A method and device for yield early warning and diagnostic analysis

Abstract

The present invention belongs to the field of semiconductor manufacturing technology, and in particular, relates to a method and device for yield early warning and diagnostic analysis, including a yield statistics & control module, an alarm grouping module, an analysis module, and a result front-end display output module, wherein the yield statistics & control module includes a Daily Bin Ratio control early warning module, a control early warning Base Line configuration module, and a Daily yield statistics module. The above-mentioned method for intelligent yield early warning and diagnostic analysis in the semiconductor industry analyzes the data in the database, statistically analyzes the bad wafers, issues early warning notifications for abnormal wafers, and automatically analyzes them, pointing out the main root causes of the bad wafers, thereby finding the problems more quickly and improving the yield of the wafers. In the whole process, due to full automation, there is no need for personnel to participate in data input and manual calculation, which greatly improves the analysis rate and ensures the accuracy of statistical analysis.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor manufacturing, and in particular to a method for intelligent yield early warning and diagnostic analysis in the semiconductor industry. Background Art

[0002] Yield is an important evaluation indicator for many manufacturing plants, especially for the semiconductor wafer manufacturing industry. Each silicon wafer undergoes many precision processes. If an imperfect process occurs during the process, it will affect the subsequent manufacturing process and product quality, and in serious cases, it will cause the wafer to be scrapped and all efforts to be wasted.

[0003] The yield of semiconductor wafers is calculated based on the ratio of defective wafers to functional wafers. The yield of wafers is generally defined as: There is a special mode in the Wafer Bin Map (WBM) to determine whether there are abnormalities in the wafer manufacturing process. That is, the probe tests the electrical characteristics of each grain in the wafer and its connection in the circuit. As a wafer foundry, customers often require the yield of wafer products to reach 83% or above, otherwise they will be judged as waste wafers, resulting in returns or price differences. Therefore, the yield of wafer products is related to the profits of the manufacturing industry, which is often a huge benefit.

[0004] The existing analysis method in the semiconductor manufacturing industry first manually screens the defective data provided by each inspection point to find out the types of defective wafers and sort them according to the order of defect types. The defective data is manually processed, and the data is manually pulled from the database to analyze the root causes of the defects.

[0005] Although this method is based on computer system analysis, it involves too much human intervention, and when analyzing the root cause, it cannot take into account all parameters, especially the massive amount of data, which makes the analysis time-consuming and laborious. Moreover, during the operation, people will inevitably make mistakes, which will affect the final root cause.

[0006] Big data is a concept that has been widely used in enterprises for data analysis, business intelligence and statistical applications in the past decade. Due to the rapid growth of data volume, the decline in data storage costs, the advancement of software technology and the maturity of cloud environments, the field has matured. Data analysis of previous insights from historical data can predict future results and even innovate and create never-before-seen business models.

[0007] The concepts of smart factories and Industry 4.0 have recently attracted increasing attention in the global semiconductor industry. The semiconductor industry occupies an important position in the global market, especially in the fields of wafer foundry, packaging and testing. Due to fierce global competition, implementing smart manufacturing to improve the efficiency of the production process, increase the degree of automation of equipment, and enhance data analysis capabilities is not only a choice, but also a necessary action.

[0008] The characteristics of big data types in the semiconductor industry are called 3Vs: massive data (Volume), the speed at which data is generated (Velocity), and there are various types of data (Variety). As semiconductor manufacturing technology advances, processes become longer and more complex, and a large amount of data is generated during the manufacturing process. Each machine at each station can generate data in each step of the process, so the speed at which data is generated is high. The data generated from different machines may have different forms, such as images or numbers. A large amount of production records and process parameter data has accumulated and must be analyzed.

[0009] Therefore, it is necessary to determine how to avoid yield loss or diagnose the cause of yield loss as soon as possible in the early stage, and optimize the current yield analysis so that wafer yield analysis can be performed faster and more accurately to save more money. To this end, we propose a yield early warning and diagnostic analysis method and device to solve the above problems. Summary of the invention

[0010] The purpose of the present invention is to solve the problems existing in the prior art and to propose a method and device for yield early warning and diagnostic analysis.

[0011] In order to achieve the above object, the present invention adopts the following technical solutions:

[0012] A device for yield warning and diagnostic analysis, including a yield statistics & control module, an alarm grouping module, an analysis module and a result front-end display output module, wherein the yield statistics & control module includes a Daily Bin Ratio control and warning module, a control and warning Base Line configuration module and a Daily yield statistics module, the alarm grouping module includes an alarm grouping module based on Bin Ratio correlation and an alarm grouping module based on Bin Map features, the analysis module includes a Wafer Bin Ratio and numerical / non-numerical parameter correlation analysis module, a Bin Map feature similarity grouping module, a Bin Map and Site Map feature matching module, a Bin Map and Defect Map matching analysis module and a Wafer abnormal production process analysis module, the yield statistics & control module is connected to the alarm grouping module, the Bin Ratio correlation alarm grouping module is connected to the Bin Map feature alarm grouping module and the Wafer Bin Ratio and numerical / non-numerical parameter correlation analysis module respectively, the Bin Map feature alarm grouping module is connected to the Bin Map feature similarity grouping module, the Bin Map and Site Map feature matching module respectively, and the The Map feature matching module, the Bin Map and Defect Map matching analysis module and the Wafer abnormal production process analysis module are connected respectively, and the analysis module is connected to the result front-end display output module.

[0013] Preferably, according to the purpose of the invention, a method for performing yield analysis of semiconductor wafers by using an intelligent yield early warning and diagnostic analysis device for the semiconductor industry is proposed, comprising the steps of:

[0014] 1) Collect statistics on daily data and summarize early warnings based on the BaseLine configuration information of management and control to obtain daily Ratio early warning information.

[0015] 2) Every day, bins with spatial correlation are integrated into each wafer map as the input of the classifier. The classifier outputs the classification result and features of the bin group and saves the bin map image information, which serves as the basic metadata for subsequent image similarity judgment.

[0016] 3). Every defect map is input into the classifier every day. The classifier outputs the classification result and features of the defect map and saves the defect map image information, which serves as the basic metadata for subsequent image similarity judgment.

[0017] 4). Based on the daily Ratio warning information, the analysis is divided into two categories: Ratio correlation analysis based on Wafer and analysis based on Wafer Bin Map (WBM).

[0018] 5). Based on the daily Ratio warning wafer, each warning case is divided into groups according to its map type and similarity coefficient. The mode type of the wafer is taken as the new bin type.

[0019] 6). Based on Wafer's Ratio correlation analysis, perform numerical (Inline / WAT / WaitTime parameters) or non-numerical (Tool / Chamber / Dcoll) correlation analysis.

[0020] 7). In the Wafer Bin Map (WBM) analysis, based on the pre-classified Wafer Bin Map (WBM) data and abnormal Wafer Bin Map (WBM) classification data within a fixed historical time, the Wafer that has been warned is searched for similarity to obtain the Bad / Good data within a fixed time based on the Bin Map.

[0021] 8). Perform numerical (Inline / WAT / WaitTime parameters) or non-numerical (Tool / Chamber / Dcoll) correlation analysis based on the Bad / Good data within a fixed time of the Bin Map.

[0022] 9). Based on the Wafer Bin Map (WBM) anomaly (Die) warned by daily Ratio, analyze the first N Inline parameters and WAT parameters of the nearest neighbors of the Wafer Bin Map (WBM) centroid.

[0023] 10). Based on the abnormal (die) wafers warned every day, find the top N most similar abnormal Wafer Bin Map (WBM) corresponding to it, and perform similar searches within a fixed historical time according to the three dimensions of same wafer / same product / cross-products.

[0024] Compared with the prior art, the advantages of the method and device for yield early warning and diagnostic analysis provided by the present invention are:

[0025] The above-mentioned method of intelligent yield warning and diagnostic analysis in the semiconductor industry analyzes the data in the database, statistically analyzes the defective wafers, issues warning notifications for abnormal wafers, and automatically analyzes them to point out the main root causes of wafer defects, thereby finding problems more quickly and improving the yield of wafers. In the entire process, due to full automation, there is no need for human participation in data input and manual calculations, which greatly improves the analysis rate while ensuring the accuracy of statistical analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a framework diagram of the intelligent yield warning and diagnostic analysis method in the semiconductor industry.

[0027] Figure 2 This is a flowchart of daily warning analysis based on existing data.

[0028] Figure 3 This is a flowchart for constructing a feature library for bin map image similarity retrieval.

[0029] Figure 4 This is a flowchart of constructing a feature library for similarity retrieval of defect maps.

[0030] Figure 5 It is a flowchart based on the map type and similar grouping corresponding to the daily Ratio warning wafer.

[0031] Figure 6 It is a flowchart for correlation analysis of numerical values ​​(Inline / WAT / WaitTime parameters) based on daily Ratio warning information.

[0032] Figure 7 It is a flowchart for non-numerical (Tool) correlation analysis based on daily Ratio warning information.

[0033] Figure 8 It is a flowchart for non-numerical (Chamber) correlation analysis based on daily Ratio warning information.

[0034] Fig. 9 It is a flowchart for non-numerical (Dcoll) correlation analysis based on daily Ratio warning information.

[0035] Fig.10 It is a flowchart that filters out Bad / Good data within a fixed time based on the warning Bin Map.

[0036] Fig.11This is a flowchart for performing correlation analysis of values ​​(Inline / WAT / WitTime parameters) based on wafers corresponding to bad / good cases retrieved by similarity.

[0037] Fig.12 This is a flowchart that analyzes the relationship between the Inline parameters and bin ratio of the failed bin in the warning from the Map dimension.

[0038] Fig.13 This is a flowchart that analyzes the relationship between the WAT parameter and bin ratio of the failed bin in the warning from the Map dimension.

[0039] Fig.14 It is a flowchart that searches for the top N most similar abnormal Wafer Bin Maps (WBM) corresponding to the abnormal (die) Wafers that are warned, and performs similar searches within a fixed historical time in three dimensions: same Wafer / same product / cross-products. DETAILED DESCRIPTION

[0040] As mentioned in the background technology, the existing yield warning and diagnostic analysis methods require manual participation in calculation and analysis, which inevitably requires a lot of time, has low analysis accuracy, and other problems.

[0041] The semiconductor industry intelligent yield early warning and diagnostic analysis method of the present invention automatically pulls data through a computer, performs statistical analysis on the data, and automatically diagnoses and analyzes the root causes of defects, thereby quickly obtaining the final early warning and analysis results, and the entire process does not require human intervention, so it has a higher analysis result.

[0042] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0043] Embodiment 1

[0044] A yield early warning and diagnostic analysis device, referring to Figure 1, including yield statistics & control module 1, alarm grouping module 2, analysis module 3 and result front-end display output module 4. Yield statistics & control module 1 includes Daily Bin Ratio control and early warning module 11, control and early warning Base Line configuration module 12 and Daily yield statistics module 13. Alarm grouping module 2 includes Bin Ratio correlation-based alarm grouping module 21 and Bin Map feature-based alarm grouping module 22. Analysis module 3 includes Wafer Bin Ratio and numerical / non-numerical parameter correlation analysis module 31, Bin Map feature similarity grouping module 32, Bin Map and Site Map feature matching module 33, Bin Map and Defect Map matching analysis module 34 and Wafer abnormal production process analysis module. Yield statistics & control module 1 is connected with alarm grouping 2 module for data transmission. Bin Ratio correlation-based alarm grouping module 21 is connected with Bin Map feature-based alarm grouping module 22 and Wafer Bin Ratio and numerical / non-numerical parameter correlation analysis module 31 for data transmission. The Map feature alarm grouping module 22 is respectively connected to the Bin Map feature similarity grouping module 32, the Bin Map and Site Map feature matching module 33, the Bin Map and Defect Map matching analysis module 34 and the Wafer abnormal production process analysis module for data transmission, and the analysis module 3 is connected to the result front-end display output module 4 for data transmission and display.

[0045] Embodiment 2

[0046] The semiconductor industry intelligent yield early warning and diagnostic analysis method of the present invention comprises the following steps:

[0047] Yield Statistics & Control Module 1 generates daily Ratio warning information. Figure 2 , obtain the wafer data of the current day, the yield benchmark data of the historical fixed time, the description data of the die (type: G / B / O / S), pre-process the data, and calculate the return material approval (RMA) / qualified rate statistics (SYL) / statistical classification monitoring system (SBL) according to the corresponding rules, and mark and warn the wafers that exceed the benchmark. If multiple grains of the same product are warned, find the wafer data of the product on the same day, perform correlation analysis according to certain rules, and if relevant, merge them into one warning information. Finally, summarize the daily Ratio warning information.

[0048] Construct a feature library for bin map image similarity retrieval. Figure 3Every day, for each wafer map, bins with spatial correlation are integrated as the input of the classifier. The classifier outputs the classification result and its features of the bin group and saves the bin map image information, which serves as the basic metadata for subsequent image similarity judgment.

[0049] Construct a feature library for defectmap image similarity retrieval. Figure 4 Every day, each defect map is input into the classifier, and the classifier outputs the classification result and features of the defect map, and saves the defect map image information, which serves as the basic metadata for subsequent image similarity judgment.

[0050] Based on the wafer of daily Ratio warning, each warning case is divided into groups according to its map type and similarity coefficient. Figure 5 , and take the majority type of wafer as the new type of bin.

[0051] Based on the above daily Ratio warning information, the correlation analysis of the values ​​(Inline / WAT / WaitTime parameters) is performed. Figure 6 , find the latest N wafers grouped by product name / type / bin, where N is selected according to the parameter type. Query the corresponding parameter type (Inline / WAT / WaitTime) data in the string table, perform correlation analysis using the Spearman correlation coefficient, obtain the Top N analysis results, standardize the format, and store them in the database for easy viewing at any time.

[0052] Based on the above daily Ratio warning information, non-numerical (Tool) correlation analysis is performed. Figure 7 , group the bin ratios in the step according to the Tool parameter. Perform a kw test on each group of bin ratio lists in each step and calculate its P-value. Filter out the Top N steps, calculate the median and variance of the bin ratio lists in the filtered steps and sort them in descending order, filter out the Tool parameter corresponding to the first descending order of each step, that is, the Tool with the worst bin ratio performance, standardize the format, and store it in the database for easy viewing at any time.

[0053] Based on the above daily Ratio warning information, non-numerical (Chamber) correlation analysis is performed. Figure 8, group the bin ratios in the step according to the chamber parameter. Perform the kw test on each bin ratio list in each step and calculate its P-value. Filter out the Top N steps, calculate the median and variance of the bin ratio list in the filtered steps and sort them in descending order, filter out the chamber parameter corresponding to the first descending order of each step, that is, the chamber with the worst bin ratio performance, standardize the format, and store it in the database for easy viewing at any time.

[0054] Based on the above daily Ratio warning information case, non-numerical (Dcoll) correlation analysis is performed. Fig. 9 , obtain the tool equipment list corresponding to each case, search for the corresponding Dcoll table according to the tool list, group by Dcoll table, and group the sub-analyses of the same equipment type (with the same Dcoll table name) in different cases. Obtain the corresponding Dcoll table data, track the wafer data within a certain period of time, calculate the Pearson correlation according to specific groups, retain the top N parameters of the Dcoll of each tool in each case, standardize the format, and store them in the database for viewing at any time.

[0055] In the Wafer Bin Map (WBM) analysis, filter out the Bad / Good data within the fixed time of the warning Bin Map. Fig.10 Based on the pre-classified Wafer Bin Map (WBM) data and abnormal Wafer Bin Map (WBM) data within a fixed historical time, the Wafer that has been warned is searched for similarity to obtain the Bad / Good data within a fixed time based on the Bin Map.

[0056] Based on the wafers corresponding to the bad / good cases retrieved similarly, perform correlation analysis on the values ​​(Inline / WAT / WitTime parameters). Fig.11 , find the latest N wafers of the corresponding product name / type / bin grouping, where N is selected according to the parameter type. Query the corresponding parameter type (Inline / WAT / WaitTime) data in the string table, use the kw algorithm to calculate the correlation between the bin ratio and the parameter data, obtain the Top N analysis results, standardize the format, and store them in the database for easy viewing at any time.

[0057] Based on the wafers corresponding to the bad / good non-cross-product cases retrieved by similarity, non-numerical (Tool) correlation analysis is performed. Here, the number of bad lots and good lots of the tool corresponding to each case on the wafer list are counted separately. For each tool under each step, the bad and good statistics are calculated according to the lot, and the tool with the largest badcount under the step is selected, that is, the tool located at the step. Sort by the number of bad lots in descending order and the number of good lots in ascending order, obtain the Top N tool parameters, standardize the format, and store them in the database for easy viewing at any time.

[0058] Based on the wafers corresponding to the bad / good non-cross-product cases retrieved by similarity, non-numerical (Chamber) correlation analysis is performed. Here, the number of bad lots and good lots of each case corresponding to the chamber on the wafer list are counted separately. For each chamber under each step, the bad / good statistics are performed according to the lot, and the tool with the largest bad count under the step is selected, that is, the chamber located in the step. Sort by the number of bad lots in descending order and the number of good lots in ascending order, obtain the Top N chamber parameters, standardize the format, and store them in the database for easy viewing at any time.

[0059] Based on the wafers corresponding to the bad / good non-cross-product cases retrieved by similarity, non-numerical (Dcoll) correlation analysis is performed. Fig. 9 , obtain the tool equipment list corresponding to each case, search for the corresponding Dcoll table according to the tool list, group by Dcoll table, and group the sub-analyses of the same equipment type (with the same Dcoll table name) under different cases. Obtain the corresponding Dcoll table data, track the wafer data within a certain period of time, calculate the Pearson correlation between the bin ratio value and the parameter data according to the specific grouping, retain the top N parameters of the Dcoll of each tool under each case, standardize the format, and store them in the database for viewing at any time.

[0060] Inline site map analyzes the relationship between inline parameters and binratio of the daily warning fail bin from the Map dimension. Fig.12According to the bin map warned every day, find the inline parameter closest to the centroid of the bin map, do a correlation analysis between the failure rate of the bin historical time period and the nearest neighbor inline parameter, select the Top N strongly correlated inline parameters, and draw a scatter plot and a dual-axis trend chart of the inline / bin ratio.

[0061] The WAT site map analyzes the relationship between the WAT parameters and the bin ratio of the daily warning fail bin from the Map dimension. Fig.13 , mainly for the fail bins warned of the jam, at the bin map level (different from the above pure numerical analysis), combined with the numerical analysis method to find the WAT parameter that causes the corresponding fail bin closest to the bin map centroid, select the TopN inline parameters with strong correlation, and draw the scatter plot and dual-axis trend chart of the inline / bin ratio. This lays the foundation for finding the Tool / Chamber that affects the corresponding bin in the subsequent Module trouble shooting (FDC / ESPC / Dcoll / offline / Inline Defect).

[0062] Based on the abnormal (die) wafers that are warned every day, find the top N most similar abnormal WaferBin Maps (WBM) corresponding to them, and perform similar searches within a fixed historical time based on the three dimensions of the same wafer / same product / cross-products. Fig.14 For similar searches of the same wafer, the Pearson correlation coefficient is used to calculate the correlation coefficient between the bin map and the defect map based on features. For similar searches of the same product or across products, minHashLSH is used to calculate the similarity between the bin map and the defect map, and product_name is used to filter out the top 10 similar searches of the same product or across products.

[0063] The present invention is based on the manufacturing and process data collected during the wafer manufacturing process to provide early warning of abnormal production and help diagnose abnormalities, thereby improving the yield. According to the semiconductor wafer manufacturing yield test data, a yield abnormality warning and abnormal factor diagnosis and analysis method are proposed. We can discover the type of abnormal phenomenon and yield changes early, and then diagnose and analyze the abnormal yield factors to find out the suspicious factors or problems that ultimately lead to it. The contribution of the present invention is to develop a method for intelligent yield warning and diagnostic analysis in the semiconductor industry. It can shorten the analysis time for finding the cause of defects, help engineers narrow the scope of suspicious factors, and improve the quality of decision-making.

[0064] The above are only preferred specific implementation modes of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and inventive concepts of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An intelligent yield early warning and diagnostic analysis device for the semiconductor industry, comprising a yield statistics and control module (1), an alarm grouping module (2), an analysis module (3) and a result front-end display output module (4), characterized in that: The yield statistics and control module (1) includes a Daily Bin Ratio control and early warning module (11), a control and early warning BaseLine configuration module (12) and a Daily yield statistics module (13). Daily data statistics are combined with the control BaseLine configuration information to perform early warning summary to obtain Ratio early warning information. Every day, bins with spatial correlation are integrated for each wafer map as input to a classifier. The classifier outputs the classification result and features of the bin group and saves the bin map image information. Every day, each defect map is input to the classifier. The classifier outputs the classification result and features of the defect map and saves the defect map image information. The alarm grouping module (2) includes an alarm grouping module (21) based on Bin Ratio correlation and an alarm grouping module (22) based on Bin Map feature. Based on daily Ratio warning information, analysis is performed in two categories: Wafer-based Ratio correlation analysis and Wafer Bin Map (WBM)-based analysis. The analysis module (3) includes a Wafer Bin Ratio and numerical or non-numerical parameter correlation analysis module (31), a Bin Map feature similarity grouping module (32), a Bin Map and Site Map feature matching module (33), and a Bin Map and Defect Map matching analysis module (34). Based on the wafers warned by daily Ratio, each warning case is divided into groups according to its map type and similarity coefficient, and the majority type of the wafer is taken as its new bin type. Based on the wafer Ratio correlation analysis, a numerical or non-numerical correlation analysis is performed. In the Wafer Bin Map (WBM) analysis, based on the pre-Wafer Bin Map (WBM) classification data and abnormal Wafer Bin Map (WBM) classification data within a fixed historical time, the warned wafer is subjected to similarity retrieval to obtain Bad or Good data within a fixed Bin Map time. Based on the Bad or Good data within a fixed Bin Map time, a numerical or non-numerical correlation analysis is performed. Based on the Wafer Bin Map (WBM) abnormality warned by daily Ratio, a Wafer Bin Map (WBM) is performed. Analyze the top N Inline parameters and WAT parameters of the nearest neighbors of the Map (WBM) centroid. Based on the abnormal wafers warned every day, find the top N most similar abnormal wafer BinMap (WBM) corresponding to it, and perform similar searches within a fixed historical time according to the three dimensions of the same wafer, the same product, or across products; The yield statistics and control module (1) is connected to the alarm grouping module (2); the Bin Ratio correlation-based alarm grouping module (21) is respectively connected to the Bin Map feature-based alarm grouping module (22) and the Wafer Bin Ratio and numerical or non-numerical parameter correlation analysis module (31); the Bin Map feature-based alarm grouping module (22) is respectively connected to the Bin Map feature similarity grouping module (32), the Bin Map and Site Map feature matching module (33) and the Bin Map and Defect Map matching analysis module (34); and the analysis module (3) is connected to the result front-end display output module (4).

2. A method for performing intelligent yield warning and diagnostic analysis in the semiconductor industry using the semiconductor industry intelligent yield warning and diagnostic analysis device according to claim 1, characterized in that: It is used to analyze the data in the database, statistically analyze the bad wafers, issue early warning notifications for abnormal wafers, and automatically analyze them to point out the main root causes of wafer defects. It includes the following steps: ① Daily data statistics are combined with the control BaseLine configuration information to perform early warning summary and obtain Ratio early warning information; ② Every day, for each wafer map, bins with spatial correlation are integrated as the input of the classifier. The classifier outputs the classification result and features of the bin group and saves the bin map image information; ③ Every day, each defect map is input into the classifier, and the classifier outputs the classification result and features of the defect map, and saves the defect map image information; ④Based on the daily Ratio warning information, the analysis is divided into two categories: Ratio correlation analysis based on Wafer and analysis based on Wafer Bin Map (WBM); ⑤ Based on the wafer with daily Ratio warning, each warning case is divided into groups according to its map type and similarity coefficient, and the mode type of the wafer is taken as its new bin type; ⑥Based on Wafer's Ratio correlation analysis, perform numerical or non-numerical correlation analysis; ⑦ In the Wafer Bin Map (WBM) analysis, based on the pre-Wafer Bin Map (WBM) classification data and abnormal Wafer Bin Map (WBM) classification data within a fixed historical time, the warning Wafer is searched for similarity with them to obtain Bad or Good data based on the Bin Map fixed time; ⑧ Perform numerical or non-numerical correlation analysis based on Bad or Good data within a fixed time of Bin Map; ⑨Based on the Wafer Bin Map (WBM) anomaly warned by daily Ratio, analyze the first N Inline parameters and WAT parameters of the nearest neighbors of the Wafer Bin Map (WBM) centroid; ⑩ Based on the abnormal wafers warned every day, find the top N most similar abnormal wafer bin maps (WBM) corresponding to them, and perform similar searches within a fixed historical time period based on the three dimensions of the same wafer, the same product, or across products.

3. The method for intelligent yield early warning and diagnostic analysis in the semiconductor industry as claimed in claim 2, characterized in that: Statistics are collected based on daily data, and early warnings are summarized based on the BaseLine configuration information of management and control.

4. The method for intelligent yield early warning and diagnostic analysis in the semiconductor industry as claimed in claim 2, characterized in that: Every day, for each wafer map, bins with spatial correlation are integrated as the input of the classifier, and the classifier outputs the classification result and characteristics of the bingroup.

5. The method for intelligent yield early warning and diagnostic analysis in the semiconductor industry as claimed in claim 2, characterized in that: Every day, each defect map is input into the classifier, and the classifier outputs the classification result and characteristics of the defect map.

6. The method for intelligent yield early warning and diagnostic analysis in the semiconductor industry as claimed in claim 2, characterized in that: Based on the wafer of daily Ratio warning, each warning case is divided into groups according to its map type and similarity coefficient.

7. The method for intelligent yield early warning and diagnostic analysis in the semiconductor industry as claimed in claim 2, characterized in that: Based on Wafer's Ratio correlation analysis, perform numerical or non-numerical correlation analysis.

8. The method for intelligent yield early warning and diagnostic analysis in the semiconductor industry as claimed in claim 2, characterized in that: Perform similarity search on the Wafers that have been warned to obtain Bad or Good data within a fixed time based on Bin Map.

9. The method for intelligent yield early warning and diagnostic analysis in the semiconductor industry as claimed in claim 2, characterized in that: Based on the Bin Map site information, the first N Inline parameters and WAT parameter analysis of the nearest neighbors of the Wafer Bin Map (WBM) centroid are analyzed.

10. The method for intelligent yield early warning and diagnostic analysis in the semiconductor industry as claimed in claim 2, characterized in that: For abnormal wafers that have been warned, similarity searches are performed within a fixed historical time period based on the three dimensions of the same wafer, the same product, or across products to find the top N most similar abnormal wafer bin maps (WBM).

Images