TW202422258A - Error factor analysis device and error factor analysis method - Google Patents

Abstract

This error factor analysis device estimates a sub-process that is an error cause from a plurality of sub-processes constituting an inspection process on the basis of a data set measured by an inspection device, and comprises: an error label affixing unit that estimates, from measurement points of sub-processes different from a sub-process including a measurement point where an error is detected, an error-related measurement point related to the measurement point where the error is detected, and affixes an error label to the measurement point where the error is detected and the error-related measurement point; an error correlation calculation unit that estimates a feature quantity having a high correlation to the occurrence of the error in each sub-process from a difference in data between the measurement points to which the error label is affixed and the measurement points to which the error label is not affixed; an abnormality degree calculation unit that, with respect to the feature quantity having a high correlation, calculates the feature quantity-based abnormality degree in each sub-process in accordance with the degree of statistical divergence of data between the measurement points to which the error label is affixed and the measurement points to which the error label is not affixed; and an abnormal process estimation unit that, on the basis of the abnormality degree in each sub-process, estimates the sub-process that is the error cause.

Description

Error factor analysis device and error factor analysis method

The present invention relates to an error factor analysis device and an error factor analysis method for analyzing factors of errors generated by an inspection device or the like.

Semiconductor inspection equipment or semiconductor measurement equipment performs inspection at each inspection point on the surface of a semiconductor wafer or performs measurement at each measurement point according to setting parameters called recipes. However, when an insufficiently adjusted recipe is used or when the characteristics of the device change over time, errors may occur during inspection or measurement, which may reduce the operating rate of the device. When analyzing the factors of such errors, the user of the device must often perform manual operations such as confirming each inspection value, each measurement value, or a photographic image taken by a scanning electron microscope (hereinafter referred to as a SEM (Scanning Electron Microscope) image), and the analysis of the error factors requires considerable time.

As one of the topics in error factor analysis, it is sometimes necessary to correctly infer the cause of the error. The inspection step or measurement step of semiconductors can be further decomposed into multiple small steps such as alignment, addressing, and length measurement. In each small step, the matching score between the template image pre-registered in the recipe and the SEM image of the measurement point is calculated. If the matching score is above the threshold, it is judged that the pattern detection is successful and transferred to the next small step.

However, if the threshold is not set properly, the error judgment may be skipped and the next step may be moved to the next step, even though the step should be judged as a pattern detection failure. In fact, the error is inferred that there is an error in the subsequent step without error. Therefore, in order to correctly analyze the error factor, it is important to correctly infer the error-causing step.

Here, as a prior art for the small step of inferring the occurrence of an error or a bad situation, the substrate inspection system of Patent Document 1 is known. For example, in the abstract of the document, it is recorded that "even an unskilled person can easily identify the cause of the bad situation.", and in paragraph 0056, it is recorded that "in the analysis information storage unit 202, an analysis program database 221, an analysis program selection table 222, a cause-countermeasure table 223, a cause-basis table 224, a display image database 225, etc. are set." In addition, in Figures 12 to 14 of the document, examples of the composition of each table are shown, and in Figure 15, specific processing of the bad cause using each table is disclosed.

Thus, Patent Document 1 discloses a technology of pre-constructing a table of characteristics of the measured values of each step and the cause step, and inferring the cause step by referring to the table. [Prior Technical Document] [Patent Document]

Patent document 1: Japanese Patent Publication No. 2006-339445

[The problem the invention is trying to solve]

Here, if the inspection and measurement recipes used in the semiconductor inspection device or the semiconductor measurement device are changed, the mechanism of error generation will also change. Therefore, in order to utilize the specific treatment of the defective cause of patent document 1, it is necessary to construct various tables according to each recipe. If it is a production step for manufacturing a large number of semiconductor products with a small variety, the same recipe can be used for a long time, so even if various tables are constructed each time the recipe is switched, the labor is not too great. However, if it is a production step for manufacturing a small number of semiconductor products with a large variety, it is necessary to switch the recipe frequently, so it is not practical to construct various tables each time the recipe is switched.

As another method, an abnormality detection method can be considered, which determines whether each step is abnormal by comparing the difference between the normal data collected in advance and the abnormal data, and infers the cause step. However, in this case, since it is necessary to define normal data for each of the dozens or hundreds of recipes that are frequently switched, the knowledge or man-hours required to identify normal data are relatively difficult.

Therefore, the purpose of the present invention is to provide a technology that can infer the cause step of the detected error without constructing a relationship table between the measured value and the error cause step, or without collecting or defining normal data, and infer the error factor by analyzing the data of the error cause step. [Technical means for solving the problem]

In order to solve the above-mentioned problem, the error factor analysis device of the present invention is a device that, when a data set measured by an inspection device includes an error, infers the sub-step of the error cause from a plurality of sub-steps constituting the inspection step based on the above-mentioned data set; and is provided with: an error label assignment unit, which infers an error-related measurement point associated with the measurement point where the error is detected from a measurement point of a sub-step different from the sub-step including the measurement point where the error is detected, and assigns an error label to the measurement point where the error is detected and the error-related measurement point. label; an error correlation calculation unit, which estimates the characteristic quantity highly correlated with the error generation according to the difference between the data of the measurement point assigned with the error label and the measurement point not assigned with the error label for each small step; an anomaly calculation unit, which calculates the anomaly based on the characteristic quantity for each small step according to the degree of statistical difference between the data of the measurement point assigned with the error label and the measurement point not assigned with the error label for the highly correlated characteristic quantity; and an abnormal step estimation unit, which estimates the small step of the error cause based on the anomaly of each small step. [Effect of the invention]

According to the error factor analysis device and error factor analysis method of the present invention, even when the recipe of a product or a measuring or inspection device is changed, the cause of the detected error can be inferred without constructing various tables according to each recipe or collecting or defining normal data.

The topics, structures and effects other than those mentioned above will be clarified by the following description of the implementation forms.

The following uses drawings to illustrate embodiments of the error factor analysis device and the error factor analysis method of the present invention. In addition, in the following, "semiconductor inspection device" not only means a device for measuring the size of a pattern formed on the surface of a semiconductor wafer, but also includes a device for inspecting whether the pattern formed on the surface of a semiconductor wafer has defects, a device for inspecting whether a bare wafer without a pattern formed has defects, and a composite device combining these devices. Furthermore, "inspection" means also used for measurement, and "inspection action" means also used for measurement action. In addition, "inspection object" not only means a wafer that becomes a measurement object or an inspection object, but also refers to a measurement object area or an inspection object area in the wafer. In addition, in the following, "error" includes not only poor measurement conditions or device failures, but also warnings or warning messages that are precursors to errors. Example 1

First, the error factor analysis device of the first embodiment of the present invention is described using FIG. 1 to FIG. 7 .

(Overview of information processing system) Figure 1 is a schematic diagram showing an example of the configuration of the information processing system 100 of this embodiment. As shown in this figure, the information processing system 100 has a semiconductor inspection device 1, a database 2, an error factor analysis device 3, a terminal 4, and a network N. Each of them will be described in turn below.

The semiconductor inspection device 1 is a device for inspecting the size of a pattern formed on the surface of a semiconductor wafer, and is connected to a database 2 or an error factor analysis device 3 via a network N.

The database 2 is a recording device that records device data, recipes, measurement results, error results, etc. sent from the semiconductor inspection device 1.

The error factor analysis device 3 is a device used to analyze the error factor when an error occurs in the inspection step implemented by the semiconductor inspection device 1. Specifically, it is a computer having hardware such as a CPU (Central Processing Unit), a semiconductor memory, and a communication device. The error factor analysis device 3 can be a local deployment used in the facility managed by the user of the semiconductor inspection device 1, or it can be a cloud used outside the facility. In addition, the function of the error factor analysis device 3 can also be incorporated into the semiconductor inspection device 1.

The terminal 4 is a device having a display functioning as a GUI (Graphical User Interface) for presenting the analysis result of the error factor analysis device 3 to the user, and is communicatively connected to the error factor analysis device 3 via a wired or wireless communication line.

(Data recorded in database 2) The data sent from semiconductor inspection device 1 and recorded in database 2, and then analyzed by error factor analysis device 3, for example, includes device data, recipes, measurement results, and error results. The following describes each data in turn.

The device data includes device-specific parameters, device error correction data, and observation condition parameters. The device-specific parameters are correction parameters used to make the semiconductor inspection device 1 operate according to the specified specifications. The device error correction data are parameters used to correct the error between semiconductor inspection devices. The observation condition parameters are parameters that specify the observation conditions of a scanning electron microscope (SEM), such as the acceleration voltage of the electron optical system.

The recipe includes wafer mapping, various parameters (alignment parameters, addressing parameters, length measurement parameters), template images, etc. Wafer mapping is the coordinate mapping of the surface of the semiconductor wafer (for example, the coordinates of the pattern). Alignment parameters are, for example, parameters used to correct the offset between the coordinate system of the surface of the semiconductor wafer and the coordinate system inside the semiconductor inspection device 1. Addressing parameters are, for example, information that specifies the characteristic pattern formed on the surface of the semiconductor wafer and existing in the inspection object area. Length measurement parameters are parameters that describe the conditions for measuring the length, for example, parameters that specify which part of the pattern is to be measured for length. Template images are reference images used to match the pattern to detect measurement points.

In addition, the recipe may include the number of evaluation points, coordinate information of the evaluation points (EP), imaging conditions when taking images, etc. In addition, the recipe may also include the coordinates of images obtained in the preparation stage for measuring the evaluation points or imaging conditions, etc.

The measurement results include length measurement results, image data, and action logs. The length measurement results record the results of measuring the length of the pattern on the surface of the semiconductor wafer. The image data is the observed image of the semiconductor wafer. The action log is the data that records the internal state of the semiconductor inspection device 1 in each action step of alignment, addressing, and length measurement. For example, the action voltage of each component, the coordinates of the observation field, etc. can be cited.

The error result is a parameter that shows which of the alignment, addressing and length measurement steps caused the error when an error occurs.

(Overview of inspection steps) Figure 2 shows an example of inspection steps of the semiconductor inspection device 1. The inspection steps illustrated here are decomposed into five action steps (sub-steps) from step P1 to step P5. That is, in step P1, the position offset between the wafer and the stage of the semiconductor inspection device is corrected by alignment in the optical (OM: Optical Microscope) mode. Then, in step P2, the position offset between the wafer and the stage is corrected by alignment in the electron beam (SEM) mode. Subsequently, in steps P3 and P4, the field of view is moved to the length measurement coordinate by addressing. Finally, in step P5, the size of the pattern formed on the surface of the wafer is measured. Through the above inspection steps, the semiconductor inspection device 1 inspects the semiconductor wafer.

Here, in each action step, the matching score between the template image registered in the recipe and the SEM image of the photographed measurement point is calculated. If the matching score is above the threshold, the pattern detection is successful and the next action step is transferred. However, if the threshold is inappropriate, although the pattern detection fails due to the detection of a different position from the original, the error judgment may be skipped and the next action step may be transferred. In this case, the action step that is different from the action step where the error is actually detected and the skipped action step becomes the cause of the error.

Therefore, in this embodiment, even when an error is detected in a subsequent action step, the action step before the error is presumed to be the cause, and the following error factor analysis device 3 is set up in such a way that the error factor can be correctly analyzed based on the data obtained in the previous action step.

(Structure and processing contents of the error factor analysis device) Figure 3 shows the detailed structure of the error factor analysis device 3 of this embodiment. As shown here, the error factor analysis device 3 has a step data selection unit 31, an error label assignment unit 32, an error correlation calculation unit 33, an abnormality calculation unit 34, an abnormality storage unit 35, and an abnormal step estimation unit 36, which outputs an abnormal step estimation result D1. Furthermore, the error correlation calculation unit 33 includes a characteristic quantity generation unit 33a, an error classification model learning unit 33b, and a model analysis unit 33c, and the abnormality calculation unit 34 includes a characteristic quantity selection unit 34a and a characteristic quantity basic calculation unit 34b. In addition, each functional unit is realized by executing a predetermined program read into a memory device by a general computer, i.e., an operation device of the error factor analysis device 3.

Fig. 4 is a process flow chart of the error factor analysis device 3. In the following, the details of the error factor analysis process of the error factor analysis device 3 are described while referring to Fig. 3 and Fig. 4 as appropriate.

((Step S1)) First, in step S1, the step data selection unit 31 uses the formula for analyzing the error factor to obtain the data set measured by the semiconductor inspection device 1 from the database 2, and divides the data according to each action step. Generally speaking, since the data set collected in the database 2 is given an identification number for distinguishing the action step at which it is measured, the identification number can be used to divide the data set into the measurement data of each action step.

Furthermore, the measurement data of the action step that may be related to the detected error is selected from the segmented measurement data. The action step that may be related to the error may be, for example, the action step that detected the error and its upstream action step, a series of action steps of repeated actions of addressing-length measurement, or action steps with close measurement timing.

((Step S2)) Then, in step S2, the error label assigning unit 32 assigns an error label to the measurement point where an error is detected for the step data selected in step S1, and also assigns an error label to the related measurement point (error-related measurement point) that is estimated to have an impact on the inspection performance of the measurement point.

FIG5 shows an example of step-specific data with error labels. FIG5 is a diagram showing an extract of step P1 data and step P5 data constituting a data set measured by semiconductor inspection device 1. In each step-specific data, the "wafer INDEX" in the first row from the left is an identification number for distinguishing semiconductor wafers. In addition, the "measurement No" in the second row from the left is a continuous number indicating the number of measurement points in the same semiconductor wafer. For example, if an error is detected in the measurement item with "Measurement No." of "12" for the semiconductor wafer whose "Wafer INDEX" is "XXX001" in the step P5 data, the method for estimating the error-related measurement point in this step will be to infer the measurement point with the same "Wafer INDEX" as the semiconductor wafer and a smaller "Measurement No." (for example, the measurement point with "Wafer INDEX" of "XXX001" and "Measurement No." of "0" or "1" in the step P1 data) as the error-related measurement point. Then, as shown in the figure, the error label assigning unit 32 assigns an error label "1" to the error-related measurement point estimated in this way and the error detection measurement point that serves as the starting point thereof.

((Step S3)) In step S3, the error-related calculation unit 33 selects one of the step-specific data selected in step S1 for which the abnormality calculation is not completed (the step P1 data or the step P5 data in the example of FIG. 5 ) as the processing object after step S4.

((Step S4)) In step S4, the error correlation calculation unit 33 generates a feature quantity suitable for input of the machine learning model from the data selected in step S3 (e.g., the step P1 data in FIG5 ). Therefore, specifically, the following processing is performed.

First, the feature quantity generating unit 33a calculates the correlation of each feature quantity with respect to the error generation based on the difference in data between the measurement points with error labels and the measurement points without error labels assigned in step S2. Next, the feature quantity generating unit 33a generates a feature quantity suitable for inputting a machine learning model that distinguishes between data with error labels and data without error labels, using the step identification data of the processing object. Here, the generation of the feature quantity can be performed by, for example, calibration or statistical processing of the measurement data, encoding of the category variables, and production of a composite feature quantity by combining a plurality of data such as alternating feature quantities.

Next, the error classification model learning unit 33b learns an error classification model, which takes the feature quantity generated by the feature quantity generating unit 33a or the error label assigned in step S2 as input, and classifies the data of the measurement points with error labels and the data of the measurement points without error labels based on the difference in tendency. The error classification model can be generated using an algorithm based on a decision tree such as Random Forest or XGBoost (Extreme Gradient Boosting), or any machine learning algorithm such as Neural Network.

Furthermore, the model analysis unit 33c calculates the correlation of each input feature quantity on the model output, i.e., the error prediction result, for the error classification model learned by the error classification model learning unit 33b. This correlation can be evaluated, for example, by calculating the variable importance (Feature Importance) based on the number of branches of each feature quantity in the model or the improvement value of the objective function, or by calculating the SHAP (Shapley Additive exPlanations) value of the correlation between the value of each feature quantity and the model output when constructing an error detection model by an algorithm based on a decision tree.

((Step S5)) In step S5, the feature quantity selection unit 34a selects the feature quantity with a higher correlation calculated in step S4. As the selection method, for example, a method of selecting the upper-order N feature quantities in order of high correlation, or a method of selecting feature quantities with a correlation greater than a preset threshold value can be used.

((Step S6)) In step S6, the feature value-based calculation unit 34b calculates the difference between the data of the measurement point with an error label and the measurement point without an error label for the error selected in step S5 as the abnormality of the action step. The difference can be, for example, the Euclidean distance or the Mahalanobis distance.

((Step S7)) In step S7, the abnormality memory unit 35 stores the abnormality of each action step calculated in step S6.

((Step S8)) In step S8, it is determined whether the processing from step S3 to step S7 has been performed on all the action step data selected in step S1. And, if the requirements are met, it proceeds to step S9. If the requirements are not met, it repeats the processing from step S3 to step S7 until the abnormality of all the action step data is calculated.

((Step S9)) In step S9, the abnormal step estimation unit 36 uses the abnormality of each action step type stored in step S7 to estimate the action step that is the cause of the error. As the method of estimating the error cause step here, for example, it can be set to the action step with the highest abnormality, or the most upstream action step with an abnormality above a preset threshold, or the action step estimated using a machine learning model that has learned the relationship between the characteristic quantity with a high correlation with the error, the abnormality of the step type, and the cause step, etc.

The estimation result of the error cause step in this step is output as the abnormal step estimation result D1. The abnormal step estimation result D1 is an error analysis result including the abnormality degree of each action step stored in the abnormality degree storage unit 35 or the error cause step estimated by the abnormal step estimation unit 36. The error analysis results are presented to the user via the GUI, i.e., the terminal 4.

FIG6 shows an example of a method for notifying the user of the error analysis result. In this example, the size of the abnormality of each action step is displayed by the length of a bar, and the step P2 estimated as the cause of the error by the processing of the flowchart of FIG4 is displayed in a different color (dark color) from other steps. In this way, even if an error is detected in step P5, the user can be informed that the cause of the error occurs in step P2 where no error is detected.

(Effect of Example 1) In the above-described example, unlike the previous abnormality detection method that must be prepared according to each recipe table or definition, the error information generated is directly used to infer the error-related measurement points contained in the data of each previous action step. In addition, the error-related measurement points are regarded as error data, and the difference between the error data and other data is calculated as the abnormality of each step.

Thus, the error cause step can be estimated without using a pre-prepared relationship table between the measured value and the error cause step, or the collection and definition of normal data. Therefore, according to this embodiment, even when the recipe is frequently updated in order to manufacture a small amount of a variety of semiconductor products, the error cause corresponding to each recipe can be easily estimated. Example 2

Next, the error factor analysis device 3 of the second embodiment will be described with reference to Fig. 7 and Fig. 8. In addition, the common points with the first embodiment will be omitted from repeated description.

As is clear from the comparison between FIG3 and FIG7, the error factor analysis device 3 of the second embodiment is the error factor analysis device 3 of the first embodiment with the addition of a matching score-based abnormality calculation unit 37 and a comprehensive abnormality calculation unit 38. The details thereof will be described in turn below.

(Matching score-based anomaly calculation unit 37) Generally speaking, the matching scores of the template image pre-registered in the recipe and the SEM image of the measurement point are recorded in the database 2. In the matching score-based anomaly calculation unit 37, for the step-specific data that may be the cause of the error selected by the step-specific data selection unit 31, the difference between the measurement point with the error label assigned to the matching score and the measurement point without the error label is calculated as the anomaly. The difference can be calculated using the Z score based on the mean value or standard deviation of the original data. The step-specific anomaly based on the matching score is stored in the anomaly storage unit 35.

(Comprehensive abnormality calculation unit 38) In the comprehensive abnormality calculation unit 38, the comprehensive abnormality is calculated from each abnormality stored in the abnormality storage unit 35. FIG8 shows a schematic diagram of the comprehensive abnormality calculation. The comprehensive abnormality of each step is calculated from the abnormality based on the feature quantity and the abnormality based on the matching score. As the calculation method, for example, a method of simply adding the abnormalities of the two steps, a method of obtaining the maximum value of the two, or a method of adding them after multiplying them by a preset weight, etc. can be used.

(Abnormal step estimation unit 36) In the abnormal step estimation unit 36, the comprehensive abnormality of each step calculated by the comprehensive abnormality calculation unit 38 is used to estimate the step that is the cause of the error based on the height or threshold of the abnormality, similarly to the first embodiment.

When the comprehensive anomaly is calculated as in the present embodiment, the GUI, i.e., the terminal 4, may display only the final comprehensive anomaly, or may display both the feature-based anomaly and the matching score-based anomaly, which are the basis for calculating the comprehensive anomaly, together with the comprehensive anomaly. In addition, in FIG8, as in FIG6, the step estimated as the cause of the error by the abnormal step estimation unit 36 is displayed in a color different from that of other steps (dark color). However, if the error step estimation method adopted by the abnormal step estimation unit 36 is, for example, a situation in which the most upstream step whose abnormality exceeds the threshold is estimated as the cause of the error, then as in the example of FIG8, the most upstream step P1 whose abnormality exceeds the threshold is sometimes estimated as the cause of the error instead of the step P2 with the largest abnormality.

(Effect of Example 2) In Example 2, there are multiple mechanisms for calculating the abnormality of each step, and the abnormality after the combination of them is calculated. In this way, information from multiple angles can be used in the estimation of the error cause step, and the estimation accuracy can be improved by reducing the omission of error characteristics. Example 3

Next, the error factor analysis device 3 of the third embodiment will be described with reference to Fig. 9 and Fig. 10. In addition, the common points with the first embodiment will be omitted from repeated description.

As can be seen from the comparison between Fig. 3 and Fig. 9, the error factor analysis device 3 of the third embodiment is different from the first embodiment and includes a step error factor estimation unit 39, an error dictionary 3A, and an error factor probability correction unit 3B. The details of these are described in turn below.

(Step-by-step error factor estimation unit 39) The step-by-step error factor estimation unit 39 uses the feature quantity or its value, correlation, etc. with a high error correlation calculated by the error correlation calculation unit 33 to retrieve items with a high similarity from the error dictionary 3A, and uses the retrieved error factor or the similarity as the error factor probability D2, and obtains one or more of them according to the step. As a calculation method for the similarity, for example, a coordinated filtering or ranking learning estimation method can be used.

(Error information accumulation section 3A) Error dictionary 3A accumulates the error information by associating the feature quantity or its value, the combination of correlation, and the error factor. As the accumulation method, the error features and their factors can be systematized in a table form based on past technical know-how, or past error data and their error factors can be stored in advance.

(Error factor probability correction unit 3B) In the error factor probability correction unit 3B, the step-specific abnormality stored in the abnormality storage unit 35 is used to correct the step-specific error factor probability D2 calculated by the step-specific error factor estimation unit 39. A schematic diagram of the correction method is shown in FIG10.

In FIG10 , the error factors and their error probabilities calculated by the step-by-step error factor estimation unit 39 are obtained as the top three of the step-by-step. The error probability is corrected by the abnormality of the step-by-step stored in the abnormality storage unit 35, and the error probability of the upper level after correction is obtained together with the step number as the error factor estimation result D3. As a correction method, for example, the value of the abnormality can be normalized to [0.0-1.0] and multiplied by the error probability, or a coefficient that makes the step with an abnormality above the threshold higher and the step with an abnormality above the threshold lower can be calculated and multiplied by the error probability, etc. The error factor estimation result D3 thus obtained is prompted to the user via the terminal 4.

(Effect of Example 3) In Example 3, the error factor corresponding to the abnormality of each step can be estimated by correcting the estimated probability of the error factor obtained for each step with the abnormality of each step. In this way, even in an example where an error occurs due to the interaction of multiple steps, the error factor in each step can be selected and prompted to the user.

(Variations) This disclosure is not limited to the above-mentioned embodiments, and includes various variations. For example, the above-mentioned embodiments are described in detail to explain this disclosure in an easy-to-understand manner, and may not have all the structures described. In addition, a part of a certain embodiment may be replaced with the structure of another embodiment. In addition, the structure of another embodiment may be added to the structure of a certain embodiment. In addition, for a part of the structure of each embodiment, a part of the structure of another embodiment may be added, deleted, or replaced.

For example, in the above-mentioned embodiments 1 to 3, although the example of estimating the error factor of the semiconductor inspection apparatus 1 is described, the error factor of an error generated in a device other than the semiconductor inspection apparatus 1 may also be estimated.

Furthermore, in the above-mentioned embodiment 2, although two abnormalities, namely, the abnormality based on the feature quantity and the abnormality based on the matching score, are used as the abnormality calculation mechanism for each step, a calculation mechanism for three or more abnormalities other than the two abnormalities may also be used.

Furthermore, in the above-mentioned embodiment 3, the error factor estimation unit 39, the error dictionary 3A, and the error factor probability correction unit 3B are added to the form of embodiment 1, but it can also be configured to add these to the form of embodiment 2.

Furthermore, in the above-mentioned embodiment 3, the information accumulated in the error dictionary 3A may include a recipe correction as information to be presented to the user with the error factor estimation result D3, and the error factor and the recipe correction may be presented together.

1: Semiconductor inspection device 2: Database 3: Error factor analysis device 3A: Error dictionary 3B: Error factor probability correction unit 4: Terminal 31: Step data selection unit 32: Error label assignment unit 33: Error correlation calculation unit 33a: Feature generation unit 33b: Error classification model learning unit 33c: Model analysis unit 34: Abnormality calculation unit 34a: Feature selection unit 34b: Abnormality calculation unit 35: Abnormality storage unit 36: Abnormal step estimation unit 37: Abnormality calculation unit based on matching score 38: Comprehensive abnormality calculation unit 39: Step-by-step error factor estimation unit 100: Information processing system D1: Abnormal step estimation result D2: Error factor probability D3: Error factor estimation result N: Network P1~P5: Step S1~S9: Step

FIG. 1 is a schematic diagram showing an example of the configuration of an information processing system of Embodiment 1. FIG. 2 is an example of an inspection step of a semiconductor inspection device. FIG. 3 is a functional block diagram of an error factor analysis device of Embodiment 1. FIG. 4 is a processing flow chart of an error factor analysis device of Embodiment 1. FIG. 5 is an example of step-specific data assigned an error label. FIG. 6 is an example of an analysis result displayed at a terminal. FIG. 7 is a functional block diagram of an error factor analysis device of Embodiment 2. FIG. 8 is a schematic diagram for explaining a method for calculating a comprehensive abnormality of Embodiment 2. FIG. 9 is a functional block diagram of an error factor analysis device of Embodiment 3. FIG10 is a schematic diagram for explaining the calculation method of the error factor estimation result of Example 3.

2: Database

3: Error factor analysis device

31: Step data selection section

32: Error labeling department

33: Error related calculation department

33a: Feature mass production department

33b: Error classification model learning department

33c: Model analysis department

34: Abnormality calculation department

34a: Feature selection section

34b: Feature quantity basic calculation unit

35: Abnormal Memory Department

36: Abnormal step estimation department

D1: Abnormal step inference results

Claims (10)

An error factor analysis device is characterized in that, when a data set measured by an inspection device contains an error, it infers the sub-step of the error cause from a plurality of sub-steps constituting the inspection step based on the data set, and has: an error label assignment unit, which infers error-related measurement points associated with the measurement points where the error is detected from the measurement points of the sub-steps different from the sub-steps containing the measurement points where the error is detected, and assigns error labels to the measurement points where the error is detected and the error-related measurement points; An error correlation calculation unit estimates a characteristic quantity highly correlated with the error generation in each small step from the difference between the data of the measurement point assigned the error label and the measurement point not assigned the error label; An anomaly calculation unit calculates the anomaly based on the characteristic quantity in each small step according to the degree of statistical difference between the data of the measurement point assigned the error label and the measurement point not assigned the error label; and An abnormal step estimation unit estimates the small step of the error cause based on the anomaly of each small step. The error factor analysis device of claim 1 further comprises: A second anomaly calculation unit, which estimates the matching score between the template image of the measurement point and the photographic image of the measurement point, and calculates the anomaly of the matching score according to the statistical difference between the matching scores of the measurement point assigned with the error label and the measurement point not assigned with the error label; and A comprehensive anomaly calculation unit, which calculates the comprehensive anomaly of each small step based on the anomaly of each small step calculated by the above-mentioned anomaly calculation unit and the anomaly of each small step calculated by the above-mentioned second anomaly calculation unit; and The above abnormal step estimation section estimates the small steps of the error cause based on the comprehensive abnormality of each small step. The error factor analysis device of claim 1 or 2 further comprises: an error dictionary that accumulates the relationship between the characteristics of the errors and their error factors or parameter corrections; and a step-by-step error factor inference unit that searches the error dictionary for items with a combination of error characteristics similar to the statistical information of the highly relevant feature quantity or measurement result, and selects candidates for error factors or parameter corrections in each small step. As in claim 3, the error factor analysis device, wherein according to the abnormality of each of the above-mentioned small steps, the similarity of the error factors or parameter corrections of each of the above-mentioned small steps is corrected, and the error factors or parameter corrections of each small step with higher similarity after correction are selected and prompted to the user. The error factor analysis device of claim 4, wherein for the number of measurement points without errors, when the probability of detecting an erroneous measurement point is below a threshold, the similarity of some items in the above selected corrected similarity is reduced. The error factor analysis device of claim 1, wherein the abnormal step estimation unit estimates the abnormality of each small step to be above the threshold and the most upstream step as the small step of the error cause. The error factor analysis device of claim 2, wherein the abnormal step estimation unit estimates the step most upstream of each small step for which the above-mentioned comprehensive abnormality is above the threshold as the small step of the error cause. The error factor analysis device of claim 1, wherein the abnormal step estimation unit constructs a model of the relationship between the abnormality of each small step, the items/correlations of the highly correlated feature quantities, and the error cause step after learning, and sets the step estimated by the model as the error cause step. The error factor analysis device of claim 2, wherein the abnormal step estimation unit constructs a model of the relationship between the above-mentioned comprehensive abnormality of each small step, the items/their correlation of the above-mentioned highly correlated feature quantities, and the error cause step after learning, and sets the step estimated by the above-mentioned model as the error cause step. A method for analyzing error factors, characterized in that, when a data set measured by an inspection device contains an error, a sub-step of inferring the cause of the error from a plurality of sub-steps constituting an inspection step is inferred based on the data set, and the method comprises: assigning an error label to a step, inferring an error-related measurement point associated with the measurement point where the error is detected from a measurement point of a sub-step different from the sub-step containing the measurement point where the error is detected, and assigning an error label to the measurement point where the error is detected and the error-related measurement point; The error correlation calculation step is to estimate the characteristic quantity highly correlated with the error in each small step from the difference between the data of the measurement point assigned with the error label and the measurement point not assigned with the error label; The anomaly calculation step is to calculate the anomaly based on the characteristic quantity in each small step according to the degree of statistical difference between the data of the measurement point assigned with the error label and the measurement point not assigned with the error label; and The abnormal step estimation step is to estimate the cause of the error based on the anomaly of each small step.