enables the integrated circuit design manufacturer to quickly find the site where the abnormal problem occurs and record the wafer batch information of the site where the abnormality occurs, and feedback all the recorded abnormal site information to the foundry manufacturer for repairing the equipment under the abnormal site. By calculating the average value of the wafer production cycle in the WIP data and using the production time of this batch entering the production line to calculate the shipment time of this batch, it can provide planning information for the integrated circuit design manufacturer and the production and processing mall, facilitating the provision of a basis for the subsequent production or scheduling of integrated circuits. In addition, statistical analysis is separately performed on the broken wafer information and batch information in the WIP data, which facilitates quickly finding the source and cause of the abnormality when there is an abnormality in the production of integrated circuit chips in the subsequent process, saving the time for finding the abnormal site.

[0208] The embodiments of the present application also provide a storage medium, which stores a program suitable for data analysis to implement the data analysis method as described above.

[0209] The embodiments of the present application also provide an electronic device, including at least one memory and at least one processor; the memory stores a program, and the processor calls the program to execute the data analysis method as described above.

[0210] Although the embodiments of the present application are disclosed as above, the present application 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 application. Therefore, the protection scope of the present application should be subject to the scope defined by the claims.

Claims

1. A data analysis method, characterized in that, Applicable to the analysis of WIP data, where the WIP data contains chip production data of multiple sites, including: Determine the current site among the various sites of the WIP data; Obtain the current wafer production cycle of the current site and the current site anomaly judgment model, where the current wafer production cycle is the respective wafer production cycles of the current site in the WIP data, and the current site anomaly judgment model is the site anomaly judgment model corresponding to the current site that has been pre-trained; Use the current site anomaly judgment model to determine whether the current wafer production cycle includes an abnormal wafer production cycle. When it is determined that the current wafer production cycle includes an abnormal wafer production cycle, obtain the abnormal wafer production cycle and determine that the predicted status information of the current site is abnormal; Among them, the steps for obtaining the current site anomaly judgment model include: Obtain the respective historical wafer production cycles of the current site in each historical WIP data and the reference status of each historical wafer production cycle of the current site corresponding to each historical WIP data; Corresponding to each historical WIP data, use the current site anomaly judgment model to obtain the training status information of each historical wafer production cycle of the current site according to each historical wafer production cycle; When the loss between the training status information and the reference status information of each historical wafer production cycle corresponding to each historical WIP data meets the loss threshold, obtain the trained current site anomaly judgment model.

2. The data analysis method according to claim 1, wherein The steps for obtaining the current wafer production cycle of the current site include: Obtain the respective wafer production cycles of each site in the WIP data; According to the current site, obtain the corresponding respective wafer production cycles to obtain the current wafer production cycle of the current site.

3. The data analysis method according to claim 1, characterized in that It further includes: When the predicted status information of all the sites in the WIP data is obtained, obtain all the abnormal sites in the WIP data.

4. The data analysis method according to claim 1, characterized in that, The steps for obtaining the WIP data include: Obtain the original WIP data, where the original WIP data includes all the production information of the wafers; Perform preprocessing of format adjustment and noise reduction on the original WIP data; Extract the information required for data analysis from the preprocessed original WIP data to obtain the WIP data.

5. The data analysis method according to claim 1, wherein It further includes: Obtain the current batch of each batch in the WIP data; Obtain the time of entering the production line and the batch sites of the current batch, where the batch sites are the respective sites required for processing the current batch; Obtain the average wafer production cycle of the batch sites, where the average wafer production cycle is the pre-obtained wafer production cycle; According to the time of entering the production line and the average wafer production cycle of the batch sites, obtain the shipping time of the current batch, and replace the current batch until the shipping times of all the batches are obtained.

6. The data analysis method according to claim 5, wherein The steps for obtaining the average wafer production cycle include: Obtain the respective historical wafer production cycles of each site in the historical WIP data; According to each of the said sites, obtain each of the historical wafer production cycles corresponding to the same said site, and obtain the historical wafer production cycles of each site. Obtain the average value of the historical wafer production cycles of each of the said sites, and obtain the average wafer production cycle corresponding to the said site and the average wafer production cycles of each of the said sites.

7. The data analysis method according to claim 1, wherein It further includes: Obtain the fragment information in the WIP data, where the fragment information includes fragment batch, fragment site, fragment time, and fragment wafer number. According to the fragment information, obtain the site where the fragments are generated, and obtain the fragment site and the site fragment information corresponding to the said fragment site.

8. The data analysis method according to claim 1, characterized in that It further includes: Obtain the batching information in the WIP data, where the batching information includes batching batch, batching site, batching time, and batching wafer number. According to the batching information, obtain the site of batching, and obtain the batching site and the site batching information corresponding to the said batching site.

9. A data analysis device, characterized in that, It includes: A current site determination module, adapted to determine the current site among the various sites of the WIP data, where the WIP data contains chip production data of multiple sites. An acquisition module, adapted to acquire the current wafer production cycle of the current site and the current site anomaly judgment model, where the current wafer production cycle is each of the wafer production cycles of the current site in the WIP data, and the current site anomaly judgment model is the site anomaly judgment model corresponding to the current site that has been pre-trained. A current site status acquisition module, adapted to use the current site anomaly judgment model to determine whether the current wafer production cycle includes an abnormal wafer production cycle. When it is determined that the current wafer production cycle includes an abnormal wafer production cycle, obtain the abnormal wafer production cycle and determine that the predicted status information of the current site is abnormal. The current site anomaly judgment model is trained by a model training module, and the model training module is adapted to: Obtain each of the historical wafer production cycles of the current site in each historical WIP data, and the reference status of each of the historical wafer production cycles of the current site corresponding to each of the historical WIP data. Corresponding to each of the historical WIP data, use the current site anomaly judgment model to obtain the training status information of each of the historical wafer production cycles of the current site according to each of the historical wafer production cycles. When the loss between the training status information and the reference status information of each of the historical wafer production cycles corresponding to each of the historical WIP data meets the loss threshold, obtain the trained current site anomaly judgment model.

10. The data analysis device according to claim 9, characterized in that The acquisition module, adapted to acquire the current wafer production cycle of the current site, includes: Obtain each of the wafer production cycles of each of the sites in the WIP data. According to the current site, obtain the corresponding each of the wafer production cycles, and obtain the current wafer production cycle of the current site.

11. The data analysis device according to claim 9, wherein It further includes: An all-site status information acquisition module, adapted to obtain all the abnormal sites in the WIP data after obtaining the predicted status information of all the sites in the WIP data according to the current site anomaly judgment module.

12. The data analysis device according to claim 9, characterized in that It further includes a WIP data acquisition module, and the WIP data acquisition module is adapted to: acquire original WIP data, where the original WIP data includes all production information of the wafers; perform preprocessing of format adjustment and noise reduction on the original WIP data; extract information required for data analysis from the preprocessed original WIP data to obtain the WIP data.

13. The data analysis device according to claim 9, characterized in that, It further includes: a current batch acquisition module, adapted to acquire the current batch of each batch in the WIP data; a status information acquisition module for the current batch, adapted to acquire the time of entering the production line and the batch sites of the current batch, where the batch sites are each of the sites required for processing the current batch; an average wafer production cycle acquisition module, adapted to acquire the average wafer production cycle of the batch sites, and the average wafer production cycle is the wafer production cycle obtained in advance; a shipping time estimation module, adapted to obtain the shipping time of the current batch according to the time of entering the production line and the average wafer production cycle of the batch sites, and replace the current batch until the shipping times of all batches are obtained.

14. The data analysis device according to claim 9, wherein The average wafer production cycle is obtained through a pre-acquisition module, and the pre-acquisition module is adapted to: acquire each historical wafer production cycle of each of the sites in the historical WIP data; obtain, according to each of the sites, each of the historical wafer production cycles corresponding to the same site to obtain the historical wafer production cycles of each site; acquire the average value of the historical wafer production cycles of each of the sites to obtain the average wafer production cycle corresponding to the site and the average wafer production cycles of each of the sites.

15. The data analysis device according to claim 9, wherein It further includes: a fragment information acquisition module, adapted to acquire the fragment information in the WIP data, where the fragment information includes the fragment batch, the fragment site, the fragment time, and the fragment wafer number; a fragment site status information acquisition module, adapted to obtain the site where the fragment is generated according to the fragment information to obtain the fragment site and the site fragment information corresponding to the fragment site.

16. The data analysis device according to claim 9, characterized in that, It further includes: a batching information acquisition module, adapted to acquire the batching information in the WIP data, where the batching information includes the batching batch, the batching site, the batching time, and the batching wafer number; a site batching status information acquisition module, adapted to obtain the site where batching occurs according to the batching information to obtain the batching site and the site batching information corresponding to the batching site.

17. A storage medium, characterized in that, The storage medium stores a program suitable for data analysis to implement the data analysis method according to any one of claims 1-8.

18. An electronic device, characterized in that, It includes at least one memory and at least one processor; the memory stores a program, and the processor calls the program to execute the data analysis method according to any one of claims 1-8.

Images