CN113642257A - Wafer quality prediction method and system - Google Patents
Wafer quality prediction method and system Download PDFInfo
- Publication number
- CN113642257A CN113642257A CN202111189877.6A CN202111189877A CN113642257A CN 113642257 A CN113642257 A CN 113642257A CN 202111189877 A CN202111189877 A CN 202111189877A CN 113642257 A CN113642257 A CN 113642257A
- Authority
- CN
- China
- Prior art keywords
- wafer
- prediction
- quality
- prediction model
- key
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06395—Quality analysis or management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/04—Manufacturing
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Abstract
The invention discloses a wafer quality prediction method and a wafer quality prediction system, and relates to the technical field of semiconductor processing. The invention comprises the following steps: acquiring sensing acquisition data and corresponding wafer quality grade, wherein the sensing acquisition data is acquired by a sensor which is arranged at a processing station; the method comprises the steps that sensing collected data are used as an input layer of a prediction model, corresponding wafer quality grades are used as an output layer of the prediction model, the prediction model is trained, and the prediction model is trained until the recognition is accurate; screening out processing stations and sensors which are sensitive to the quality grade of the wafer, wherein the processing stations and the sensors are respectively key processing stations and key sensors; when the wafer to be measured is processed, the key sensors in the key processing stations acquire sensing acquisition data, and the sensing acquisition data is input into the prediction model to obtain the wafer prediction quality grade. According to the invention, the sensor is used for acquiring the sensing acquisition data generated in the wafer processing process, the quality grade of the wafer is predicted, and the problem of low wafer processing detection efficiency is solved.
Description
Technical Field
The invention belongs to the technical field of semiconductor processing, and particularly relates to a wafer quality prediction method and a wafer quality prediction system.
Background
Wafer processing needs to use a plurality of processing stations to perform a plurality of processing, and in order to ensure the final yield, wafer quality detection needs to be performed for a plurality of times in the processing process of the wafer. At present, when quality detection is carried out on a wafer, the detection efficiency is low from the perspective of electrical data of the wafer. And because the wafers produced by batch processing cannot be detected completely, the detection is carried out by adopting a random sampling mode in the traditional mode, and the detection efficiency cannot be further improved on the basis of ensuring the detection accuracy.
Disclosure of Invention
The invention aims to provide a wafer quality prediction method and a wafer quality prediction system, which can be used for predicting the quality grade of a wafer by acquiring sensing acquisition data generated in the wafer processing process through a sensor, thereby solving the problem of low wafer processing detection efficiency.
In order to solve the technical problems, the invention is realized by the following technical scheme:
the invention provides a wafer quality prediction method, which comprises the following steps: acquiring sensing acquisition data and corresponding wafer quality grade, wherein the sensing acquisition data is acquired by a sensor which is arranged at a processing station;
using the sensing collected data as an input layer of a prediction model, using the corresponding wafer quality grade as an output layer of the prediction model, and training the prediction model until the recognition is accurate;
Screening out the processing stations and the sensors which are sensitive to the quality grade of the wafer, wherein the processing stations and the sensors are respectively key processing stations and key sensors;
and when the wafer to be measured is processed, inputting the sensing acquisition data acquired by the key sensor in the key processing station into the prediction model so as to obtain the wafer prediction quality grade.
In one embodiment of the invention, the method further comprises:
in the processing process of the wafers to be detected at the key processing station, at least one group of the sensing acquisition data of the wafers to be detected is acquired through the key sensor;
respectively inputting the sensing and collecting data into the prediction model to obtain the wafer prediction quality grade;
selecting the wafer to be measured corresponding to the lowest value in the wafer prediction quality grades for online measurement to obtain an online measurement result;
and judging the quality of the wafers to be detected according to the on-line measurement result.
In one embodiment of the invention, the method further comprises:
comparing the on-line measurement result with the wafer prediction quality grade to judge whether the prediction model is accurate or not;
And if not, retraining the prediction model according to the online measurement result.
In one embodiment of the invention, the method further comprises:
obtaining a plurality of wafer prediction quality grades of at least one wafer in a key processing station;
and obtaining a wafer prediction final yield interval according to the plurality of wafer prediction quality grades.
In one embodiment of the invention, the method further comprises:
actually measuring the actual final yield of the wafer;
comparing the actual final yield of the wafer with the predicted final yield interval of the wafer to judge whether the prediction model is accurate or not;
and if not, retraining the prediction model according to the actual final yield of the wafer.
In an embodiment of the present invention, the step of using the sensor data as an input layer of a prediction model and using the wafer quality grade as an output layer of the prediction model to train the prediction model and train the prediction model until the recognition is accurate includes:
dividing the sensing collected data and the corresponding wafer quality grade into a training set and a testing set;
Taking the sensing collected data in the training set as an input layer of the prediction model, and taking the wafer quality grade in the training set as an output layer for training the prediction model;
comparing and matching the sensing acquisition data in the test set with the sensing acquisition data in the training set to obtain a wafer quality grade corresponding to the sensing acquisition data in the test set;
and comparing the wafer quality grade with the wafer quality grade corresponding to the test set, and if the accuracy is higher than a set value, judging that the prediction model is accurately identified.
In an embodiment of the present invention, the step of screening out the processing stations and the sensors sensitive to the quality level of the wafer, which are respectively key processing stations and key sensors, includes:
selecting a plurality of wafers with different wafer quality grades, wherein the wafer quality grades are actual measurement results;
acquiring the sensing acquisition data acquired by the sensor in the process of processing the wafer;
for a plurality of wafers with different wafer quality grades, if the sensor data acquired by the sensor are the same or similar, the sensor is taken as a non-key sensor;
After the non-key sensors are eliminated, obtaining the key sensors;
the processing station where the key sensor is located is the key processing station.
In an embodiment of the present invention, a plurality of key sensors in the same key processing station acquire a plurality of sensor acquisition data of the same wafer to be tested;
and if the plurality of wafer prediction quality grades obtained after the plurality of sensing collected data are input into the prediction model are not completely the same, selecting a mode of the plurality of wafer prediction quality grades as the wafer prediction quality grade.
In an embodiment of the present invention, a plurality of key processing stations perform a plurality of processes on the same wafer to be tested;
acquiring the wafer prediction quality grade of the wafer to be detected after each processing, and summarizing the wafer prediction quality grades into a plurality of wafer prediction quality grades;
selecting a mode of the plurality of wafer prediction quality grades as the wafer prediction quality grade.
The present invention also provides a wafer quality prediction system, comprising:
the data acquisition unit is used for acquiring sensing acquisition data and corresponding wafer quality grades, wherein the sensing acquisition data is acquired by a sensor, and the sensor is arranged on a processing station;
The model training unit is used for taking the sensing collected data as an input layer of a prediction model, taking the corresponding wafer quality grade as an output layer of the prediction model, training the prediction model and training the prediction model until the recognition is accurate;
the screening unit is used for screening the processing stations and the sensors which are sensitive to the quality grade of the wafer, and the screening units are respectively key processing stations and key sensors;
and the prediction output unit is used for inputting the sensing acquisition data acquired by the key sensor in the key processing station into the prediction model when the wafer to be detected is processed so as to obtain the predicted quality grade of the wafer.
According to the wafer prediction method, the prediction model is established through the sensing collected data and the wafer quality grade, the wafer to be detected is predicted according to the prediction model, and the wafer prediction quality grade generated by the prediction model can be further processed to obtain the wafer prediction final yield interval. Moreover, when a batch of wafers to be detected are subjected to online measurement in a sampling inspection mode, the wafers to be detected with relatively low predicted wafer quality level can be selected as samples for online measurement, and the wafer quality detection efficiency can be further improved.
Of course, it is not necessary for any product in which the invention is practiced to achieve all of the above-described advantages at the same time.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a first flowchart illustrating a wafer quality prediction method according to an embodiment of the present invention;
FIG. 2 is a second flowchart illustrating a wafer quality prediction method according to an embodiment of the present invention;
FIG. 3 is a third flowchart illustrating a wafer quality prediction method according to an embodiment of the present invention;
FIG. 4 is a flowchart illustrating the steps of training a predictive model according to one embodiment of the present invention;
FIG. 5 is a schematic view of a target map conversion summary of the present invention in accordance with one embodiment;
FIG. 6 is a schematic flow chart diagram illustrating the steps for screening key process stations and key sensors according to one embodiment of the present invention
FIG. 7 is a block diagram of a wafer quality prediction system according to an embodiment of the present invention.
In the drawings, the components represented by the respective reference numerals are listed below:
1-a data acquisition unit, 11-a sensor;
2-a model training unit;
3-a screening unit;
4-prediction output unit.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1, the present invention provides a method for predicting wafer quality to improve the problem of low wafer processing inspection efficiency. The present invention may include the step of, step S1, first acquiring the sensor data and the corresponding wafer quality grade using the sensor 11. The sensor 11 is disposed at a processing station, and the sensing Data acquired by the sensor 11 may be real-time sensor Data (FDC Raw Data) of a machine. The wafer quality levels may include high, medium, and low levels. Wherein the high level may be a corresponding high yield group, e.g. 91% to 100%. The medium level may be a corresponding medium yield group, for example, 85% to 90%. The low level may be a corresponding low yield group, which may be 80% to 85%, for example. Meanwhile, in order to conveniently sample and detect the wafer, the high grade, the medium grade and the low grade in the wafer quality grade can be respectively corresponding to sigma, 2 sigma and 3 sigma. Specifically, since the mass of a plurality of wafers after processing is normally distributed, σ corresponding to the high level indicates that the wafer belongs to the first 68.27% of the mass distribution from the high level to the low level, 2 σ corresponding to the middle level indicates that the wafer belongs to the first 95.45% of the mass distribution from the high level to the low level, and 3 σ corresponding to the low level indicates that the wafer belongs to the first 99.74% of the mass distribution from the high level to the low level. During each processing of the wafer, the sensor 11 in the processing station acquires sensing data for analyzing the processing quality of the wafer. At step S2, the predictive model may be a convolutional neural network model having an input layer and an output layer. And in the process of training the prediction model, the sensing collected data is used as an input layer of the prediction model. And taking the corresponding wafer quality grade as an output layer of the prediction model. The method is used for training the prediction model and training the prediction model until the recognition is accurate. Because the accurate prediction model can be identified according to the sensing collected data collected by the sensor 11, the wafer quality grade can be predicted, and the wafer quality detection efficiency is higher.
In order to refer to fig. 1, in step S3, because the data collected by the sensor 11 is redundant, the processing station and the sensor sensitive to the wafer quality level may need to be screened out and labeled as a critical processing station and a critical sensor, respectively. And step S4, when the wafer to be measured is processed, inputting the sensing acquisition data acquired by the key sensors in the key processing stations into the prediction model, so as to obtain the wafer prediction quality grade. The quality of the wafer to be tested can be judged according to the predicted quality grade of the wafer, and the predicted quality grade of the wafer can be further processed to obtain the score of the wafer. Specifically, the high level, the middle level, and the low level of the wafer quality level may be respectively assigned to 10 points, 5 points, and 0 points, that is, the score ratios may be respectively 100%, 50%, and 0%. In one specific application of the present embodiment, if the wafer manufacturing process has 10 processes, and the full process score of each process is 10, the total score is 100. A wafer currently passes only the first 5 passes, the full score is 50, and if the algorithm calculates the score to be 40, the current score ratio of the wafer is 40/50= 80%. If the wafer is processed through all 10 processes, the full score is 100, and the algorithm score is 70, the final score ratio of the wafer is 70/100= 70%. The class to which the score ratio is closest is the predicted class for the wafer, e.g., 70% is closer to 50% as described above, and belongs to the "medium yield group", and the predicted future yield for the wafer may be between 86-90. The quality of the wafer to be detected is predicted through the prediction model so as to obtain the predicted quality grade of the wafer, the quality of the wafer can be obtained according to the preset conversion rule between the predicted quality grade of the wafer and the yield interval, and the predicted quality grade of the wafer can be predicted and generated according to the real-time sensor Data (FDC Raw Data) of the machine acquired in the wafer processing process, so that the problem of low wafer processing detection efficiency is solved.
Referring to fig. 2, in order to improve the efficiency of the on-line measurement, the wafer with the lower quality level may be selected from a plurality of wafers for the on-line measurement, and if the wafer with the lower quality level is qualified, it may be determined that there is no quality problem with other wafers. The specific step is to add the following step, step S5.1, to the step S4, in the process of processing the wafers to be measured at the key processing station, a group of sensing collected data of the wafers to be measured is acquired by the key sensor. And S6.2, respectively inputting the sensing and collecting data into a prediction model to obtain the wafer prediction quality grade. And S7.3, selecting the wafer to be measured corresponding to the lowest value in the predicted quality grades of the wafers for online measurement, and then obtaining an online measurement result. And S8.1, judging the quality of the wafers to be detected according to the online measurement result. When a plurality of wafers are processed in the same processing channel, the mass distribution of the wafers is in normal distribution, and if the parts with relatively low mass can be selected for detection and qualification, the quality of the rest of the wafers can be presumed to be qualified. Therefore, in this embodiment, the wafer to be measured with the lowest predicted wafer quality level is selected to perform the on-line measurement, and the obtained on-line measurement result represents the quality of the wafer to be measured with the lowest relative distribution of the quality among the batch of wafers to be measured. In this embodiment, the wafer to be detected with relatively low mass distribution can be selected as the sample for sampling detection, so that the detection efficiency and accuracy are higher.
Referring to fig. 2, after the prediction model is trained in step S2, the problem of inaccurate recognition still exists, and in order to overcome the problem, the online measurement result may be compared with the wafer prediction quality level to determine whether the prediction model is accurate, and if not, the prediction model is retrained according to the online measurement result to improve the recognition accuracy of the prediction model.
Referring to fig. 3, the sensor 11 in each processing station can acquire the predicted quality level of the wafer in the processing process, and can collect the predicted quality levels generated in all the previous critical processing stations, so as to obtain the final yield interval. The specific steps may include the following steps. Step S5.2, a plurality of wafer predicted quality levels of at least one wafer in the critical processing station may be obtained. And S6.2, summarizing to obtain a wafer prediction final yield interval according to the plurality of wafer prediction quality grades. According to the previous process of wafer processing, when the wafer does not complete all the processing processes, the predicted final yield interval of the wafer can be calculated and predicted, and therefore the technical effect of improving the wafer quality detection efficiency is achieved.
Referring to fig. 3, the prediction model is trained and tested by using the data in the training set and the test set, and may not be completely and accurately identified in the subsequent prediction judgment, and may also be supplemented and corrected by combining with the subsequently updated data. The specific steps may include actually measuring the actual final yield of the wafer. And comparing the actual final yield of the wafer with the predicted final yield of the wafer to judge whether the prediction model is accurate, and if not, retraining the prediction model according to the actual final yield of the wafer, so that the recognition accuracy of the prediction model can be further improved.
Referring to fig. 4, the training of the prediction model in step S2 may specifically include the following steps. And S2.1, dividing the sensing collected data and the corresponding wafer quality grade into a training set and a testing set. And S2.2, taking the sensing and collecting data in the training set as an input layer of the prediction model, and taking the wafer quality grade in the training set as an output layer for training the prediction model. And S2.3, comparing and matching the sensing collected data in the test set with the sensing collected data in the training set to obtain the wafer quality grade corresponding to the sensing collected data in the test set. And S2.4, comparing the wafer quality grade with the wafer quality grade corresponding to the test set, and judging that the prediction model is accurately identified if the accuracy is higher than a set value. By checking the accuracy of the prediction model, the misjudgment of the prediction model is avoided.
Referring to fig. 1 to 5, in the process of testing and calibrating a model by using a test set, a CATS algorithm may be used to compare and match the sensor data collected in the test set with the sensor data collected in a training set, so as to obtain a wafer quality grade corresponding to the sensor data collected in the test set, and draw a wafer target map. Then, target maps of a plurality of wafers can be collected to obtain wafer prediction results, the accuracy of the prediction model can be determined according to the wafer prediction results, and the recognition accuracy of the prediction model can be further improved.
Referring to fig. 6, the screening key sensors and the screening key processing stations in step S3 may specifically include the following. And S3.1, selecting a plurality of wafers with different actually measured wafer quality grades. And S3.2, acquiring sensing acquisition data acquired by the sensor 11 in the wafer processing process. And step S3.3, regarding a plurality of wafers with different wafer quality grades, if the sensor data acquired by the sensor 11 are the same or similar, taking the sensor 11 as a non-critical sensor. And S3.4, removing the non-key sensors to obtain key sensors. And S3.5, the processing station where the key sensor is located is the key processing station. Because tens or even tens of processing procedures are performed in the wafer processing process, but the information collected by each sensor is not valuable, the key sensors and key processing stations can be screened out to reduce the workload of data processing and improve the workload of predictive model training and predictive model identification.
Referring to fig. 1 to 6, during a wafer processing process, after a plurality of sensor 11 in the same processing station collects a plurality of sensor collection data and inputs the sensor collection data into a prediction model, a plurality of different wafer prediction quality levels may be generated, so that the wafer prediction quality levels need to be processed. The specific processing steps may be as follows. And acquiring a plurality of sensing acquisition data of the same wafer to be detected by a plurality of key sensors in the same key processing station. And if the plurality of wafer prediction quality grades obtained after the plurality of sensing collected data are input into the prediction model are not completely the same, selecting a mode of the plurality of wafer prediction quality grades as the wafer prediction quality grade. Through mutual verification of the plurality of sensors 11, misjudgment caused by accidental errors of data acquired by the single sensor 11 can be effectively avoided.
Referring to fig. 1-6, a wafer processed at a plurality of critical processing stations may result in a plurality of different predicted quality levels for the wafer, so that the wafer may be processed with different data. The specific processing steps may be as follows. And multiple key processing stations carry out multiple processing on the same wafer to be detected. And acquiring the wafer prediction quality grade of the wafer to be detected after each processing, and summarizing the wafer prediction quality grades into a plurality of wafer prediction quality grades. The mode in the plurality of wafer prediction quality grades is selected as the wafer prediction quality grade, and misjudgment caused by accidental errors of data acquired by a single sensor can be effectively avoided.
Referring to fig. 7, the present invention provides a wafer quality prediction system, which may include a data obtaining unit 1, a model training unit 2, a screening unit 3, and a prediction output unit 4. The data acquisition unit 1 may include a sensor 11, the sensor 11 may be configured to acquire sensing data, and the data acquisition unit 1 may also be configured to acquire a wafer quality level corresponding to the sensing data. The model training unit 2 may be used for training of a predictive model, which comprises an input layer and an output layer. And taking the sensing collected data as an input layer of the prediction model, taking the corresponding wafer quality grade as an output layer of the prediction model, and training the prediction model until the recognition is accurate. The screening unit 3 may be used to screen out processing stations and sensors 11 that are sensitive to wafer quality levels, being critical processing stations and critical sensors, respectively. When the wafer to be measured is processed, the prediction output unit 4 inputs the sensing acquisition data acquired by the key sensors in the key processing stations into the prediction model so as to obtain the wafer prediction quality grade. The sensor 11 in the data acquisition unit 1 acquires the sensing acquisition data in the wafer processing process without additionally using equipment to acquire other data, then the model training unit 2 is used for training a prediction model with accurate recognition, and the sensing acquisition data acquired in the wafer processing process to be detected is input and predicted according to the prediction model to obtain the wafer prediction quality grade for predicting and judging the wafer quality. The quality of the wafer can be predicted and judged by directly using the sensing and collecting data generated in the wafer processing process, so that the wafer quality detection efficiency can be improved.
Please refer to fig. 1 to 7, and the following tables 1 to 3 for specific description. In table 1, Wafer-01 is processed at the same processing station 5m. eem10, and the three sensors sensor1, sensor2 and sensor3 collect sensing data, and the Wafer quality grade is obtained according to the actual measured yield. And training the prediction model according to the sensing collected data and the wafer quality grade corresponding to the sensing collected data. Wherein, since the actual yield of the wafer is 98%, the wafer is of high grade H, the wafer score is 10, and the score ratio is 100%. Eem10, three sensors, namely, sensor1, sensor2 and sensor3, collect sensing data, and obtain Wafer quality grade according to actual measured yield. And training the prediction model according to the sensing collected data and the wafer quality grade corresponding to the sensing collected data. Since the actual yield of the wafer is 80%, the wafer is of high grade L, the wafer score is 0, and the score ratio is 0%. Eem10, three sensors, namely, sensor1, sensor2 and sensor3, collect sensing data, and obtain Wafer quality grade according to actual measured yield. And training the prediction model according to the sensing collected data and the wafer quality grade corresponding to the sensing collected data. Wherein, since the actual yield of the wafer is 98%, the wafer is of high grade M, the wafer score is 5, and the score ratio is 50%.
TABLE 1
In table 2, for the sampling measurement application, a wafer with a low score ratio is selected for the on-line measurement. In table 2, the predicted results of the Wafer-04 collected by the sensors 1, 2, and 3 are respectively the high level H, the low level L, and the Wafer scores are respectively 10, 0, and 0, so the score ratio is 10/30=33.3%, and corresponds to 3 σ. The Wafer-05 collected and predicted by the sensors 1, 2, and 3 respectively has a high level H, and a medium level M, and the Wafer scores are 10, and 5, respectively, so that the score ratio is 25/30=83.3%, corresponding to 1 σ. The predicted results of the Wafer-06 collected by the sensors 1, 2, 3 are respectively the high level H, the middle level M, and the corresponding Wafer scores are respectively 10, 5, and 5, so the score ratio is 20/30=66.7%, corresponding to 2 σ. And selecting the Wafer-04 for online measurement, wherein if the Wafer-04 is qualified, the Wafer-05 and the Wafer-06 are also qualified.
TABLE 2
In table 3, a specific application of the final yield interval is predicted for predicting the wafer of a certain wafer. For Wafer-07, in the processing process of 7 processing stations, 1c.cdg20, 1t.cdg20, 2d.iia20, 2i.ppu10, 1f.wwm30, 2i.iid20 and 1m.cda10, the Wafer prediction quality grades are respectively high grade H, medium grade M, low grade L, high grade H, medium grade M and low grade L, the Wafer scores are respectively 10, 5, 0, 10, 5 and 0, so the score ratio is 57.1%, and since 57.1% is close to 50%, it can be predicted that Wafer-07 belongs to the medium yield group, that is, the Wafer prediction final yield interval is 86% -90%.
TABLE 3
In summary, a prediction model is established by using the wafer with the measured yield and the acquired sensing acquisition data in the processing process, and then the sensing acquisition data acquired in the processing process of the wafer to be measured is input into the prediction model, so that the predicted quality grade of the wafer can be obtained, and the predicted final yield interval of the wafer can be obtained according to the predicted quality grade of the wafer generated in the multi-pass processing. When a batch of wafers to be detected are subjected to online measurement in a sampling mode, the wafers to be detected with lower predicted quality level of the wafers can be selected, if the wafers to be detected with lower predicted quality level of the wafers can pass detection, the quality of the whole batch of wafers to be detected can pass detection, and the wafer quality detection efficiency can be further improved through the sampling mode. By collecting the sensing acquisition data generated in the wafer processing process and establishing the prediction model, the wafer prediction quality grade can be directly obtained according to the sensing acquisition data generated in the wafer processing process, and the problem of low wafer processing detection efficiency is solved.
The preferred embodiments of the invention disclosed above are intended to be illustrative only. The preferred embodiments are not intended to be exhaustive or to limit the invention to the precise embodiments disclosed. Obviously, many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the invention and the practical application, to thereby enable others skilled in the art to best utilize the invention. The invention is limited only by the claims and their full scope and equivalents.
Claims (10)
1. A method for predicting wafer quality, comprising:
acquiring sensing acquisition data and corresponding wafer quality grade, wherein the sensing acquisition data is acquired by a sensor which is arranged at a processing station;
using the sensing collected data as an input layer of a prediction model, using the corresponding wafer quality grade as an output layer of the prediction model, and training the prediction model until the recognition is accurate;
screening out the processing stations and the sensors which are sensitive to the quality grade of the wafer, wherein the processing stations and the sensors are respectively key processing stations and key sensors;
and when the wafer to be measured is processed, inputting the sensing acquisition data acquired by the key sensor in the key processing station into the prediction model so as to obtain the wafer prediction quality grade.
2. The method of claim 1, further comprising:
in the processing process of the wafers to be detected at the key processing station, at least one group of the sensing acquisition data of the wafers to be detected is acquired through the key sensor;
Respectively inputting the sensing and collecting data into the prediction model to obtain the wafer prediction quality grade;
selecting the wafer to be measured corresponding to the lowest value in the wafer prediction quality grades for online measurement to obtain an online measurement result;
and judging the quality of the wafers to be detected according to the on-line measurement result.
3. The method of claim 2, further comprising:
comparing the on-line measurement result with the wafer prediction quality grade to judge whether the prediction model is accurate or not;
and if not, retraining the prediction model according to the online measurement result.
4. The method of claim 1, further comprising:
obtaining a plurality of wafer prediction quality grades of at least one wafer in a key processing station;
and obtaining a wafer prediction final yield interval according to the plurality of wafer prediction quality grades.
5. The method of claim 4, further comprising:
actually measuring the actual final yield of the wafer;
comparing the actual final yield of the wafer with the predicted final yield interval of the wafer to judge whether the prediction model is accurate or not;
And if not, retraining the prediction model according to the actual final yield of the wafer.
6. The method according to any one of claims 1 to 5, wherein the step of using the sensor collected data as an input layer of a prediction model, using the wafer quality grade as an output layer of the prediction model, training the prediction model, and training the prediction model until recognition is accurate comprises:
dividing the sensing collected data and the corresponding wafer quality grade into a training set and a testing set;
taking the sensing collected data in the training set as an input layer of the prediction model, and taking the wafer quality grade in the training set as an output layer for training the prediction model;
comparing and matching the sensing acquisition data in the test set with the sensing acquisition data in the training set to obtain a wafer quality grade corresponding to the sensing acquisition data in the test set;
and comparing the wafer quality grade with the wafer quality grade corresponding to the test set, and if the accuracy is higher than a set value, judging that the prediction model is accurately identified.
7. The method of any of claims 1 to 5, wherein the step of screening out the processing stations and sensors sensitive to the wafer quality level as critical processing stations and critical sensors, respectively, comprises:
selecting a plurality of wafers with different wafer quality grades, wherein the wafer quality grades are actual measurement results;
acquiring the sensing acquisition data acquired by the sensor in the process of processing the wafer;
for a plurality of wafers with different wafer quality grades, if the sensor data acquired by the sensor are the same or similar, the sensor is taken as a non-key sensor;
after the non-key sensors are eliminated, obtaining the key sensors;
the processing station where the key sensor is located is the key processing station.
8. The method according to any one of claims 1 to 5, wherein a plurality of key sensors in the same key processing station acquire a plurality of sensor acquisition data of the same wafer to be tested;
and if the plurality of wafer prediction quality grades obtained after the plurality of sensing collected data are input into the prediction model are not completely the same, selecting a mode of the plurality of wafer prediction quality grades as the wafer prediction quality grade.
9. The method of any one of claims 1 to 5, wherein a plurality of key processing stations perform a plurality of processes on the same wafer to be tested;
acquiring the wafer prediction quality grade of the wafer to be detected after each processing, and summarizing the wafer prediction quality grades into a plurality of wafer prediction quality grades;
and selecting a mode of the plurality of wafer prediction quality grades as the wafer prediction quality grade.
10. A wafer quality prediction system, comprising:
the data acquisition unit is used for acquiring sensing acquisition data and corresponding wafer quality grades, wherein the sensing acquisition data is acquired by a sensor, and the sensor is arranged on a processing station;
the model training unit is used for taking the sensing collected data as an input layer of a prediction model, taking the corresponding wafer quality grade as an output layer of the prediction model, training the prediction model and training the prediction model until the recognition is accurate;
the screening unit is used for screening the processing stations and the sensors which are sensitive to the quality grade of the wafer, and the screening units are respectively key processing stations and key sensors;
And the prediction output unit is used for inputting the sensing acquisition data acquired by the key sensor in the key processing station into the prediction model when the wafer to be detected is processed so as to obtain the predicted quality grade of the wafer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111189877.6A CN113642257A (en) | 2021-10-13 | 2021-10-13 | Wafer quality prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111189877.6A CN113642257A (en) | 2021-10-13 | 2021-10-13 | Wafer quality prediction method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113642257A true CN113642257A (en) | 2021-11-12 |
Family
ID=78426535
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111189877.6A Pending CN113642257A (en) | 2021-10-13 | 2021-10-13 | Wafer quality prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113642257A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101408538A (en) * | 2008-11-04 | 2009-04-15 | 陕西科技大学 | Method for evaluating leather hand feeling quality based on neural network |
| CN110991495A (en) * | 2019-11-14 | 2020-04-10 | 国机智能技术研究院有限公司 | Method, system, medium, and apparatus for predicting product quality in manufacturing process |
| CN112183919A (en) * | 2020-05-22 | 2021-01-05 | 海克斯康制造智能技术(青岛)有限公司 | Quality prediction system and quality prediction method |
-
2021
- 2021-10-13 CN CN202111189877.6A patent/CN113642257A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101408538A (en) * | 2008-11-04 | 2009-04-15 | 陕西科技大学 | Method for evaluating leather hand feeling quality based on neural network |
| CN110991495A (en) * | 2019-11-14 | 2020-04-10 | 国机智能技术研究院有限公司 | Method, system, medium, and apparatus for predicting product quality in manufacturing process |
| CN112183919A (en) * | 2020-05-22 | 2021-01-05 | 海克斯康制造智能技术(青岛)有限公司 | Quality prediction system and quality prediction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115082482B (en) | Metal surface defect detection method | |
| CN107862338B (en) | Marine environment monitoring data quality management method and system based on double inspection method | |
| US7062081B2 (en) | Method and system for analyzing circuit pattern defects | |
| CN109816031B (en) | Transformer state evaluation clustering analysis method based on data imbalance measurement | |
| CN109285791B (en) | Design layout-based rapid online defect diagnosis, classification and sampling method and system | |
| CN108829878B (en) | Method and device for detecting abnormal points of industrial experimental data | |
| CN113763312A (en) | Detecting defects in semiconductor samples using weak marks | |
| KR101797400B1 (en) | Method and apparatus for diagnosing fault based on probabilistic density | |
| CN115100199A (en) | Method for detecting wafer low-texture defects | |
| CN112805719A (en) | Classifying defects in semiconductor samples | |
| CN113191399A (en) | Method for improving yield of semiconductor chips based on machine learning classifier | |
| CN115526258A (en) | Power system transient stability evaluation method based on Spearman correlation coefficient feature extraction | |
| US20030169064A1 (en) | Selective trim and wafer testing of integrated circuits | |
| CN113283288B (en) | Nuclear power station evaporator eddy current signal type identification method based on LSTM-CNN | |
| US6539272B1 (en) | Electric device inspection method and electric device inspection system | |
| CN116452581B (en) | Intelligent voltage source state detection system and method based on machine vision | |
| US6828776B2 (en) | Method for analyzing defect inspection parameters | |
| CN116662800A (en) | Rolling bearing fault diagnosis method based on self-adaptive attention mechanism | |
| CN113642257A (en) | Wafer quality prediction method and system | |
| TWI801973B (en) | Device and method for estimating error factors | |
| CN112397404B (en) | Semiconductor process detection method and system | |
| CN107506824B (en) | Method and device for detecting bad observation data of power distribution network | |
| CN112116014A (en) | Test data outlier detection method for distribution automation equipment | |
| CN111638427A (en) | Transformer fault detection method based on nuclear capsule neuron coverage | |
| TWI745767B (en) | Optical inspection secondary image classification method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20211112 |