Intelligent silicon-based OLED yield prediction method and device and storage medium

Abstract

The invention relates to the technical field of data processing, and discloses an intelligent silicon-based OLED yield prediction method and device and a storage medium. The method comprises the following steps: acquiring production data of each intelligent silicon-based OLED; deleting a first data type and a type value corresponding to the first data type in the production data, wherein the obtained target data comprises multiple second data types and type values corresponding to the second data types; transposing the first data format of the target data into a second data format, wherein the data in the row direction in the obtained transposed data comprises type values of a plurality of second data types of the same intelligent silicon-based OLED; training based on the transposed data to obtain a bad prediction model; determining a Shapley value of each feature type in a plurality of second data types based on the transposed data and a bad prediction model; generating multi-dimensional interpretation information based on the Shapley value of each feature type; the training efficiency of the bad prediction model and the interpretability of yield prediction can be improved.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing technology, and in particular to an intelligent silicon-based OLED yield prediction method, device and storage medium. Background Art

[0002] Smart silicon-based organic light-emitting diode (OLED) is a new type of display device that combines semiconductor technology and OLED display technology. Smart silicon-based OLED is made of single-crystal silicon as the active driving backplane. In order to improve the product qualification rate and production efficiency of smart silicon-based OLED, it is necessary to perform yield management on smart silicon-based OLED. Yield management of smart silicon-based OLED refers to a series of measures taken during the production process to improve the qualification rate and production efficiency of smart silicon-based OLED and reduce waste and rework. By performing yield management on smart silicon-based OLED, product quality can be improved.

[0003] In traditional semiconductor yield management, production procedures and process parameters are generally adjusted based on manual experience, and data-driven methods are used to assist in analyzing and predicting yield issues in the production process. The data-driven method collects and analyzes a large amount of production data to identify the root causes of defects, predict future yield trends, and take corresponding improvement measures.

[0004] However, the production data of smart silicon-based OLEDs collected through traditional data-driven methods is complex, resulting in low efficiency in the analysis and prediction of yield issues. Summary of the invention

[0005] In view of this, the present disclosure proposes an intelligent silicon-based OLED yield prediction method, device and storage medium, which can solve the problem of low efficiency in predicting yield using traditional data-driven methods, improve the training efficiency of defect prediction models, and thus improve the yield prediction efficiency. At the same time, multi-dimensional explanatory information can be generated based on the Shapley value of each feature type, thereby intuitively reflecting the root cause of the defect type and improving the interpretability of yield prediction.

[0006] According to one aspect of the present disclosure, a smart silicon-based OLED yield prediction method is provided, the method comprising:

[0007] Acquire production data of each smart silicon-based OLED; wherein the production data includes multiple data types and type values ​​corresponding to each data type;

[0008] Deleting a first data type and a type value corresponding to the first data type in the production data to obtain target data, wherein the target data includes a plurality of second data types and type values ​​corresponding to the second data types;

[0009] Transposing the first data format of the target data into a second data format to obtain transposed data; the transposed data is N×M structured data, and the row-wise data in the transposed data includes type values ​​of multiple second data types of the same smart silicon-based OLED; both N and M are integers greater than 1;

[0010] A bad prediction model is obtained by training based on the first part of the transposed data;

[0011] Determine, based on the second portion of data in the transposed data and the bad prediction model, a Shapley value of each feature type in the plurality of second data types;

[0012] Multi-dimensional explanation information is generated based on the Shapley value of each feature type, and the multi-dimensional explanation information is used to describe the influence of each feature type on yield prediction from different dimensions.

[0013] In a possible implementation, the target data is structured data represented by a plurality of first data tables, and the row-wise data in each first data table includes a type identifier of a second data type of the same smart silicon-based OLED and a type value corresponding to the second data type; accordingly,

[0014] The step of transposing the first data format of the target data into a second data format to obtain transposed data includes:

[0015] Creating a second data table corresponding to the second data format, wherein the first column of the second data table represents the product identification of each smart silicon-based OLED, the first row represents the type identification of each second data type, and the type identifications are arranged in a preset order;

[0016] For each type identifier and type value in the same row of each first data table corresponding to each product identifier, the type value is converted to the position indicated by the product identifier and the type identifier in the second data table to obtain the transposed data.

[0017] In a possible implementation manner, before the training based on the first part of the transposed data to obtain the bad prediction model, the method further includes:

[0018] Performing at least one level of screening on the transposed data to obtain screened transposed data;

[0019] The first portion of data and the second portion of data are determined based on the filtered transposed data.

[0020] In a possible implementation, the performing at least one level of screening on the transposed data to obtain screened transposed data includes:

[0021] Determining the variance of the type values ​​corresponding to each second data type;

[0022] Based on the comparison result between the variance and a preset variance threshold, performing a primary screening on the transposed data;

[0023] For any two second data types after the primary screening, determining the correlation between the two second data types based on the type values ​​corresponding to the two second data types;

[0024] Based on the comparison result between the correlation and a preset correlation threshold, the transposed data is subjected to secondary screening to obtain the screened transposed data.

[0025] In a possible implementation, the multiple second data types include the feature type and the output type, the feature type is the second data type input to the bad prediction model, and the output type is the second data type corresponding to the prediction result of the bad prediction model; the multi-dimensional explanation information includes a Shapley overall ranking diagram, and the Shapley overall ranking diagram is used to indicate the feature type that causes the bad result and the output type that the bad prediction model can predict; accordingly,

[0026] The multi-dimensional explanation information is generated based on the Shapley value of each feature type, including:

[0027] Determine the mean absolute value of the Shapley value for each feature type;

[0028] For each output type, determining the proportion of the output type in the average absolute value of each feature type;

[0029] The first coordinate axis of the Shapley overall ranking diagram displays each feature type, the second coordinate axis displays the average absolute value corresponding to each feature type, and the color legend displays the proportion corresponding to various output types; wherein the first coordinate axis is perpendicular to the second coordinate axis.

[0030] In a possible implementation, the multiple second data types include the feature type and the output type, the feature type is the second data type input to the bad prediction model, and the output type is the second data type corresponding to the prediction result of the bad prediction model; the multi-dimensional explanation information includes a Shapley heat map, and the Shapley heat map is used to indicate the Shapley value ranking of each feature type predicting the output type for each output type to be analyzed predicted by the bad prediction model, and indicates whether each feature type has a positive or negative impact on the output type; accordingly,

[0031] The multi-dimensional explanation information is generated based on the Shapley value of each feature type, including:

[0032] For each output type to be analyzed, determining a first Shapley absolute value when predicting the output type based on each feature type;

[0033] Clustering the second part of each group of data according to explanation similarity to obtain multiple sample sets; wherein explanation similarity refers to multiple groups of data with the same output type and the same feature type;

[0034] Determining a second Shapley absolute value when predicting the output type based on each sample set;

[0035] Determine the prediction result corresponding to each sample set based on the bad prediction model;

[0036] The first coordinate axis of the Shapley heat map displays each feature type, the second coordinate axis displays the first Shapley absolute value corresponding to each feature type, and the third coordinate axis displays each sample set corresponding to each feature type, and the color representation of the second Shapley absolute value is displayed between the first coordinate axis and the third coordinate axis, and the color representation is used to indicate the value of the second Shapley absolute value; wherein the first coordinate axis and the second coordinate axis are perpendicular to the third coordinate axis;

[0037] The change curve of the prediction result corresponding to each sample set is displayed at a specified position of the Shapley heat map.

[0038] In a possible implementation, the multiple second data types include the feature type and the output type, the feature type is the second data type input to the bad prediction model, and the output type is the second data type corresponding to the prediction result of the bad prediction model; the multi-dimensional explanation information includes a Shapley decision diagram, and the Shapley decision diagram is used to indicate the change process of the prediction result after gradually adding each feature type for each output type to be analyzed that can be predicted by the bad prediction model; accordingly,

[0039] The multi-dimensional explanation information is generated based on the Shapley value of each feature type, including:

[0040] For each group of the second part of data, the type values ​​corresponding to each feature type in the second part of data are added to the input data one by one according to the preset input order, and the prediction results corresponding to various input data are respectively predicted by the bad prediction model to obtain the change process of the prediction results corresponding to the second part of data;

[0041] Clustering the changing processes of different prediction results that meet the preset conditions to obtain the changing processes of the clustered prediction results;

[0042] Get the mean of the prediction results corresponding to the second part of each group of data;

[0043] The first coordinate axis of the Shapley decision diagram displays each feature type in the input order, and the second coordinate axis displays each prediction result, and based on the change process of the prediction result after clustering, the prediction value change curve starts from the mean of the prediction result on the second coordinate axis and is extended along the direction of the first coordinate axis; wherein the first coordinate axis is perpendicular to the second coordinate axis.

[0044] In a possible implementation, the multiple second data types include the feature type and the output type, the feature type is the second data type input to the bad prediction model, and the output type is the second data type corresponding to the prediction result of the bad prediction model; the multi-dimensional explanation information includes a Shapley dependency graph, and the Shapley dependency graph is used to indicate, for each output type to be analyzed that can be predicted by the bad prediction model, how the Shapley value corresponding to the type value of the feature type to be analyzed is affected by the type value of another feature type; accordingly,

[0045] The multi-dimensional explanation information is generated based on the Shapley value of each feature type, including:

[0046] For the output type to be analyzed and the feature type to be analyzed, each Shapley value is displayed through the first coordinate axis of the Shapley dependency graph, and different type values ​​corresponding to the feature type are displayed through the second coordinate axis, and color points are displayed between the first coordinate axis and the second coordinate axis. The position of the color point is determined by the Shapley value corresponding to the type value of the feature type to be analyzed when predicting the output type to be analyzed, and the color of the color point indicates the size of the type value of the other feature type.

[0047] According to another aspect of the present disclosure, there is provided an intelligent silicon-based OLED yield prediction device, comprising: a processor; and a memory for storing processor executable instructions; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0048] According to another aspect of the present disclosure, a non-volatile computer-readable storage medium is provided, on which computer program instructions are stored, wherein the computer program instructions implement the above method when executed by a processor.

[0049] According to another aspect of the present disclosure, a computer program product is provided, including a computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of an electronic device, the processor in the electronic device executes the above method.

[0050] The method obtains production data of each smart silicon-based OLED; deletes the first data type and the type value corresponding to the first data type in the production data to obtain target data, wherein the target data includes multiple second data types and type values ​​corresponding to the second data types; transposes the first data format of the target data to the second data format to obtain transposed data; the transposed data is N×M structured data, and the data in the row direction of the transposed data includes type values ​​of multiple second data types of the same smart silicon-based OLED; a defect prediction model is obtained by training based on the first part of the data in the transposed data; the Shapley value of each feature type in the multiple second data types is determined based on the second part of the data in the transposed data and the defect prediction model; and multi-dimensional explanation information is generated based on the Shapley value of each feature type; after the first data format of the target data is transposed to the second data format, the type value of the same row can be directly used for the training of the defect prediction model, without calling the target data in the first data format to find the second data type corresponding to each type value during the training process. Therefore, the problem of low efficiency of the traditional data-driven method in predicting yield can be solved, and the training efficiency of the defect prediction model can be improved, thereby improving the yield prediction efficiency.

[0051] In addition, by generating multi-dimensional explanatory information based on the Shapley value of each feature type, the root cause of the defective type can be intuitively reflected, thereby improving the interpretability of yield prediction.

[0052] Further features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

[0054] Figure 1 A flow chart showing a method for predicting yield of an intelligent silicon-based OLED according to an embodiment of the present disclosure is shown;

[0055] Figure 2 A schematic diagram showing a data transposition process according to an embodiment of the present disclosure;

[0056] Figure 3 A schematic diagram showing a Shapley global ranking diagram according to an embodiment of the present disclosure;

[0057] Figure 4 A schematic diagram showing a Shapley heat map according to an embodiment of the present disclosure;

[0058] Figure 5 and Figure 6 A partial schematic diagram of a Shapley decision diagram with an output type of Class 12 according to an embodiment of the present disclosure is shown;

[0059] Figure 7 A Shapley dependency graph is shown when a feature type to be analyzed is VTH_LIN_value_rangeb and another feature type is MOB_SAT_value_std according to an embodiment of the present disclosure;

[0060] Figure 8 A schematic diagram showing an output processing process according to an embodiment of the present disclosure;

[0061] Fig. 9 A block diagram showing an intelligent silicon-based OLED yield prediction device according to an embodiment of the present disclosure is shown;

[0062] Fig.10 A block diagram of an intelligent silicon-based OLED yield prediction device according to another embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0063] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0064] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0065] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the main purpose of the present disclosure.

[0066] Figure 1The flowchart of the intelligent silicon-based OLED yield prediction method according to an embodiment of the present disclosure is shown. In this embodiment, the method is an intelligent silicon-based OLED yield prediction method based on the Shapley model, and the method is used in an electronic device with computing capabilities as an example for explanation. The electronic device can be a user terminal or a server. The user terminal can be a computer, a tablet computer, a mobile phone, etc. This embodiment does not limit the implementation of the electronic device. Figure 1 As shown, the method includes:

[0067] Step 101, obtaining production data of each smart silicon-based OLED.

[0068] The production data includes multiple data types and type values ​​corresponding to each data type.

[0069] The production data is used to indicate the production process and production results of smart silicon-based OLED (or samples).

[0070] Illustratively, data indicative of the production process include, but are not limited to: key process parameters, electrical parameter data, and production parameter data generated during the production process.

[0071] Among them, key process parameters refer to process parameters that may affect the production results (such as yield) of smart silicon-based OLEDs if their values ​​change during the production process. These process parameters need to be monitored and controlled to ensure that the produced smart silicon-based OLEDs meet the predetermined quality standards. Key process parameters include, but are not limited to: production temperature, evaporation rate, pressure, and / or production time, etc. This embodiment does not limit the parameter content of key process parameters.

[0072] Electrical parameter data refers to data obtained through electrical testing during the semiconductor manufacturing process. Electrical parameter data include, but are not limited to: when the transistor is working in the linear region, the gate voltage corresponding to when the transistor just starts to conduct current, when the transistor is working in the saturation region, the gate voltage corresponding to when the transistor just starts to conduct current, when the transistor is working in the saturation region, the mobility corresponding to when the transistor is working in the saturation region, current-voltage (IV) characteristics, etc. This embodiment does not limit the parameter content of the electrical parameter data.

[0073] The production parameter data is used to characterize the parameter data generated in real time during the production process. Specifically, it can be the data obtained after characterization processing of the parameter data generated in real time during the production process. Since the parameter data generated in real time will change over time, resulting in a large amount of parameters in the parameter data, the characteristics of the parameter data can be retained after characterization processing, and large storage resources and processing resources will not be occupied. For example: the parameter data is the residence time, production temperature and other parameter data of the smart silicon-based OLED passing through a certain production equipment.

[0074] Optionally, the method of characterizing the parameter data to obtain production parameter data includes, but is not limited to: obtaining the maximum value of the parameter data, or obtaining the minimum value of the parameter data, or calculating the mean and / or variance of the parameter data, etc. Different types of parameter data may have different characterization processing methods, and this embodiment does not limit the implementation method of the characterization processing.

[0075] The data indicating the production results include, but are not limited to: defect types and defect rates corresponding to each defect type.

[0076] During the production process of smart silicon-based OLEDs, data from the production process will be collected through various collection components to obtain key process parameters, electrical parameter data, and production parameter data; after the production is completed, the smart silicon-based OLED will be tested for defects to obtain the production results of the smart silicon-based OLED. If the production results indicate that the smart silicon-based OLED is defective (or unqualified), the defective type of the smart silicon-based OLED will also be detected. The defective types of each smart silicon-based OLED are counted to obtain the defective rate corresponding to each defective type. Thus, a set of production data corresponding to each smart silicon-based OLED can be obtained, and the production data includes multiple data types.

[0077] Schematically, the production data is tree-structured data, for example, a factory workshop is used as the root node of the tree structure, multiple production equipment in the factory workshop are first-level child nodes of the root node, and the data generated by each production equipment is a second-level child node of the first-level child node corresponding to the production equipment. At this time, production data of different data types can be queried with each other, for example, the factory workshop and production equipment can be obtained through the data in the second-level child node. In other embodiments, the production data can be other data structures, and this embodiment does not limit the data structure of the production data.

[0078] Optionally, in this embodiment, production data whose production results indicate that the smart silicon-based OLED is bad is taken as an example for explanation. In other embodiments, production data whose production results indicate that the smart silicon-based OLED is qualified and bad may also be analyzed. This embodiment does not limit the type of production data.

[0079] Step 102, delete the first data type and the type value corresponding to the first data type in the production data to obtain target data; the target data includes multiple second data types and the type values ​​corresponding to the second data types.

[0080] Among them, when the type value corresponding to the first data type changes, the prediction result of the smart silicon-based OLED remains basically unchanged, and the second data type is different from the first data type.

[0081] "Basically remain unchanged" means unchanged or the change value is less than the preset fluctuation range. Since the first data type has little effect on the prediction result, there is no need to analyze the impact of the first data type on the prediction result. The first data type and the corresponding type value can be deleted, thereby saving computing resources. For example: the factory workshop has little effect on the prediction result, so the factory workshop is the first data type.

[0082] The deletion method includes but is not limited to: obtaining a whitelist data type; comparing a data type in the production data with the whitelist data type; deleting a first data type that matches the whitelist data type and a type value corresponding to the first data type.

[0083] Alternatively, production data of each data type is displayed; when a deletion instruction to delete a first data type is received, the first data type indicated by the deletion instruction and the type value corresponding to the first data type are deleted.

[0084] The deletion instruction may be generated through a human-computer interaction interface, or may be sent by other devices. This embodiment does not limit the method for obtaining the deletion instruction.

[0085] Step 103, transpose the first data format of the target data into a second data format to obtain transposed data; the transposed data is N×M structured data, and the data in the row direction of the transposed data includes type values ​​of multiple second data types of the same smart silicon-based OLED; N and M are both integers greater than 1.

[0086] In one example, the target data is structured data represented by multiple first data tables, and the row-wise data in each first data table includes a type identifier of a second data type of the same smart silicon-based OLED and a type value corresponding to the second data type. The second data type and type value of the same smart silicon-based OLED stored in different first data tables are associated through the product identifier of the smart silicon-based OLED.

[0087] Optionally, the product identification may be a primary key of production data of the smart silicon-based OLED, or a product number, etc. This embodiment does not limit the implementation method of the product identification.

[0088] The type identifier may be a type name of the second data type, or a type code, etc. This embodiment does not limit the implementation method of the type identifier.

[0089] Accordingly, the first data format of the target data is transposed to the second data format to obtain transposed data, including: creating a second data table corresponding to the second data format, the first column of the second data table representing the product identification of each smart silicon-based OLED, the first row representing the type identification of each second data type, and the type identifications are arranged in a preset order; for the type identification and type value of the same row in each first data table corresponding to each product identification, converting the type value to the position indicated by the product identification and type identification in the second data table to obtain transposed data.

[0090] For example: Reference Figure 2 In the data transposition process shown, the target data includes two first data tables 201 and 202. The first rows of the first data tables 201 and 202 each include attributes corresponding to a column of data. The second row of the first data table 201 includes a type identifier class0 of a second data type corresponding to the smart silicon-based OLED 213a, and a type value 0.0017 corresponding to class0; the third row of the first data table 201 includes a type identifier class1 of another second data type corresponding to the smart silicon-based OLED 213a, and a type value 0.0317 corresponding to class1. Accordingly, the attributes of the first row of the first data table 201 include attributes of class0 and class1, which are bad types, and attributes of 0.0017 and 0.0317, which are bad rates.

[0091] The second row of the first data table 202 includes a type identifier VFB1_max of a second data type corresponding to the smart silicon-based OLED 213a, and a type value 0.13 corresponding to VFB1_max; the third row of the first data table 202 includes a type identifier VFB1_avg of another second data type corresponding to the smart silicon-based OLED 213a, and a type value 0.18 corresponding to VFB1_avg. Accordingly, the attributes of the first row of the first data table 202 include the attributes of VFB1_max and VFB1_avg, which are parameters, and the attributes of 0.13 and 0.18, which are parameter values.

[0092] In the second data table 203, the first column indicates the product identification 213a of each smart silicon-based OLED, and the first row indicates the type identification VFB1_max, VFB1_avg, class0 and class1 of each second data type; for the type identification class0 and the type value 0.0017 in the same row in each of the first data tables 201 and 202 corresponding to each product identification 213a, the type value 0.0017 is converted to the position indicated by the product identification 213a and the type identification class0 in the second data table 203, and the type identification class1 and the type value 0.0317 in the same row are converted to the position indicated by the product identification 213a and the type identification class1 in the second data table 203. The other data in the first data table are transposed in the same way to obtain transposed data.

[0093] According to the second data table 203, a group of data for each smart silicon-based OLED is located in the same row, and the data in the same row constitutes a multi-input and multi-output data structure corresponding to the bad prediction model, that is, including the type value of the feature type and the type value of the output type. At this time, the model can determine the second data type corresponding to the data according to the position of each data, without having to look up the type value corresponding to each second data type in the first data table during the training process, thereby improving the model training efficiency. Among them, the feature type refers to the second data type to be input into the bad prediction model, and the output type refers to the output type corresponding to the prediction result of the bad prediction model.

[0094] Step 104: Train a bad prediction model based on the first part of the transposed data.

[0095] Optionally, before training the bad prediction model based on the first part of the transposed data, the transposed data may be screened, specifically including: performing at least one level of screening on the transposed data to obtain screened transposed data; and determining the first part of the data and the second part of the data described below based on the screened transposed data.

[0096] In one example, the transposed data is subjected to at least one level of screening to obtain the screened transposed data, including: determining the variance of the type value corresponding to each second data type; performing a first level of screening on the transposed data based on a comparison result between the variance and a preset variance threshold; for any two second data types after the first level of screening, determining the correlation between the two second data types based on the type values ​​corresponding to the two second data types; performing a second level of screening on the transposed data based on a comparison result between the correlation and a preset correlation threshold to obtain the screened transposed data.

[0097] Based on the comparison result between the variance and the preset variance threshold, the transposed data is screened at the first level, including: if the variance corresponding to the second data type is less than the variance threshold, the second data type and the type value corresponding to the second data type are deleted. The variance threshold can be 10 -10 , or it may be other values ​​set according to the screening requirements. This embodiment does not limit the value of the variance threshold.

[0098] Optionally, the correlation between the two second data types may be determined by calculating the Pearson correlation coefficient between the two second data types, and accordingly, based on the comparison result between the correlation and a preset correlation threshold, the transposed data is subjected to secondary screening, including: if the correlation threshold corresponding to the second data type is greater than the correlation threshold, the second data type and the type value corresponding to the second data type are deleted. The correlation threshold may be 0.99, or may be other values ​​set according to screening requirements, and this embodiment does not limit the value of the correlation threshold.

[0099] In other embodiments, the correlation between the two second data types may be determined in other ways, such as determining the correlation based on Kendall rank correlation. This embodiment does not limit the way to determine the correlation.

[0100] In other embodiments, the transposed data may be subjected to only one level of screening, or may be subjected to more levels of screening. This embodiment does not limit the screening method.

[0101] In this embodiment, by screening the transposed data, it is possible to avoid the problem that the transposed data may have redundant information or contain too little information, achieve feature engineering dimensionality reduction of the transposed data, and improve the model calculation efficiency.

[0102] Optionally, before screening the data or training the bad prediction model, the transposed data may be preprocessed, and the preprocessing process includes but is not limited to: missing value processing, outlier processing, and / or standardization processing, etc.

[0103] Wherein, a missing value refers to a second data type of a certain smart silicon-based OLED, the second data type does not have a type value, but the same second data type of other smart silicon-based OLEDs has a type value. In this case, the type value that the smart silicon-based OLED does not have is a missing value. The method of processing missing values ​​includes but is not limited to: deleting a group of data with missing values; or filling the missing value with the average value of the data of the second data type corresponding to the missing value, etc. This embodiment does not limit the method of processing missing values.

[0104] An abnormal value refers to a type value of a second data type that exceeds a corresponding preset value range, or a change in the type value that exceeds a corresponding preset change range. The method of handling abnormal values ​​includes but is not limited to: deleting a group of data with abnormal values; or correcting the abnormal value to a normal value by a preset correction algorithm. This embodiment does not limit the method of handling abnormal values.

[0105] Standardization processing refers to scaling the type value to a standard range, generally scaling the type value to a range with a mean of 0 and a standard deviation of 1. Standardization processing methods include but are not limited to: normalization processing, logarithmic conversion processing, etc., and this embodiment does not limit the standardization processing method.

[0106] After screening and / or preprocessing the transposed data, the processed transposed data is divided into a preset ratio to obtain a training set, a validation set, and a test set, and the training set and the validation set are used to train the pre-created mathematical model to obtain a bad prediction model. Among them, the training set is the main source for the model to learn data features and establish prediction rules so that the model can learn patterns and relationships in the data. The validation set is used to adjust model parameters and determine the model structure during the model training process to evaluate the performance of the model on unseen data and prevent the model from overfitting. The test set is used to evaluate the performance of the bad prediction model.

[0107] In this embodiment, the training set and the validation set constitute the first part of data, and the test set constitutes the second part of data below. In other embodiments, the first part of data may also include only the training set, and correspondingly, the second part of data may include the validation set and the test set. This embodiment does not limit the determination method of the first part of data and the second part of data.

[0108] Optionally, the mathematical model used to train the bad prediction model includes but is not limited to: random forest model, elastic network, partial least squares regression, eXtreme Gradient Boosting (XGBoost), support vector regression (SVR), and this embodiment does not limit the implementation of the mathematical model. Since the random forest model has a good prediction effect in the yield prediction of smart silicon-based OLEDs and has strong interpretability, this embodiment takes the bad prediction model obtained by training the random forest model as an example for explanation.

[0109] In this embodiment, the multiple second data types include feature types for inputting into the bad prediction model, and output types corresponding to the prediction results of the bad prediction model. Optionally, the output types in the second data type are at least two, for example: the output type includes multiple bad types, and accordingly, the prediction result includes the bad rate corresponding to each bad type. At this time, it is necessary to ensure that the output of the bad prediction model is a multi-dimensional output. In one example, the bad prediction model is a multi-output model, and at this time, the output dimension of the multi-output model corresponds to the output type one by one.

[0110] In another example, if a single bad prediction model cannot achieve multi-dimensional output, the bad prediction model can be set to include a single output regression model corresponding to each output type, and the single output regression models corresponding to different output types are different. Among them, the single output regression model refers to a regression model that outputs only one output type, and the regression model is used to indicate the relationship between the type value of the feature type and the type value of the output type, so as to predict the type value of the output type based on the type value of the feature type.

[0111] Since the correlation between different defect types is small during the production of smart silicon-based OLEDs, different output types corresponding to different single-output regression models will not affect the accuracy of the defect prediction model. Optionally, the model types of the single-output regression models corresponding to different output types are the same or different, for example: the single-output regression model corresponding to output type 1 is a random forest model, and the single-output regression model corresponding to output type 2 is XGBoost. In this case, the model types of the single-output regression models corresponding to different output types are different.

[0112] A bad prediction model is obtained based on training of the first part of data, including: inputting the type value of the feature type in the first part of data into a pre-created mathematical model to obtain a model result; comparing the model result with the type value and model result of the same group of output types, iteratively updating the model parameters of the mathematical model based on the comparison result, to obtain a bad prediction model.

[0113] During the training process, by reading the same row of data in the first part of the data, the type values ​​corresponding to multiple output types can be directly obtained, which can be compared with the corresponding model results in turn without returning to the first data table to query the type value corresponding to each output type. Therefore, the efficiency of model training can be improved.

[0114] Step 105 : determining the Shapley value of each feature type in the plurality of second data types based on the second portion of data in the transposed data and the bad prediction model.

[0115] The Shapley value of each feature type is used to indicate the contribution of the feature type to the prediction result.

[0116] The second part of data includes multiple groups, each group of the second part of data includes the type value of each second data type corresponding to a certain smart silicon-based OLED. For example: Figure 2 A set of second part data corresponding to the smart silicon-based OLED 213a includes type values ​​of 0.13 and 0.18 of two characteristic types, and type values ​​of 0.0017 and 0.0317 of two output types.

[0117] Assume that the type values ​​corresponding to the p feature types of a set of the second part of the data corresponding to a certain smart silicon-based OLED in the second part of the data constitute a data set P, that is, the data dimension of the data set P is p, and the defect prediction model v can obtain the prediction results corresponding to each output type (such as the defect rate corresponding to each defect type) based on any subset S of P, and v(φ) = 0, φ is an empty set. For the subset S of P, v(S) represents the average prediction result obtained after the combined action of the features in the S set, that is, the prediction value of these features. Shapley value of the i-th feature type Calculated by the following formula:

[0118]

[0119] Among them, |S| represents the number of elements in subset S; v(S∪{i})-v(S) represents the contribution of the type value of the i-th feature type to the prediction result, P\{i} represents the set formed by removing the type value corresponding to feature type i from the set P, S belongs to P\{i}, and SU{i} represents the union of the type value corresponding to the set S and feature type i. The larger the Shapley absolute value of the i-th feature type, the higher the importance (or contribution) of feature type i to the prediction. A positive Shapley value indicates that feature type i will have a positive impact on the prediction result, making the prediction result higher; a negative Shapley value indicates that feature type i will have a negative impact on the prediction result, making the prediction result lower. By comparing the absolute values ​​and positive and negative values ​​of Shapley between different feature types i, the importance of each feature type in the bad prediction model and its impact on the prediction result can be explained.

[0120] Step 106 : Generate multi-dimensional explanation information based on the Shapley value of each feature type. The multi-dimensional explanation information is used to describe the influence of each feature type on yield prediction from different dimensions.

[0121] Optionally, the multi-dimensional explanation information may be text information and / or chart information. This embodiment does not limit the implementation method of the multi-dimensional explanation information.

[0122] In order to improve the intuitiveness of obtaining multi-dimensional explanation information, this embodiment is described by taking the generation of multi-dimensional explanation information in the form of a chart as an example.

[0123] In one example, the multi-dimensional explanation information includes a Shapley global ranking diagram, which is used to indicate the type of features that cause the adverse results and the type of output that the adverse prediction model can predict.

[0124] Accordingly, multi-dimensional explanatory information is generated based on the Shapley value of each feature type, including: determining the average absolute value of the Shapley value of each feature type; for each output type, determining the proportion of the output type in the average absolute value of each feature type; displaying each feature type through the first coordinate axis of the Shapley overall ranking diagram, displaying the average absolute value corresponding to each feature type through the second coordinate axis, and displaying the proportion corresponding to various output types through the third coordinate axis; wherein the first coordinate axis and the third coordinate axis are perpendicular to the second coordinate axis.

[0125] Since each feature type may have different type values ​​in different groups of the second part of the data, and the Shapley values ​​corresponding to different type values ​​may be different, the absolute values ​​of the Shapley values ​​corresponding to the same feature type are taken and averaged to obtain the average absolute value.

[0126] For each type value of each feature type, there is a corresponding prediction result, which includes type values ​​of multiple output types (i.e., predicted values). Accordingly, for each output type, there may be multiple type values ​​of feature types that are predicted by the bad prediction model to obtain the type value of the output type. Therefore, by statistically analyzing the Shapley values ​​corresponding to the type values ​​of each feature type that can obtain the predicted value of the output type, the absolute value of the Shapley value is taken and averaged; the quotient of the average value and the average absolute value is determined to obtain the proportion of the output type in the average absolute value of each feature type.

[0127] refer to Figure 3 The Shapley overall ranking diagram shown is taken as an example, with the first coordinate axis as the left vertical axis, the second coordinate axis as the horizontal axis, and the color legend located on the right side of the coordinate axis. The first coordinate axis shows the various feature types, which are specifically represented by the type identification of each feature type; the average absolute value of each feature type is represented by a bar graph, and the height of the bar graph is indicated by the numerical value corresponding to the second coordinate axis; the feature types displayed in the first coordinate axis are arranged in descending order according to the bar graph, and the higher the bar graph, the greater the influence of the feature type on the prediction result. The proportion of each output type in the average absolute value of each feature type is represented by a rectangular legend, and the proportion of each output type is arranged in order from large to small, and the size of the proportion is represented by rectangular legends of different colors, among which blue represents a large proportion and purple represents a small proportion. For example: Figure 3In the figure, the Shapley values ​​of each feature type mainly affect the output variables in the blue part (such as the bad type Class12), and the Shapley values ​​of each feature type corresponding to the output variables in the purple part are very small. The performance of the bad prediction model is mainly determined by the output variables with a relatively high proportion, such as Figure 3 In the figure, the blue output types (Class 12, Class 0) account for a large proportion. The performance of the bad prediction model is mainly determined by the prediction results corresponding to the output types Class 12 and Class 0. The Shapley mean absolute value of the feature type VFB1_value_max is the largest, indicating that it has the greatest impact on the prediction results of the bad prediction model.

[0128] In other embodiments, the first coordinate axis may be the horizontal axis, and correspondingly, the second coordinate axis may be the vertical axis; the average absolute value of each feature type may not be represented by a bar chart, but by other methods such as a line chart; the proportion corresponding to each output type may also be represented by a color legend of different colors. This embodiment does not limit the display method of the Shapley overall ranking diagram.

[0129] In another example, the multi-dimensional explanatory information includes a Shapley heat map, which is used to indicate the Shapley value ranking of each feature type predicting the output type for each output type to be analyzed predicted by the bad prediction model, and indicates whether each feature type has a positive or negative impact on the output type.

[0130] Optionally, the output type to be analyzed may be any output type, or may be the output type with the top n proportions determined by the Shapley overall ranking diagram, where n is a positive integer. This embodiment does not limit the method for determining the output type to be analyzed.

[0131] Accordingly, multi-dimensional explanation information is generated based on the Shapley value of each feature type, including: for each output type to be analyzed, determining the first Shapley absolute value when predicting the output type based on each feature type; clustering each group of second part data according to explanation similarity to obtain multiple sample sets; determining the second Shapley absolute value when predicting the output type based on each sample set; determining the prediction result corresponding to each sample set based on the bad prediction model; displaying each feature type through the first coordinate axis of the Shapley heat map, displaying the first Shapley absolute value corresponding to each feature type through the second coordinate axis, displaying each sample set corresponding to each feature type through the third coordinate axis, and displaying the color representation of the second Shapley absolute value between the first coordinate axis and the third coordinate axis, the color representation is used to indicate the value of the second Shapley absolute value; wherein the first coordinate axis and the second coordinate axis are perpendicular to the third coordinate axis; and displaying the change curve of the prediction result corresponding to each sample set at a specified position of the Shapley heat map.

[0132] Here, explanation similarity refers to multiple sets of data with the same output type and the same feature type.

[0133] The method for determining the first Shapley absolute value includes: statistically averaging the absolute values ​​of Shapley values ​​corresponding to each feature data in each group of second part data that can predict the output type to be analyzed, to obtain the first Shapley absolute value corresponding to each feature type.

[0134] The second Shapley absolute value is determined by averaging the absolute values ​​of Shapley values ​​corresponding to each feature type in a sample set that can predict the output type to be analyzed, to obtain the second Shapley absolute value corresponding to the sample set.

[0135] refer to Figure 4The Shapley heat map shown in the figure takes the first coordinate axis of the Shapley heat map as the left vertical axis, the second coordinate axis as the right vertical axis, and the third coordinate axis as the horizontal axis as an example. The first coordinate axis shows the various feature types, which are specifically represented by the type identifiers of the various feature types; the first Shapley absolute value corresponding to each feature type is represented by a bar graph on the second coordinate axis. The higher the bar graph, the larger the first Shapley absolute value. The third coordinate axis shows the various sample sets corresponding to each feature type, which are specifically represented by the set number of the sample set, for example: represented by a set number of 0-400. The color representation of the second Shapley absolute value is displayed between the first and third coordinate axes. Red indicates that the type value of the feature type in the sample set can improve the predicted value in the prediction result, and blue indicates that the type value of the feature type in the sample set can reduce the predicted value in the prediction result. The depth of the color represents the size of the second Shapley absolute value. The darker the second Shapley absolute value, the larger the second Shapley absolute value, and the lighter the second Shapley absolute value, the smaller the second Shapley absolute value.

[0136] The change curve of the prediction results is displayed above the first coordinate axis, the second coordinate axis, and the third coordinate axis, indicating the change curve of the prediction results corresponding to each sample set. The higher the predicted value of the output type in the prediction result, the higher the corresponding function height. The gray horizontal line indicates the mean of the prediction results of each sample set.

[0137] In other embodiments, the first coordinate axis and the second coordinate axis may also be the horizontal axis, and correspondingly, the third coordinate axis may be the vertical axis; the first Shapley absolute value may also not be represented by a bar graph, but by other methods such as a line graph; this embodiment does not limit the display method of the Shapley heat map.

[0138] In another example, the multi-dimensional explanation information includes a Shapley decision diagram, which is used to indicate the change process of the prediction result after gradually adding each feature type for each output type to be analyzed that can be predicted by the bad prediction model.

[0139] Accordingly, multi-dimensional explanatory information is generated based on the Shapley value of each feature type, including: for each group of second part data, the type values ​​corresponding to each feature type in the second part data are added to the input data in sequence according to a preset input order, and the prediction results corresponding to various input data are predicted respectively by the bad prediction model to obtain the change process of the prediction results corresponding to the second part data; the change process of different prediction results that meet the preset conditions is clustered to obtain the change process of the clustered prediction results; the mean of the prediction results corresponding to each group of second part data is obtained; each feature type is displayed in the input order through the first coordinate axis of the Shapley decision diagram, and each prediction result is displayed through the second coordinate axis, and based on the change process of the clustered prediction results, the prediction value change curve starts from the mean of the prediction results on the second coordinate axis and extends along the direction of the first coordinate axis; wherein the first coordinate axis is perpendicular to the second coordinate axis.

[0140] The preset conditions include, but are not limited to: the similarity of the change process between different prediction results is greater than or equal to a preset similarity threshold.

[0141] refer to Figure 5 and Figure 6 The following is a partial schematic diagram of a Shapley decision diagram for output type Class 12, where: Figure 5 is the top of the Shapley decision diagram, Figure 6 The prediction value change curve in the middle part of the Shapley decision diagram has a small change amount, that is, the prediction value change curve is basically vertical, so this embodiment takes the top and bottom of the Shapley decision diagram as an example for schematic description. Figure 5 In the example, the first axis of the Shapley decision diagram is the left vertical axis and the second axis is the horizontal axis. The first axis shows the various feature types, which are specifically represented by the type identifiers of the feature types; the second axis shows the various prediction results, specifically showing the prediction values ​​corresponding to the current output type Class12; starting from the mean of the prediction results on the second axis, the prediction value change curve is extended along the direction of the first axis. The color of the prediction value change curve corresponding to each group of the second part of the data indicates the size of the final prediction value when the input data includes all feature types. Blue means that its final prediction value is relatively low, and red means that its final prediction value is relatively high.

[0142] In other embodiments, the first coordinate axis may also be the horizontal axis, and correspondingly, the second coordinate axis may be the vertical axis; the preset value change curve may also not have color, or other colors may be used to represent the size of the final predicted value. This embodiment does not limit the display method of the Shapley decision diagram.

[0143] In yet another example, the multi-dimensional explanation information includes a Shapley dependency graph, which is used to indicate how the prediction result and Shapley value corresponding to the type value of each feature type are affected by the type value of another feature type.

[0144] Accordingly, multi-dimensional explanatory information is generated based on the Shapley value of each feature type, including: for the feature type to be analyzed, each Shapley value is displayed through the first coordinate axis of the Shapley dependency graph, and different type values ​​corresponding to the feature type are displayed through the second coordinate axis, and color points are displayed between the first coordinate axis and the second coordinate axis. The position of the color point is determined by the Shapley value corresponding to the type value of the feature type to be analyzed when predicting the output type to be analyzed, and the color of the color point indicates the size of the type value of another feature type.

[0145] Among them, the feature types to be analyzed can be all feature types, or can also be feature types whose average absolute values ​​are in the top m positions determined by the Shapley overall ranking diagram, where m is a positive integer. This embodiment does not limit the method for determining the feature types to be analyzed.

[0146] refer to Figure 7 The Shapley dependence diagram when the feature type to be analyzed is VTH_LIN_value_rangeb and the other feature type is MOB_SAT_value_std. Among them, the feature type VTH_LIN_value_rangeb represents the gate voltage corresponding to the transistor working in the linear region when the transistor just starts to conduct current. The other feature type MOB_SAT_value_std represents the standard deviation of the mobility corresponding to the transistor working in the saturation region. Figure 7In the example, the first coordinate axis of the Shapley dependency graph is the left vertical axis, the second coordinate axis is the horizontal axis, and the third coordinate axis is the right vertical axis. The first coordinate axis displays various Shapley values; the second coordinate axis displays different type values ​​of VTH_LIN_value_rangeb, and the third coordinate axis displays the type value of another feature type MOB_SAT_value_std. Each color point displayed between the first coordinate axis and the second coordinate axis represents the Shapley value corresponding to the type value corresponding to the feature type VTH_LIN_value_rangeb in a set of second part data. The color in the color legend in the third coordinate axis represents the size of the type value of another feature type MOB_SAT_value_std. The redder the color, the larger the type value of another feature type MOB_SAT_value_std. If the type value of the second coordinate axis corresponds to a single Shapley value, it means that the type value does not have an interaction with another feature type MOB_SAT_value_std. However, if the type value of the second coordinate axis corresponds to multiple Shapley values, it means that the Shapley value is affected by another characteristic type MOB_SAT_value_std in addition to the characteristic type VTH_LIN_value_rangeb.

[0147] according to Figure 7 It can be seen that when the type value of VTH_LIN_value_rangeb is in the range of 0.2 to 1, the Shapley value decreases as it increases. When the interval range is relatively large, the change in the Shapley value is not greatly affected, and the impact on the poor prediction results is negative. When the type value of VTH_LIN_value_rangeb is near zero, its Shapley value will be affected by another feature type, MOB_SAT_value_std. When the interval range of the type value of VTH_LIN_value_rangeb is 5 and the type value of another feature type, MOB_SAT_value_std, is large, the Shapley value is relatively low.

[0148] In summary, the smart silicon-based OLED yield prediction method provided in the present embodiment obtains production data of each smart silicon-based OLED; deletes the first data type and the type value corresponding to the first data type in the production data to obtain target data, wherein the target data includes multiple second data types and type values ​​corresponding to the second data types; transposes the first data format of the target data to the second data format to obtain transposed data; the transposed data is N×M structured data, and the data in the row direction of the transposed data includes type values ​​of multiple second data types of the same smart silicon-based OLED; a defect prediction model is obtained by training based on the first part of the data in the transposed data; based on the second part of the data in the transposed data and the defect prediction model, the Shapley value of each feature type in the multiple second data types is determined; and multi-dimensional explanation information is generated based on the Shapley value of each feature type; after the first data format of the target data is transposed to the second data format, the type value of the same row can be directly used for the training of the defect prediction model, without calling the target data in the first data format to find the second data type corresponding to each type value during the training process, therefore, the problem of low efficiency of the traditional data-driven method in predicting yield can be solved, and the training efficiency of the defect prediction model can be improved, thereby improving the yield prediction efficiency.

[0149] In addition, by generating multi-dimensional explanatory information based on the Shapley value of each feature type, the root cause of the defective type can be intuitively reflected, thereby improving the interpretability of yield prediction.

[0150] In order to more clearly understand the processing of production data in this application, the following is an example to introduce the processing process. Figure 8 In the process shown, production data is obtained by collecting data from various data sources 801. Figure 8 The production data includes production data with poor test results, and the production data with poor test results includes multiple groups, each group of data includes: key process parameters, electrical parameter data and production parameter data, and poor test data, and the poor test data includes but is not limited to poor type and poor rate.

[0151] After the production data is collected, the production data is integrated 802. Specifically, data irrelevant to the bad prediction in the production data is deleted, that is, the first data type and the type value corresponding to the first data type are deleted, and each second data type obtained after the deletion and the type value corresponding to the second data type are associated based on the product identifier to obtain the target data.

[0152] After the data is integrated to obtain the target data, the first data format of the target data is transposed into the second data format, and the obtained structured transposed data is preprocessed 803. Specifically, the transposed data is processed for missing values, outliers, and standardization to obtain preprocessed data. The implementation methods of the transposition method and each preprocessing are referred to in the above embodiment, and will not be described in detail in this embodiment.

[0153] After obtaining the preprocessed data, feature engineering dimension reduction is performed on the preprocessed data 804. Specifically, a primary screening (i.e., pre-screening) is performed based on the variance of the type value corresponding to each second data type; after the primary screening is completed, a secondary screening (i.e., fine screening) is performed based on the correlation between any two second data types to obtain the screened transposed data.

[0154] Finally, the yield prediction is performed based on the filtered transposed data 805. Specifically, a defect prediction model is obtained by training the first part of the filtered transposed data, and the feature data in the second part of the data is predicted using the defect prediction model, and the Shapley value of each feature type is determined according to the prediction result and the second part of the data, so as to generate multi-dimensional explanation information based on the Shapley value.

[0155] Fig. 9 FIG. 1 is a block diagram of an intelligent silicon-based OLED yield prediction device according to an embodiment of the present disclosure. Fig. 9 As shown, the device includes: a data acquisition module 910, a data deletion module 920, a data transposition module 930, a model training module 940, a Shapley value calculation module 950 and a model interpretation module 960.

[0156] The data acquisition module 910 is used to acquire the production data of each smart silicon-based OLED; wherein the production data includes multiple data types and type values ​​corresponding to each data type;

[0157] A data deletion module 920, configured to delete a first data type and a type value corresponding to the first data type in the production data, and obtain target data, wherein the target data includes a plurality of second data types and type values ​​corresponding to the second data types;

[0158] The data transposition module 930 is used to transpose the first data format of the target data into a second data format to obtain transposed data; the transposed data is N×M structured data, and the row-wise data in the transposed data includes type values ​​of multiple second data types of the same smart silicon-based OLED; the N and the M are both integers greater than 1;

[0159] A model training module 940, configured to obtain a bad prediction model based on the first part of the transposed data;

[0160] A Shapley value calculation module 950, configured to determine a Shapley value of each feature type in the plurality of second data types based on the second portion of data in the transposed data and the bad prediction model;

[0161] The model explanation module 960 is used to generate multi-dimensional explanation information based on the Shapley value of each feature type, and the multi-dimensional explanation information is used to describe the influence of each feature type on yield prediction from different dimensions.

[0162] In some embodiments, the functions or modules included in the device provided by the embodiments of the present disclosure can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0163] The embodiment of the present disclosure also provides a computer-readable storage medium on which computer program instructions are stored, and the computer program instructions implement the above method when executed by a processor. The computer-readable storage medium can be a volatile or non-volatile computer-readable storage medium.

[0164] An embodiment of the present disclosure further proposes an electronic device, comprising: a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0165] The embodiments of the present disclosure also provide a computer program product, including a computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of an electronic device, the processor in the electronic device executes the above method.

[0166] Fig.10 1 is a block diagram of an intelligent silicon-based OLED yield prediction device 1900 according to an exemplary embodiment. For example, the device 1900 may be provided as a terminal device. Fig.10 , the apparatus 1900 includes a processing component 1922, which further includes one or more processors, and a memory resource represented by a memory 1932 for storing instructions, such as an application, that can be executed by the processing component 1922. The application stored in the memory 1932 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 1922 is configured to execute instructions to perform the above method.

[0167] The device 1900 may also include a power supply component 1926 configured to perform power management of the device 1900, a wired or wireless network interface 1950 configured to connect the device 1900 to a network, and an input / output interface 1958 (I / O interface). The device 1900 may operate based on an operating system stored in the memory 1932, such as Windows Server 2000. TM , MacOS X TM , Unix TM ,Linux TM , FreeBSD TM or similar.

[0168] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions, which can be executed by the processing component 1922 of the device 1900 to perform the above method.

[0169] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. An intelligent silicon-based OLED yield prediction method, characterized in that: The method comprises: Acquire production data of each smart silicon-based OLED; wherein the production data includes multiple data types and type values ​​corresponding to each data type; Deleting a first data type and a type value corresponding to the first data type in the production data to obtain target data, wherein the target data includes a plurality of second data types and type values ​​corresponding to the second data types; Transposing the first data format of the target data into a second data format to obtain transposed data; the transposed data is N×M structured data, and the row-wise data in the transposed data includes type values ​​of multiple second data types of the same smart silicon-based OLED; both N and M are integers greater than 1; A bad prediction model is obtained by training based on the first part of the transposed data; Determine, based on the second portion of data in the transposed data and the bad prediction model, a Shapley value of each feature type in the plurality of second data types; Multi-dimensional explanation information is generated based on the Shapley value of each feature type, and the multi-dimensional explanation information is used to describe the influence of each feature type on yield prediction from different dimensions.

2. The method according to claim 1, characterized in that The target data is structured data represented by a plurality of first data tables, and the row-wise data in each first data table includes a type identifier of a second data type of the same smart silicon-based OLED and a type value corresponding to the second data type; accordingly, The step of transposing the first data format of the target data into a second data format to obtain transposed data includes: Creating a second data table corresponding to the second data format, wherein the first column of the second data table represents the product identification of each smart silicon-based OLED, the first row represents the type identification of each second data type, and the type identifications are arranged in a preset order; For each type identifier and type value in the same row of each first data table corresponding to each product identifier, the type value is converted to the position indicated by the product identifier and the type identifier in the second data table to obtain the transposed data.

3. The method according to claim 1, characterized in that: Before obtaining the bad prediction model through training based on the first part of the transposed data, the method further includes: Performing at least one level of screening on the transposed data to obtain screened transposed data; The first portion of data and the second portion of data are determined based on the filtered transposed data.

4. The method according to claim 3, characterized in that: The performing at least one level of screening on the transposed data to obtain screened transposed data includes: Determine the variance of the type value corresponding to each second data type; Based on the comparison result between the variance and a preset variance threshold, performing a primary screening on the transposed data; For any two second data types after the primary screening, determining the correlation between the two second data types based on the type values ​​corresponding to the two second data types; Based on the comparison result between the correlation and a preset correlation threshold, the transposed data is subjected to secondary screening to obtain the screened transposed data.

5. The method according to any one of claims 1 to 4, characterized in that: The multiple second data types include the feature type and the output type, the feature type is the second data type input to the bad prediction model, and the output type is the second data type corresponding to the prediction result of the bad prediction model; the multi-dimensional explanation information includes a Shapley overall sorting diagram, and the Shapley overall sorting diagram is used to indicate the feature type that causes the bad result and the output type that the bad prediction model can predict; accordingly, The multi-dimensional explanation information is generated based on the Shapley value of each feature type, including: Determine the mean absolute value of the Shapley value for each feature type; For each output type, determining the proportion of the output type in the average absolute value of each feature type; The first coordinate axis of the Shapley overall ranking diagram displays each feature type, the second coordinate axis displays the average absolute value corresponding to each feature type, and the color legend displays the proportion corresponding to various output types; wherein the first coordinate axis is perpendicular to the second coordinate axis.

6. The method according to claim 1, characterized in that The multiple second data types include the feature type and the output type, the feature type is the second data type input to the bad prediction model, and the output type is the second data type corresponding to the prediction result of the bad prediction model; the multi-dimensional explanation information includes a Shapley heat map, and the Shapley heat map is used to indicate the Shapley value ranking of each feature type predicting the output type for each output type to be analyzed predicted by the bad prediction model, and indicates whether each feature type has a positive or negative impact on the output type; accordingly, The multi-dimensional explanation information is generated based on the Shapley value of each feature type, including: For each output type to be analyzed, determining a first Shapley absolute value when predicting the output type based on each feature type; Clustering the second part of each group of data according to explanation similarity to obtain multiple sample sets; wherein explanation similarity refers to multiple groups of data with the same output type and the same feature type; Determining a second Shapley absolute value when predicting the output type based on each sample set; Determine the prediction result corresponding to each sample set based on the bad prediction model; The first coordinate axis of the Shapley heat map displays each feature type, the second coordinate axis displays the first Shapley absolute value corresponding to each feature type, and the third coordinate axis displays each sample set corresponding to each feature type, and the color representation of the second Shapley absolute value is displayed between the first coordinate axis and the third coordinate axis, and the color representation is used to indicate the value of the second Shapley absolute value; wherein the first coordinate axis and the second coordinate axis are perpendicular to the third coordinate axis; The change curve of the prediction result corresponding to each sample set is displayed at a specified position of the Shapley heat map.

7. The method according to claim 1, characterized in that The multiple second data types include the feature type and the output type, the feature type is the second data type input to the bad prediction model, and the output type is the second data type corresponding to the prediction result of the bad prediction model; the multi-dimensional explanation information includes a Shapley decision diagram, and the Shapley decision diagram is used to indicate the change process of the prediction result after gradually adding each feature type for each output type to be analyzed that can be predicted by the bad prediction model; Accordingly, The multi-dimensional explanation information is generated based on the Shapley value of each feature type, including: For each group of the second part of data, the type values ​​corresponding to each feature type in the second part of data are added to the input data one by one according to the preset input order, and the prediction results corresponding to various input data are respectively predicted by the bad prediction model to obtain the change process of the prediction results corresponding to the second part of data; Clustering the changing processes of different prediction results that meet the preset conditions to obtain the changing processes of the clustered prediction results; Get the mean of the prediction results corresponding to the second part of each group of data; The first coordinate axis of the Shapley decision diagram displays each feature type in the input order, and the second coordinate axis displays each prediction result, and based on the change process of the prediction result after clustering, the prediction value change curve starts from the mean of the prediction result on the second coordinate axis and is extended along the direction of the first coordinate axis; wherein the first coordinate axis is perpendicular to the second coordinate axis.

8. The method according to claim 1, characterized in that: The multiple second data types include the feature type and the output type, the feature type is the second data type input to the bad prediction model, and the output type is the second data type corresponding to the prediction result of the bad prediction model; the multi-dimensional explanation information includes a Shapley dependency graph, and the Shapley dependency graph is used to indicate, for each output type to be analyzed that can be predicted by the bad prediction model, how the Shapley value corresponding to the type value of the feature type to be analyzed is affected by the type value of another feature type; accordingly, The multi-dimensional explanation information is generated based on the Shapley value of each feature type, including: For the output type to be analyzed and the feature type to be analyzed, each Shapley value is displayed through the first coordinate axis of the Shapley dependency graph, and different type values ​​corresponding to the feature type are displayed through the second coordinate axis, and color points are displayed between the first coordinate axis and the second coordinate axis. The position of the color point is determined by the Shapley value corresponding to the type value of the feature type to be analyzed when predicting the output type to be analyzed, and the color of the color point indicates the size of the type value of the other feature type.

9. An intelligent silicon-based OLED yield prediction device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to implement the method described in any one of claims 1 to 8 when executing the instructions stored in the memory.

10. A non-volatile computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images