Diagnostic system

Abstract

The present disclosure provides a diagnostic system that can appropriately determine the cause even for an error considering multiple factors or multiple composite factors. The diagnostic system according to the present disclosure includes a learner that learns by establishing a correspondence between at least any one of a recipe for operating a specified inspection device, log data describing the state of the device, or sample data describing the characteristics of the sample and the type of error of the device, and uses the learner to estimate the cause of the error (see FIG. 4).

Description

Technical Field

[0001] The present disclosure relates to a diagnostic system for diagnosing the state of an inspection apparatus for inspection samples. Background Art

[0002] A scanning electron microscope (SEM) used for measurement and inspection of semiconductor devices is controlled by a control program (hereinafter also referred to as a recipe in some cases) for setting measurement and inspection conditions. For example, in a scanning electron microscope, a CD-SEM (Critical Dimension-SEM) is also used to measure a sample manufactured through mass production for spot observation and confirm its completion status.

[0003] In Patent Document 1 described below, even when the set conditions of the recipe no longer suit the measurement of the sample due to changes in the manufacturing conditions of the sample or the like and an error occurs, the cause of the error occurrence is also determined. This document discloses a recipe diagnostic apparatus that displays the change over time of a score indicating the degree of agreement of pattern matching, the deviation of coordinates before and after pattern matching, or the change amount of a lens before and after autofocus.

[0004] Prior Art Documents

[0005] Patent Documents

[0006] Patent Document 1: Japanese Patent Application Laid-Open No. 2010-87070 (corresponding to US Patent Publication US2011 / 0147567) Summary of the Invention

[0007] Problems to be Solved by the Invention

[0008] By evaluating changes in the degree of agreement of pattern matching or the like using the apparatus of Patent Document 1 and adjusting the recipe at an appropriate timing, the operation rate of a CD-SEM or the like can be maintained at a high level. However, since various reasons are considered as the cause of the error occurrence, sometimes appropriate adjustment cannot be performed only by evaluating the change in the degree of agreement of pattern matching or the like. In addition, as the cause of the error occurrence, not only changes in the manufacturing conditions of the sample are considered, but also changes in the external environment, changes in the apparatus conditions of a CD-SEM or the like, etc. are considered. In addition, a case where the initial set conditions of the recipe itself are not suitable for measurement is also considered.

[0009] In view of the above problems, the present disclosure provides a diagnostic system that can appropriately determine the cause even for an error considering multiple factors or multiple composite factors.

[0010] Means for Solving the Problems

[0011] The diagnostic system according to the present disclosure includes a learner that learns by establishing a correspondence between at least any one of a recipe for operating a specified inspection device, log data describing the state of the device, or sample data describing the characteristics of the sample and the type of error of the device, and uses the learner to estimate the cause of the error.

[0012] Effect of the Invention

[0013] According to the diagnostic system of the present disclosure, even when multiple factors or multiple composite factors are considered as causes of errors in the inspection device, the cause determination can be appropriately performed. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 FIG. is an example showing a measurement system 10 including a plurality of image acquisition tools.

[0015] Figure 2 It shows the management Figure 1 FIG. is an example showing a diagnostic system for the exemplified measurement system.

[0016] Figure 3 FIG. is an example of a measurement object pattern formed on a sample.

[0017] Figure 4 It more specifically shows Figure 1 FIG. is a diagram showing the exemplified computer system 103.

[0018] Figure 5 FIG. is a flowchart showing a learning process of a system for performing determination of error causes, reliability of generated measurement recipes, prediction diagnosis of errors, etc.

[0019] Figure 6 FIG. is a schematic diagram showing a state of learning teacher data.

[0020] Figure 7 FIG. is a flowchart showing a prediction stage of a learning model constructed using the exemplified learning stage. Figure 5 FIG. is a flowchart showing a process of causing a learner to learn triggered by an error generated during recipe execution.

[0021] Figure 8 FIG. is an example of a report describing the result of the computer system 103 estimating the cause of an error.

[0022] Figure 9 FIG. is a flowchart showing a process of causing a learner to learn triggered by an error generated during recipe execution.

[0023] [[ID=�3]] Figure 10 FIG. is a flowchart showing a process of device repair and relearning of a learning model when an error occurs.

[0024] Figure 11This is a flowchart showing a learning process of a learner that performs predictive diagnosis.

[0025] Figure 12 Instructions for use follow Figure 11 Flowchart of the process of learning a learning model to diagnose the precursors of errors.

[0026] Figure 13 This figure shows an example of a GUI screen for setting learning conditions for causing a learner to learn using error type information and log data output from a CD-SEM or the like when an error occurs.

[0027] Figure 14 This is a diagram showing an example of a diagnostic system that performs unsupervised learning and diagnoses error causes using a learned model generated based on the unsupervised learning.

[0028] Figure 15 Schematic diagram of a learning model constructed through clustering.

[0029] Figure 16 This is a flowchart showing the process of making a learning model learn through unsupervised learning.

[0030] Figure 17 This is a schematic diagram showing how a search around is performed.

[0031] Figure 18 This is a diagram illustrating an example of abnormality determination performed in S2003.

[0032] Figure 19 It is a structural diagram of the device 4100 involved in the third embodiment of the present disclosure.

[0033] Figure 20 This diagram shows an example of a measurement system in which three devices (CD-SEM) 4100 ( 4100 - 1 , 4100 - 2 , and 4100 - 3 ) are connected to a computer system 4120 via a bus or a network.

[0034] Figure 21 This is an example of an analysis screen for performing analysis of the stage position accuracy in each device.

[0035] Figure 22 This is an example of scanning an overlapping test screen.

[0036] Figure 23 This is a flowchart showing a series of processes from the start to the end of the scan overlap test.

[0037] Figure 24 It is a flowchart showing the detailed steps of S4508.

[0038] Figure 25This is an example of a screen showing the execution result of a scan overlap test. Detailed implementation

[0039] <Embodiment 1>

[0040] The following describes a system that outputs the cause of an error generated by a measurement or inspection device (hereinafter simply referred to as an inspection device) based on the device conditions of the input inspection device, the inspection conditions of the inspection device, etc. In addition, a diagnostic system that detects the omen of error generation based on the input of device conditions, etc. is described.

[0041] Figure 1 This is a diagram showing an example of a measurement system 10 including a plurality of image acquisition tools. The image acquisition tool is an SEM that forms an image by detecting electrons (secondary electrons, backscattered electrons, etc.) obtained by scanning an electron beam on a pattern formed on a semiconductor wafer, for example. In this Embodiment 1, as an example of the SEM, a CD-SEM, which is a type of measurement device that measures the size and shape of a pattern based on a detection signal, is described, but it is not limited thereto, and it can be a review SEM that inspects foreign matters and defects based on the given coordinate information. In addition, for example, a focused ion beam device that generates an image based on the scanning of an ion beam can also be used as an image generation tool.

[0042] Figure 1 The system includes three CD-SEMs 100, 101, and 102. Further, in Figure 1 In the illustrated system, the computer system 103 is connected to these CD-SEMs via a bus or a network. An input / output device 104 that inputs and outputs data to and from the computer system 103 is connected to the computer system 103. The computer system 103 can access a measurement recipe storage medium 105 that stores a recipe as an operation program of the CD-SEM, a log information storage medium 113 that stores the device state of the CD-SEM up to the present, a sample information storage medium 106 that stores sample information that is the measurement object of the CD-SEM, and a design data storage medium 114 that stores the design data of a semiconductor device.

[0043] The computer system 103 is composed of one or more computer subsystems. The computer system 103 includes: a computer-readable medium 108; and a processing unit 107 that executes each component (module) stored in the computer-readable medium 108. An analysis component 109 that analyzes the information stored in the storage medium that is accessible and connected to the computer system 103 as described above is stored in the computer-readable medium 108. The analysis component 109 includes a recipe analysis component 110, a sample analysis component 111, and a device analysis component 112.

[0044] In the measurement recipe storage medium 105, the number of measurement points, the coordinate information of the measurement points (Evaluation Point: EP), the imaging conditions at the time of image capture, the imaging sequence, etc. are stored corresponding to the type of sample (e.g., semiconductor wafer). In addition, the coordinates, imaging conditions, etc. of the image obtained in the preparation stage for measuring the measurement points are stored together with the measurement points.

[0045] The image obtained in the preparation stage is, for example, a low-magnification (wide field of view) image for determining the correct field-of-view position, an image for adjusting the optical conditions of the beam at a position other than the measurement target pattern, etc. The low-magnification image is an image captured as a pattern (Addressing Pattern: AP) that includes a unique shape in a known positional relationship with the measurement target pattern. The low-magnification image is used to determine the addressing pattern position by performing pattern matching using a template image that includes a pattern of the same shape as the AP pre-registered on the low-magnification image, and further to determine the measurement target pattern in a known positional relationship with the addressing pattern. The image for adjusting the optical conditions is, for example, an image for auto focus (AF) adjustment, auto astigmatism correction (Auto Astigmatism (AST)), and auto brightness / contrast control (ABCC).

[0046] The imaging conditions at the time of image capture are the acceleration voltage of the beam, the field-of-view (FOV) size, the probe current, the lens conditions, the number of frames (accumulated number of images), etc. These imaging conditions and coordinates are set for each acquired image. The imaging sequence is, for example, the control process of the CD-SEM up to the measurement.

[0047] In the measurement recipe storage medium 105, various measurement conditions are stored in addition to the above examples, and the computer system 103 can read out the stored information as needed.

[0048] In the log information storage medium 113, the device information up to the present is stored in association with the recipe, sample information, or the time when the device information was acquired, etc. Specifically, it uses the position information of the addressing pattern in the low-magnification image at the time of determining the position of the addressing pattern (e.g., deviation from a given position), the time required for AF, the number of images, the gain, bias voltage of the detector output when performing ABCC, the size information of the pattern, the peak height, etc. In addition, it is also possible to store together the outputs of various sensors provided in the CD-SEM, control signals such as voltage values, current values, and DAC values supplied to electrodes, coils, detectors, etc.

[0049] Manufacturing conditions of the semiconductor wafer to be measured and the like are stored in the sample information storage medium 106. For example, when the object to be measured is a resist pattern, there are the type of exposure machine used for forming the resist pattern, exposure conditions (dose, focus value, etc.), the type of resist material, film thickness, size value, and the like.

[0050] Layout data of the semiconductor pattern is stored in the design data storage medium 114.

[0051] The computer system 103 determines the cause of errors generated in the CD-SEM, the reliability of the generated measurement recipe, and the omen diagnosis of errors based on the output of the CD-SEM and the information stored in the above storage media. Specific processing contents will be described later.

[0052] Figure 2 It represents management Figure 1 FIG. 1 is an example of a diagnostic system showing an exemplary measurement system. The diagnostic system 206 performs determination of the cause of errors, reliability of the generated measurement recipe, omen diagnosis of errors, etc. based on the outputs of a plurality of measurement systems 10 including a plurality of CD-SEMs 100, 101, 102, a storage medium 201 storing various information as exemplified above, an input / output device 104, and a computer system (subsystem) 103. Specific processing contents will be described later. Figure 1 FIG. 2 is an example of a measurement object pattern formed on a sample. As a camera sequence, for example, when measuring the width of the upper end of the measurement object pattern 1502, first, a low-magnification image 1501 for addressing is obtained, and the position of the addressing pattern 1504 is determined by pattern matching using the template image 1503. If the position of the addressing pattern can be determined by setting the distance between the addressing pattern 1504 and the measurement object pattern 1502 within the deflection range of the beam deflector built in the CD-SEM, then thereafter, the beam can be irradiated onto the measurement field of view 1506 only by beam deflection without accompanying stage movement.

[0053] Figure 3 When setting the camera sequence, a process for performing positioning of the field of view 1506 to a given position of the pattern 1505 for performing addressing, AST, positioning of the field of view 1507 to a given position for performing AF, positioning of the field of view 1508 to a given position for performing ABCC, and positioning of the field of view 1509 to a position for performing measurement is determined, and condition setting is performed so as to control the CD-SEM in such an order. Furthermore, the position, size, etc. of the measurement cursor 1511 are also determined, where the measurement cursor 1511 determines the measurement reference on the measurement high-magnification image 1510 obtained by beam scanning of the field of view 1509.

[0054] ​

[0055] Figure 4 is a more specific representation Figure 1 of the exemplary computer system 103. As Figure 4 exemplified, the computer system 103 includes an input interface 404, a teacher data generation unit 405, a teacher data storage unit 406 that stores the teacher data generated by the teacher data generation unit 405, a learning unit 407, and a learning model storage unit 408 that stores the learning model learned based on the teacher data. Further, the computer system 103 includes: a first estimator (409), a second estimator (410), a third estimator (411), and a fourth estimator (412) that estimate results by inputting input data into the learning model stored in the learning model storage unit 408. The outputs (estimation results) of these estimators are displayed on a display device of the input / output device 104 via the output interface 413.

[0056] Regarding the data input via the input interface 404, in the learning stage, information such as the type of error is input from the label information storage medium 401, and recipe information, CD-SEM log data, sample information, etc. at the time of error occurrence are input from the learning information storage medium 402. In the estimation stage, recipe information, log data, etc. stored in the estimation information storage medium 403 are used. In addition, the computer system 103 includes a CPU, a GPU, etc. not shown in the figure.

[0057] The teacher data generation unit 405 generates teacher data when information such as the type of error and recipe information are input via the input interface 404. The learning unit 407 uses the teacher data stored in the teacher data storage unit 406 to generate a learning model for error type estimation. The learning unit 407 generates a learning model (learner) based on the data stored in the teacher data storage unit 406 in response to a request input from the input / output device 104, and stores the generated learning model in the learning model storage unit 408.

[0058] The estimators estimate error types, etc. based on the learning model. The first estimator (409) estimates the error type based on the input of recipe information, for example. The second estimator (410) estimates the error type based on the input of log data, for example. The third estimator (411) estimates the error type based on sample information, for example. The fourth estimator (412) estimates the error type based on the outputs of the three estimators, for example. Since there are cases where the cause of error generation is not only one reason but a combination of multiple reasons, in Figure 4 the exemplified system, the fourth estimator (412) estimates the main cause or associated causes of the error based on the outputs of the three estimators. However, the estimation process is not limited to this, and for example, one estimator can also be used to estimate the error cause.

[0059] It is also possible to feedback the information estimated by the estimation unit as new teacher data. As shown by the arrow of the dash-dot line, it is also possible to output the information estimated in the estimation unit and the judgment result of the operator, etc. as teacher data to the teacher data storage unit 406. Figure 4 The arrow shown by the solid line represents the flow of data in the learning stage, and the dashed line represents the flow of data in the estimation stage.

[0060] The learning model is constituted by, for example, a neural network. The neural network outputs error type information, etc. from the output layer by successively propagating the information input to the input layer to the intermediate layer and the output layer. The intermediate layer is constituted by a plurality of intermediate units. The information input to the input layer is weighted by the coupling coefficients between the respective input units and the respective intermediate units, and input to the respective intermediate units. The input to the intermediate unit becomes the value of the intermediate unit by addition. The value of the intermediate unit is non-linearly transformed by an input-output function. The output of the intermediate unit is weighted by the coupling coefficients between the respective intermediate units and the respective output units, and input to the respective output units. The input to the output unit becomes the output value of the output layer by addition.

[0061] By promoting learning, parameters (constants, coefficients, etc.) such as the coupling coefficients between units and the coefficients describing the input-output functions of the respective units are gradually optimized. The storage unit 305 stores these optimized values as the learning result of the neural network. The same applies when using a learning device other than the neural network as the learning device. The same also applies in the following embodiments.

[0062] Figure 4 The exemplified system is a system that inputs at least one of the acquired device information and inspection conditions to an estimator (learning device) that makes an estimation based on a learning model, and outputs at least one of the error generation cause of the inspection device, the device adjustment condition of the inspection device, and the inspection condition of the inspection device, wherein the learning model is learned using teacher data including at least one of device information output from a CD-SEM, etc. (for example, device information stored as log data), the measurement conditions of the CD-SEM (recipe setting conditions, etc.), and at least one of the error generation cause of the inspection device, the device adjustment condition of the inspection device, and the inspection condition of the inspection device.

[0063] Figure 5 It is a flowchart showing the learning process of a system that represents the determination of the implementation error cause, the reliability of the generated measurement recipe, the omen diagnosis of errors, etc. This flowchart is implemented by the computer system 103. The following uses Figure 5 The exemplified flowchart to explain Figure 4 The learning stage of the exemplified system.

[0064] First, the measurement conditions and device conditions of the CD-SEM are initially set (S301). The initial setting conditions are, for example, measurement conditions appropriately set in the recipe and normal device conditions corresponding to the conditions set in the recipe.

[0065] Next, at least one parameter of the device conditions and the measurement conditions is changed (S302), and the measurement process using the CD-SEM is performed under the set conditions (S303). For example, consider changing the conditions set in the recipe to Figure 3 The example of a low-magnification image 1501 with a changed FOV size. If the FOV size is too small, depending on the stage's stopping accuracy, the addressing pattern 1504 may extend beyond the frame of the low-magnification image, potentially causing an addressing error. On the other hand, if the FOV size is too large, the distance to other nearby measurement patterns becomes closer, and due to the charge attached by the beam irradiation, there is a possibility that the beam will be deflected when measuring other measurement patterns, resulting in measurement errors caused by beam drift. This may become a cause of error depending on the recipe's setting conditions. Furthermore, if the beam's energy reaching the sample changes due to power supply anomalies, the charge condition and the image's appearance may differ, potentially causing addressing errors and focusing errors.

[0066] By changing the device and measurement conditions in S302 to create a state prone to errors, the presence or absence of errors and the type of errors when the device is operated in this state are determined and set as label information for the learner. Teacher data is then generated (S304) based on a dataset containing the type of parameter changed, the degree of change, and the combination of the changed parameter with other parameters. The learner is then trained (S305), thereby constructing a learner capable of identifying the type of errors.

[0067] Parameter changes are expected to be implemented both in the case of increasing and decreasing the initial value, and multi-stage changes are also expected. Furthermore, since errors may occur under the combined conditions of different types of parameter changes, it is expected that teacher data will be generated for each combination of various parameter changes.

[0068] Figure 6 is a diagram showing the state of learning teacher data. Figure 7 As illustrated, by generating teacher data using the conditions of the device (parameter changes that lead to errors) and information on the types of errors as a data set, a learner that can identify the causes of errors can be constructed.

[0069] It is also possible to construct teacher data using not only the type of error but also the amount of change as parameter adjustment. In this case, it is possible to estimate not only the type of error but also the adjustment conditions for the recipe used to correct the error. Furthermore, teacher data can be generated from image data generated when an error occurs, along with or instead of the measurement conditions registered in the recipe, to serve as learning data. If there is a correlation between features appearing in the image and the error, teacher data can be generated based on the learning data, enabling early diagnosis of the error.

[0070] Furthermore, it is also possible to use a dynamic image or multiple continuous images instead of a simple still image as the teacher data. For example, the teacher data can be generated based on multiple images (continuous images) obtained when autofocus is executed, or a dynamic image generated from these multiple images. When autofocus is executed, the same FOV is basically scanned continuously, but due to multiple scans, charge is accumulated and there is a possibility that the image drifts. In such a case, there is a case where error-specific information that does not appear in the still image is contained in the dynamic image, etc. In such a case, by generating teacher data based on dynamic images and continuous images, error estimation can be implemented with high precision.

[0071] Figure 7 It means using experience Figure 5 Flowchart of the estimation phase of a learning model constructed during the learning phase. Upon receiving an error signal output by the CD-SEM (S601), the computer system 103 performs an estimation process for the cause of the error. The computer system 103 collects all parameter information selected as or becoming the subject of evaluation corresponding to the error signal from various storage media or the CD-SEM (S602). Depending on the type of error, parameters may be determined. In such cases, parameters are selected as a pre-processing step for performing estimation using the learner. Alternatively, tolerance values may be prepared in advance, and only parameters with significant fluctuations exceeding these tolerances may be selected for estimation (S603).

[0072] The computer system 103 inputs the parameters selected or collected as described above into the learner (estimation unit) (S604), thereby obtaining an estimation result of the error factor, etc. as the output of the learner (S605). The computer system 103 outputs the estimated error factor (S606).

[0073] exist Figure 7In the flowchart, the process of specifying the cause of an error when an error signal output from a CD-SEM or the like is received is illustrated. However, instead of being based on the error signal, it is also possible to receive information such as a decrease in the matching score during addressing or an incorrect detection position of the address as the result of dimensional measurement, which is not recognized as an error but implies the possibility of inappropriate measurement or inspection not being performed, and to infer the state that causes a phenomenon such as a decrease in measurement accuracy. For example, in the case where the dimensional measurement result of a pattern is outside the given allowable range, although there is a possibility that the pattern itself is deformed, on the other hand, there is also a case where the edge is blurred due to a focus error or the like, and the dimensional measurement result shows an abnormal state. Therefore, by performing an inference process triggered by changes in dimensional measurement results or the like (that is, when the measurement accuracy deviates from the previously determined allowable range), it is possible to determine the timing of optimization of the recipe and the occurrence of an abnormal state due to the device.

[0074] Figure 8 This is an example of a report describing the result of the computer system 103 inferring the cause of an error. The output information of the report includes the error information output by the CD-SEM and the inference result inferred by the learner. Specifically, based on the type of error output by the CD-SEM, the type of error is displayed or printed in the Error Detail column. Here, for example, in the case where addressing is not performed appropriately, an addressing error is reported, and in the case where focus adjustment is not performed appropriately, a focus error is reported.

[0075] For example, in the case of an addressing error, when Figure 3 image recognition using the template image 1503 as exemplified in cannot be performed (for example, when performing pattern matching using the template image 1503 in the low-magnification image 1501, the search for a part showing a consistency above a given threshold fails, or multiple candidates are found, etc.), the CD-SEM outputs error information (type, degree, evaluation value for evaluating the error, etc., in this example, an addressing error, the score indicating the consistency of the pattern matching, the threshold for evaluating the score, etc.) to the computer system 103.

[0076] When the computer system 103 receives an error signal, it infers the cause of the error and reports the result. As Figure 8 exemplified, the computer system 103 outputs the inference result of the inferencer together with the recipe information, the sample information, and the device information (log data) as an error score. The error score is output as, for example, the accuracy of the cause of the error. The score is, for example, the output of a neural network in which the parameters of each neuron are adjusted using the error backpropagation method. In addition, it is also possible to coefficientize the degree of contribution to the error for each parameter included in the recipe information and calculate the error score by multiplying the output of the inferencer by this value.

[0077] Figure 9It is a flowchart showing the process of causing the learner to learn triggered by an error generated during the execution of a recipe. This flowchart is implemented by the computer system 103. The computer system 103 can perform the learning process before starting to use the image acquisition tool, or can perform it during the execution of the recipe as shown in this flowchart, or can use both of them.

[0078] First, at least one of the error information output from the CD-SEM at the time of error generation, the recipe information at this time, the log data, and the sample information is collected (S1301), and teacher data is generated based on these data sets (S1302). Furthermore, log data, etc. are collected regularly, and teacher data labeled as normal is generated (S1303). A learner is constructed to estimate two states, the normal state and the error state, and the learner is used to estimate the measurement state, thereby enabling the prediction of error omens.

[0079] For example, log data, etc. are collected regularly, and the log data, etc. are input to two estimators, the normal state evaluation estimator and the error generation estimator, and thus estimation results are obtained from each estimator. If the score of the normal state decreases and the score of the error state increases in the estimation results, then there is a possibility of an error occurring later. The prediction diagnosis of error generation can be implemented through the evaluation of the outputs of the two estimators.

[0080] Figure 10 It is a flowchart showing the process of device repair and relearning of the learning model at the time of error generation. This flowchart is implemented by the computer system 103. In addition to the flowcharts described above, the computer system 103 can, as shown in this flowchart, at the time of error generation, perform the repair process and then execute the recipe, and perform relearning following the result of the re-execution.

[0081] When executing the recipe (S1401) and performing the measurement process, if an error occurs, the computer system 103 follows Figure 7 the flowchart to determine the cause of the error (S1402). The determination of the cause of the error in S1402 can be performed by an operator who is proficient in handling the device. If the cause of the error is known (when the determination result of the learner is above a given score, or when the judgment of the operator is reliable), the recipe is corrected and the device is adjusted (S1403). On the other hand, if the cause of the error is unknown, the error repair operation is performed after stopping the device (S1404).

[0082] When the cause is unknown, operations requiring expertise need to be performed. Therefore, up to S1402, it is implemented by the measurement system 10 owned by the user. When it is difficult for the system or the user's engineer to determine the cause, it can be sent to the management system of the measurement system 10 (for example Figure 2The diagnostic system) conveys this meaning. In this case, the upper learning model that has learned from the teacher data provided by multiple users or the experts belonging to the management company judge the cause of the error, and perform the repair work.

[0083] After performing the adjustments etc. in S1403 to S1404, the formulation is executed again (S1405). If no error occurs by executing again, the response in S1403 or S1404 is correct. Therefore, the computer system 103 generates teacher data (S1406) using the error information and the correction information that describes the correction content implemented in S1403 or S1404 as a data set, and performs relearning of the learning model using it (S1407).

[0084] By implementing Figure 10 the operations represented by the flowchart, a learner that can output an appropriate response strategy based on the input of the type of error generated can be generated.

[0085] Figure 11 is a flowchart showing the learning process of the learner that performs omen diagnosis. This flowchart is implemented by the computer system 103. The computer system 103 can replace Figure 9 the flowchart described in or use it in combination with, and construct a learner that can perform omen diagnosis through the process of this flowchart.

[0086] After receiving an error signal (S801), the computer system 103 reads out the past data stored in the log data etc. (S802). The past data mentioned here is past formulation, past log data, past sample data, etc. For example, looking back a given time from when the error occurred, read out the amount of field movement during addressing, the time required for focus adjustment, the size measurement result, etc. It is also possible to selectively read out the singular points where the parameters in the past data change rapidly (S803). Specifically, it is possible to selectively read out the index value indicating the change of the parameter, the amount of change per given time, or the information on whether it exceeds a given allowable value. Instead, it is also possible to obtain all the parameters acquired at a given timing. Furthermore, instead of the parameter itself, it is also possible to output the change rate of the parameter, the characteristic change as an index value, a mark, and read it out.

[0087] The computer system 103 sets the type of error as a label, generates teacher data (S804) based on the information related to the type of error and the read or extracted parameters, and makes the learner learn using the generated teacher data (S805).

[0088] Figure 12 is a flowchart explaining the process of diagnosing the omen of an error using the learning model that has been learned following Figure 11 This flowchart is implemented by the computer system 103.

[0089] The computer system 103 collects log data in units of given time (S1601), Figure 11 The learning model that has been trained in the illustrated process is fed with collected data to perform error prediction diagnosis (S1602). Since the learner learns based on information such as the type of error that has actually occurred and been identified, and log data before the error occurs, if there is a causal relationship between the events before the error occurs and the error, the occurrence of an error can be inferred before the error actually occurs.

[0090] When the learner predicts a subsequent error with a given degree of accuracy, the computer system 103 can generate a warning signal (S1603) to prompt the device user to take measures such as maintenance or change measurement conditions. If the log data shows a characteristic change before the error occurs, the computer system 103 generates teacher data based on information related to the parameter state, the time from the characteristic change to the error occurrence, the number of measurement points, the number of wafers, the number of batches, and other information. The teacher data is then used to train the learner, thereby creating a learner that outputs information such as the time until the error occurs.

[0091] Figure 13 This figure shows an example of a GUI screen for setting learning conditions for a learner to learn using error type information and log data output from a CD-SEM or the like when an error occurs. Figure 13 The illustrated GUI screen is provided with: a log data display column 1701 for displaying the time changes of the log data output from the CD-SEM, etc. and stored in the log information storage medium 113; a display column 1702 for error information output from the CD-SEM, etc.; and a setting column 1703 for setting the data to be learned.

[0092] The log data display section 1701 displays the transition of multiple parameters stored in the log data and a bar 1704 indicating the timing of error occurrence. Furthermore, the log data display section 1701 displays a pointer 1705 that can be operated using a pointing device (not shown), and further displays a left slider 1706 and a right slider 1707 that can be moved along the horizontal axis of the graph using the pointer 1705.

[0093] By selecting a time using the left slider 1706 and the right slider 1707, parameter change information within that range can be selected as training data. For example, by selecting a parameter that exhibits specific behavior at a specific time associated with an error, efficient learning can be implemented. Alternatively, time selection can be performed by inputting a time into the time setting field 1708 provided in the setting field 1703.

[0094] <Implementation Method 2>

[0095] In Embodiment 1, a structural example of using teacher data to train a learner was described. Instead, a learning model can be generated through unsupervised learning based on parameters during normal operation (when no error occurs). For example, the learner stored in the computer system 103 periodically, or during a process that is relatively more prone to errors compared to other processes, saves the parameters when no error occurs to perform unsupervised learning. The learner that has been trained through unsupervised learning generates and outputs a non-error score for the process when no error occurs. In addition, the learner generates and outputs an error score when an error occurs through unsupervised learning.

[0096] The computer system 103 determines the omen of error occurrence by receiving the output score from the learner and determining whether it is a non-error score or an error score. Errors are caused by a combination of factors such as recipe setting conditions, equipment conditions, and sample conditions, and it is sometimes difficult to correctly identify the cause. However, relevant relationships can be extracted by applying machine learning.

[0097] When the recipe setting conditions, equipment conditions, sample conditions, etc. are input to the learning model generated through unsupervised learning, if the score obtained from the learner through the input is abnormal compared to the non-error score of the non-error learning model, the omen of error occurrence can be detected based on this. Additionally, after learning, data when an error occurs can also be input to the learner, and the output score is compared with the non-error score to determine the range of the non-error score. In this case, when the computer system 103 exceeds the set score range, an alarm indicating the occurrence of an error is generated.

[0098] In the case of unsupervised learning, data when no error occurs can be selectively input for learning. In the mass production process of semiconductor devices, errors rarely occur frequently, and it is sometimes difficult to collect the data required for learning that uses data (parameters) when an error occurs. Since more data can be obtained when no error occurs compared to when an error occurs, learning based on a sufficient amount of learning can be performed.

[0099] Figure 14 FIG. is an example of a diagnostic system that performs unsupervised learning and uses the learned model generated based on this unsupervised learning to diagnose the cause of an error. Figure 14 The illustrated system includes a preprocessing unit 1801 that generates learning data 1802 for the learning unit 407 based on learning information output from a CD-SEM and a management device that manages the CD-SEM. The preprocessing unit 1801 receives data in a normal operating state of the CD-SEM or the like and generates learning data 1802.

[0100] The learning data 1802 includes at least one of information related to object processing, information related to measurement conditions, and device information. Object processing in a CD-SEM, for example, is SEM alignment, addressing, AF adjustment, AST, ABCC, etc. that implement the matching of the coordinate system of the sample stage of the electron microscope and the coordinate system recognized by the electron microscope. In addition, measurement conditions are, for example, the FOV size obtained during addressing, the number of images obtained during AF adjustment, the frame accumulation number, the distance between the EP point and the AF adjustment pattern (or deflection signal amount), the direction, or various lens conditions, etc. In addition, it can also be the actual distance (deflection signal amount) between the EP point and the AF adjustment pattern during measurement. Device information, for example, when there are multiple CD-SEMs that are managed objects of the computer system 103, is information about device attributes such as the identification information of the device and information related to the environment where the device is placed.

[0101] The preprocessing unit 1801 generates a data set based on the above one or more parameters. The learning unit 407 performs clustering on multiple combinations of the above multiple parameters using the learning data 1802, and through clustering, generates one or more clusters for each combination of the parameters.

[0102] Figure 15 It is a schematic diagram of a learning model constructed by clustering. As Figure 15 illustrated, by generating one or more clusters (e.g., classification I, classification II, classification III) for each combination of multiple parameters (e.g., parameter A and parameter B), a learning model 1901 is generated. The learning model 1901 contains multiple clusters for each combination of multiple parameters, and the estimation unit 1803 performs an estimation process using this learning model.

[0103] The computer system 103 determines the error cause of the evaluation object data based on the evaluation object data 1804 output from the preprocessing unit 1801. Specifically, it determines whether the relevant data for each combination of the multiple parameters contained in the evaluation object data is included in one or more clusters contained in the learning model, and determines the parameter related to the relevant data that is not included as abnormal. More specifically, it determines whether the evaluation object data 1902 is included within the range defined for the classification I, II, III of the relevant data 1. In Figure 15 the example, since in relevant 1, the evaluation object data is included within the range 1904 (the range defined by the cluster of classification I), it is determined to be normal. On the other hand, the evaluation object data 1903 of relevant 2 does not belong to any of the classification IV, V, VI defined in relevant 2, so this data is determined to be abnormal.

[0104] By performing estimation using a learner that has performed unsupervised learning as described above, abnormal parameters can be identified. Furthermore, if the measurement target data is not included in all or a predetermined number of categories set for multiple related data, there is a possibility that the learning model has not been properly learned, and therefore it is desirable to recreate the model.

[0105] Once the semiconductor device manufacturing process reaches mass production after the research and development phase, the frequency of errors decreases, making it difficult to create a learning model using data obtained during errors as teacher data. On the other hand, if an error occurs and it takes time to identify its cause, it can also reduce semiconductor device manufacturing efficiency. Therefore, although the frequency is low, rapid device repair is desirable. The learning model generated through unsupervised learning, as described above, can perform appropriate inferences even when the error frequency is low.

[0106] Figure 16 This is a flowchart showing the process of making a learning model learn through unsupervised learning. Figure 16 The illustrated flowchart further includes a step of causing another learning model to perform learning when the CD-SEM is not operating normally or when there is a possibility of this.

[0107] First, the computer system 103 determines that no errors have occurred in the CD-SEM based on data output from the CD-SEM, etc., and then determines the device status at that time (S2001, S2002). The device status is determined by referring to, for example, the evaluation target data 1804. Next, it determines whether the measurement was performed under predetermined measurement conditions. If so, it determines whether the acquired image is abnormal (S2003). If not, it determines whether the measurement was performed under recovery conditions. The abnormality determination process in S2003 will be described later.

[0108] Based on the device information when it is determined that it is not abnormal in the abnormality determination process of S2003, the learning model based on normal data (first learning unit) is learned or re-learned (S2004). Figure 16 The learning data generated by the process illustrated in the example (learning data for the first learning unit) is obtained under conditions that prevent errors from occurring and, as described later, have a low potential for error occurrence. Therefore, unsupervised learning allows the construction of a model suitable for error factor determination. Specifically, by clustering (a) error-free and normal object processing content, (b) measurement conditions, and (c) device information, using parameters belonging to these clusters allows inference that no errors have occurred. In other words, using parameters outside of these clusters allows inference that no errors have occurred.

[0109] According to this flowchart, a learning model for determining the errors that will occur and their causes can be constructed based on device conditions and the like for a state that, although not an error, has the potential to become an error. The second learning model (second learning unit) generated in S2005 is generated based on device conditions and the like during measurement under recovery conditions. By recovery, it does not mean ideal measurement conditions, but rather a process for performing a process prepared in advance to avoid errors and the like. As a specific example, there is searching based on the field of view around the search. As Figure 3 illustrated, addressing is performed to determine the position of the measurement target pattern 1502, but there are cases where the addressing pattern is not included in the low-magnification image 1501 obtained for addressing. When no pattern with a high degree of consistency is found through template matching (there is no part with a consistency above a given value), recovery can be performed by implementing a search around. The search around is a process of searching for a suitable field of view by cycling the field of view in the peripheral area of the current field of view. An example of the search around is described later in Figure 17 .

[0110] On the other hand, performing a search around is considered to mean that the low-magnification image 1501 cannot be obtained appropriately, indicating a state where the device conditions are not set appropriately or a state with a high possibility of becoming an error in the future. Therefore, when performing the recovery process, device conditions and the like are selectively collected, and a model based on them (second learning unit) is generated, whereby a model for predicting the omen of an error can be constructed.

[0111] The recovery process includes not only the search around. For example, there are also: (a) when performing autofocus, when no lens condition with a focus evaluation value above a given value can be found, the variation range of the lens condition is expanded to perform autofocus; (b) a process of repeating (retrying) the same process multiple times, etc. The retry process is not limited to the above and refers to the overall process selectively performed when a certain abnormal condition occurs.

[0112] The second learning model clusters (a) the content of the target process, (b) the measurement conditions, and (c) the device information when further performing the recovery process in the case where there is no error but at least any one of (a) the content of the target process, (b) the measurement conditions, and (c) the device information is abnormal. Thus, if these parameters belonging to the cluster are used, it can be presumed that no error has occurred, and if the recovery process is performed, it can be presumed whether recovery can be achieved through this recovery process.

[0113] Regarding the cases that cannot be recovered by recovery processing, the fourth learning model (fourth learning unit) that clusters these parameters can also be generated in the same way (S2006). Thus, if these parameters belonging to the cluster are used, it can be presumed that no error has occurred, and if the recovery processing is performed, it can be presumed whether the recovery can be achieved by the recovery processing. For example, the estimated scores of the third model and the fourth model can be compared, and based on which one is higher, it can be presumed whether recovery is possible.

[0114] It is also possible to collect device conditions, etc. when it is determined in S2003 that the acquired image, etc. is abnormal, and construct the third learning model (third learning unit) (S2006). According to the model constructed in this way, it is possible to determine a state where appropriate measurement has not been performed although no error occurs. That is, by clustering (a) the content of the target process, (b) the measurement conditions, and (c) the device information, where there is no error and the measurement conditions are normal but the image becomes abnormal, if these parameters belonging to the cluster are used, it can be presumed that although no error has occurred, the image has become abnormal.

[0115] In the case where an error signal is received from a CD-SEM or the like, the computer system 103 constructs the fourth learning model (S2007). Since an error signal is received from the CD-SEM and the type of error is determined, in this case, supervised learning with the type of error as a label can be performed.

[0116] Figure 17 is a schematic diagram showing the state of performing circumferential search. As Figure 17 illustrated, by moving the field of view to surround the periphery of the initial low-magnification image 1501, a circumferential search that performs pattern matching at each field-of-view position is performed, so that the addressing pattern 1504 can be found.

[0117] Figure 18 is a diagram illustrating an example of the abnormality determination performed in S2003. Figure 18 The upper figure of shows the surrounding image 2101 of the measurement target pattern 2102. By addressing, the field-of-view position can be determined at the field-of-view position 2105 including the measurement target pattern 2102, and the image 2107 can be acquired. On the other hand, due to the failure of addressing, the influence of charging, the pattern deformation caused by process variation, etc., the field of view deviates, and the image 2108 at the field-of-view position 2106 is acquired. When the pattern 2103 is measured, a pattern different from the target is measured. In this case, even if there is no error in the device, the output data will be abnormal data.

[0118] Therefore, in order to detect whether such an abnormality has occurred, for example, image data (template) identical to the image 2107 can be prepared in advance, and the degree of consistency of pattern matching can be evaluated during the abnormality determination in S2003, thereby determining whether the acquired image is appropriate (whether it is not an image at the wrong position). In the length measurement image 2107 obtained at the correct field of view position, other patterns 2104 are captured, and the degree of consistency is higher than that in the case of performing template matching for the image 2108. Therefore, it is also possible to determine that abnormal data has been output when the degree of consistency is lower than a given value. In addition, for the abnormality determination, the sharpness of the image, the amount of field of view movement, etc. can also be used as evaluation objects for determination.

[0119] The above-mentioned abnormality determination is implemented in order to appropriately select the model to be the learning object, and it can also be implemented not during the actual recipe execution but after a certain amount of data has been accumulated.

[0120] The learning of the learning model can be implemented in real time during the measurement process such as CD-SEM, or can be implemented at the stage where a certain amount of data has been accumulated offline. Furthermore, it is considered that in a computer system that manages multiple CD-SEMs, when there is an abnormality specific to a particular device, an abnormality caused by the hardware of the device will occur when performing measurements using the same recipe. Therefore, a model caused by the hardware can be created separately, or the identification information of the device can be included in the learning data when creating the above-mentioned first to fourth learning models.

[0121] <Embodiment 3>

[0122] Figure 19 It is a structural diagram of the device 4100 according to Embodiment 3 of the present disclosure. The device 4100 is configured as a CD-SEM as an example. The electrons emitted from the electron source 4101 held in the housing 4124 maintained at a high vacuum are accelerated by the primary electron acceleration electrode 4126 to which a high voltage is applied by the high voltage power supply 4125. The electron beam 4106 (charged particle beam) is converged by the converging electron lens 4127. After the beam current amount of the electron beam 4106 is adjusted by the aperture 1828, it is deflected by the scanning coil 4129 and two-dimensionally scanned on the wafer 4105. The electron beam 4106 is focused by the electron objective lens 4130 disposed directly above the sample, i.e., the semiconductor wafer (hereinafter simply referred to as the wafer) 4105, and is incident on the wafer 4105. The secondary electrons 4131 generated as a result of the incidence of the primary electrons (electron beam 1806) are detected by the secondary electron detector 4132. Since the amount of the detected secondary electrons reflects the shape of the sample surface, the shape of the surface can be imaged based on the information of the secondary electrons.

[0123] The wafer 4105 is held on the electrostatic chuck 4107 while ensuring a certain flatness, and is fixed to the X-Y stage 4104. In Figure 19 it is described with a cross-sectional view of the housing and its internal structure as seen from the side. Therefore, the wafer 4105 can move freely in either the X direction or the Y direction, and any position within the wafer plane can be measured. In addition, the X-Y stage 4104 is equipped with a wafer transfer lifting mechanism 4133, which is an elastomer that can move up and down for disassembling and assembling the wafer 4105 relative to the electrostatic chuck 4107, and can perform the transfer of the wafer 4105 between the loading chamber (preliminary evacuation chamber) 4135 through cooperative operation with the transfer robot 4134. The computer system 4120 performs positioning control of the X-Y stage 4104 based on the detection signal and measurement time from a position detector (such as a laser displacement meter) that detects the position of the X-Y stage 4104 in real time, and records tracking information (log information, movement history information) (the relationship between time and position) related to the moving position of the X-Y stage 4104 in the storage device described later.

[0124] Describe the operation when transporting the wafer 4105, which is the object to be measured, to the electrostatic chuck 4107. First, the wafer placed in the wafer cassette 4136 is transported into the loading chamber 4135 by the transfer robot 4138 in the microenvironment 4137. The loading chamber 4135 can be evacuated and opened to the atmosphere through a vacuum exhaust system (not shown). By opening and closing a valve (not shown) and the operation of the transfer robot 4134, the wafer 4105 is transported onto the electrostatic chuck 4107 while maintaining the vacuum level inside the housing 4124 at a level that is practically acceptable. A surface potentiometer 4139 is installed on the housing 4124. The surface potentiometer 4139 is fixed at a position in the appropriate height direction from the front end of the probe so that the surface potential of the electrostatic chuck 4107 or the wafer 4105 can be measured non-contactingly.

[0125] Each component of the device 4100 can be controlled using a general-purpose computer. In Figure 19 it shows an example of the structure of the control system implemented by the computer system 4120. The computer system 4120 at least includes a processor such as a CPU (Central Processing Unit), a storage unit such as a memory, and a storage device such as a hard disk (including an image storage unit). This storage device can be a structure including the same storage media as the measurement recipe storage medium 105, log information storage medium 113, sample information storage medium 106, and design data storage medium 114 detailed in Figure 1 or can be configured to record in each storage medium the information related to Figure 1The same information. Further, for example, the computer system 4120 can also be configured as a multi-processor system. The main processor constitutes the control related to each component of the electro-optical system within the housing 4124. Regarding the control related to the X-Y stage 4104, the transfer robots 4134 and 4138, and the surface potentiometer 4139, and the image processing for generating the SEM image based on the signal detected by the secondary electron detector 4132, they are respectively constituted by sub-processors.

[0126] An input / output device (user interface) 4141 is connected to the computer system 4120. The input / output device 4141 has: an input device for user input of instructions, etc.; and a display device for displaying the GUI screen for inputting these, the SEM image, etc. The input device can be, for example, a device such as a mouse, a keyboard, or a voice input device that allows the user to input data and instructions. The display device can be, for example, a display device. Such an input / output device (user interface) can be a touch panel that can perform data input and display.

[0127] It is known that when using a CD-SEM to measure the length of a photoresist (hereinafter also referred to as "resist") used in semiconductor lithography ArF exposure technology, etc., the resist shrinks due to the irradiation of the electron beam. In order to reduce the shrinkage amount and measure the length of a fine resist pattern with high precision, it is desired to minimize the irradiation amount of the electron beam on the resist. Therefore, it is necessary to avoid irradiating the same area of such a resist with the electron beam multiple times to perform the length measurement.

[0128] In a CD-SEM, etc., as a method for avoiding irradiating the same area of the resist with the electron beam multiple times, the following method is considered: Before actually executing the measurement recipe (a set or program of the process, processing method, parameters, and specified data supplied to the CD-SEM, etc.), the area for irradiating (scanning) the electron beam is pre-allocated according to the information such as the process and parameters of the measurement recipe, and a measurement recipe is created that does not repeatedly irradiate (scan) the electron beam on the same area.

[0129] Figure 19When the computer system 4120 of the device (CD-SEM) 4100 shown starts the execution of a measurement recipe according to an instruction from an input / output device (user interface) 4141, it controls the X-Y stage 4104 in accordance with information such as the process specified in the measurement recipe, moves the sample 4105 to be measured to a given position, and performs the measurement of the target pattern. Generally, stages used in semiconductor inspection devices and semiconductor manufacturing devices mostly use sliding mechanisms. It is known that the sliding surface characteristics of such sliding mechanisms change over time due to driving conditions such as speed, acceleration, interval, and moving distance, and the wear state of the sliding surface changes. Therefore, even if the X-Y stage 4104 has been initially adjusted for positioning in advance, there may be cases where problems such as the following occur: corresponding to the change over time of the sliding surface characteristics (for example, the clearance (play) between components increases due to wear of the sliding surface), the working range (movable range) of the stage changes slightly; or the stop position accuracy (positioning accuracy) deteriorates. Therefore, in order to create a measurement recipe that does not repeatedly irradiate (scan) the electron beam on the same area as described above, it is necessary to continuously set the required control parameters, etc., based on the change over time of the stop position accuracy of such a stage.

[0130] However, when there are multiple CD-SEMs of the same model, it is desirable to use the same measurement recipe without change in the same measurement and inspection process. This is because, in the case where the same measurement recipe cannot be used and parameter settings of the measurement recipe need to be performed separately for each device, not only does it take time to adjust the parameters for each device, but also the parameters cannot be shared, etc., and the management of the measurement recipe becomes complicated.

[0131] But in order to use the same measurement recipe in multiple devices, even in the case where the small changes in the working range (movable range) of the stage and the variations in the stop position accuracy caused by the change over time as described above are different for each device, the measurement recipe needs to operate without problems for each device. That is, in order to create a measurement recipe common (the same) to each device that does not repeatedly irradiate (scan) the electron beam on the same area of the sample, it is necessary to set the parameters of the measurement recipe, etc., taking into account the change over time of the stage that is different for each device.

[0132] Figure 20FIG. 0 is a diagram showing an example of a measurement system in which three apparatuses (CD-SEMs) 4100 (4100-1, 4100-2, 4100-3) are connected to a computer system 4120 via a bus or a network. The computer system 4120 can access, via the bus or the network, a measurement recipe storage medium 105 that stores an operation program (i.e., a recipe) of the CD-SEM, a log information storage medium 113 that stores the device state of the CD-SEM up to the present, a sample information storage medium 106 that stores sample information, which is the measurement object of the CD-SEM, and a design data storage medium 114 that stores design data of a semiconductor device. In addition, an input / output device 4141 capable of inputting / outputting data to / from the computer system 4120 is connected to the computer system 4120.

[0133] The computer system 4120 is composed of one or more computer subsystems. The computer system 4120 includes: a computer-readable medium 4208; and a processing unit 107 that executes each component (module) stored in the computer-readable medium 4208. Various components 4214 for processing information stored in a storage medium that is accessibly connected to the computer system 103 as described above and information indicated by a user via an input / output device (user interface) 4141 are stored in the computer-readable medium 4208. The various components 4214 include: a wafer information processing component 4209 that processes wafer information and in-chip information related to a wafer on which processing is performed in the apparatus 4100; a recipe information processing component 4210 that processes the measurement sequence, various alignment information, etc.; a stage information processing component 4211 that processes log information recording the stage movement position; a scan overlap test component 4212 that processes scan overlap test information; and a scan overlap test result processing component 4213 that processes information on scan overlap test results. In addition, regarding each common constituent element with Figure 1 the Figure 1 description is the same.

[0134] Figure 21 FIG. 11 is an example of an analysis screen for analyzing the stage position accuracy in each apparatus. The function related to this screen is processed by the stage information processing component 4211.

[0135] The user specifies the target device for analyzing the implementation of stage position accuracy via the Unit menu 4301. The stage information processing component 4211 reads out the tracking information (log information, movement history information) related to the stage movement of the specified target device from the log information storage medium 113 or the storage unit and storage device within the computer system 4120, and displays it on the stage position accuracy information display unit 4304. In addition, the stage information processing component 4211 displays the log information corresponding to the movement axis (X-axis or Y-axis) selected by the movement axis selection button 4303 on the stage position accuracy information display unit 4304.

[0136] On the stage position accuracy information display unit 4304, with the consecutive number of measurement points / check points (MP / IP No.) as the horizontal axis and the deviation amount between the target position and the stop position related to the stage movement as the vertical axis, the tracking information related to past stage movements is displayed. In the stage position accuracy information display unit 4304 of FIG. 43, as an example, the tracking information 4305 for the consecutive numbers 1 to 5 of the measurement points / check points and the tracking information 4306 for the consecutive numbers 50001 to 50005 of the measurement points / check points are displayed. The tracking information for each measurement point / check point is displayed as the deviation amount and the clearance amount. For example, the tracking information for each measurement point / check point 1 is displayed as the deviation amount x1 and the clearance amount W x1 is displayed, and the tracking information for each measurement point / check point 50001 is displayed as the deviation amount x 50001 and the clearance amount W x50001 is displayed. Here, the clearance amount refers to the variation range of the deviation amount between the target position and the stop position related to the stage movement. The clearance amount is the variation range of the deviation amount generated by the mechanical clearance (gap) that the sliding mechanism constituting the stage inevitably has. For example, it can be calculated based on the detection signal from the position detector (such as a laser displacement meter) that detects the position of the X-Y stage 4104 in real time and the measurement time. Alternatively, the clearance amount can be obtained by statistically processing the data related to the detection signal and the measurement time from the position detector during past multiple stage movements (using, for example, standard error, standard deviation, confidence interval of the mean, etc.). In addition, if the sliding surface characteristics of the sliding mechanism constituting the stage change over time, the clearance amount also changes over time. Generally, as the number of stage movements increases, the gap (clearance) between components becomes larger due to wear of the sliding surface, so the clearance amount also tends to be larger compared to the initial value.

[0137] If the user inputs the desired value in the tolerance 4314 of the stage deviation limit setting unit 4308 and presses the Apply button, the stage information processing component 4211 displays the width (bars) 4320, 4321 of the tolerance based on the input value. The value of the tolerance set here is used as the following Figure 24is used as the processing parameter in the step of checking the overlapping area in the scanning overlap test execution process.

[0138] The user performs parameter setting from the log information statistics processing setting unit 4309 and parameter setting from the omen diagnosis setting unit 4413 in the scanning overlap parameter setting unit 4322. Each parameter set in these setting units is used as the processing parameter when the log information statistics processing 4415 or the omen diagnosis 4416 is selected in the stage position accuracy factor setting unit 4409 of the subsequent scanning overlap test screen ( Figure 22 ).

[0139] The user specifies the object range of the measurement point / check point for implementing the statistical processing of the tracking information from the measurement point / check point number setting unit 4311 in the log information statistics processing setting unit 4309, and specifies the method (average value or maximum value) of the statistical processing of the tracking information from the measurement processing setting unit 4310. When it is desired to consider the gap amount in the statistical processing, a check mark is placed in the gap information application check box 4312 for specification.

[0140] The user performs parameter setting for predicting the tracking information at future measurement points / check points based on the previously obtained tracking information in the omen diagnosis setting unit 4313. If the user inputs the desired value (number or range) in the measurement point / check point number setting unit 4316 and presses the Presumption button, the stage information processing component 4211 predicts the tracking information of the measurement points / check points in the object range based on the previously obtained tracking information, and displays the prediction result on the stage position accuracy information display unit 4304. In Figure 21 , as an example, the tracking information 4307 (deviation amount x 90001 , gap amount W x90001 ) predicted at the measurement point / check point number 90001 is displayed.

[0141] If the user presses the Save button 4318 after determining various scanning overlap parameters, the stage information processing component 4211 stores the various scanning overlap parameters in the log information storage medium 113.

[0142] The user performs the above operations and settings on each moving axis (X-axis or Y-axis) of each device via the Figure 21 stage position accuracy analysis screen.

[0143] Figure 22 is an example of the scanning overlap test screen. The functions related to this screen are processed by the scanning overlap test component 4212.

[0144] The user designates an IDS file that describes the measurement sequence and various alignment information via the File menu 4411. In addition, the user designates an IDW file that describes wafer information and chip internal information related to the wafer via the File Load button for the IDW file. Based on these designations, the scan overlap test component 4212 reads out the designated IDS file and IDW file from the log information storage medium 113 or the storage section and storage device within the computer system 4120, and displays them on the recipe information display section 4402.

[0145] The IDS file and IDW file are files created or edited by the user using desired conditions and parameter settings via another setting screen (not shown), and are recorded in the log information storage medium 113 or the storage section and storage device within the computer system 4120. The functions related to the creation and editing of the IDS file and IDW file are processed by the wafer information processing component 4209 and the recipe information processing component 4210.

[0146] On the scan overlap test screen, the scan overlap test component 4212 displays the name 4401 of the IDS file, the name 4403 of the IDW file, the alignment point information 4405, and the measurement point / check point information 4404. The alignment point information 4405 contains information on the alignment points for measurement (alignment chips, internal chip coordinates, alignment conditions, images for automatic detection, etc.), and the measurement point / check point information 4404 contains information on the measurement points (length measurement chips, internal chip coordinates, length measurement conditions, etc.).

[0147] The user sets the device settings 4407 and the stage position accuracy factor setting section 4409 that are the objects of the scan overlap test in the scan overlap setting section 4417.

[0148] The user can select and specify in the device settings 4407 Figure 20 "ALL" for the case of designating all the devices (devices 4100-1 to 4100-3) connected to the system shown and "Selected" for the case of designating specific devices. When "Selected" is selected, the devices to be the objects are designated via the unit buttons.

[0149] When performing the scan overlap test, considering the stage position accuracy factor (conditions, setting parameters) determined from the stage position accuracy analysis screen in Fig. 43, the user selects the log information statistical processing 4415 or the omen diagnosis 4416 in the stage position accuracy factor setting section 4409. When not considering the stage position accuracy factor, "None" 4414 is selected.

[0150] With the above settings, if the Start button 4418 is pressed, the scan overlap test is executed.

[0151] Figure 23 is a flowchart showing a series of processes from the start to the end of the scan overlap test. The user creates and edits IDW files and IDS files (S4501 - S4502). The computer system 4120 analyzes the stage position accuracy of the apparatus 4100 (S4503). The user sets the scan overlap parameters on the Figure 21 screen (S4504). The computer system 4120 reads in the IDS file and the IDW file (S4505). The user designates two or more apparatuses with overlapping test scan positions in the unit menu 4301 (S4506). The user designates each parameter for the overlap of the test scan positions (S4507). When the computer system 4120 uses the same recipe in each apparatus 4100 according to the settings, the computer system 4120 tests in which apparatus 4100 the overlapping part of the scan positions is generated (S4508). The computer system 4120 displays the test results for the user to confirm (S4509). If there is a place where the scan positions overlap (i.e., the beam irradiation is needlessly repeated at the same position), the process returns to S4501 and the same processing is repeated. If not, the computer system 4120 updates the recipes and other related parameters of each apparatus 4100 (S4511).

[0152] Figure 24 is a flowchart showing the detailed steps of S4508. The computer system 4120 reads out the log data of each designated apparatus 4100 (S4601). The computer system 4120 calculates the stage position accuracy parameters in each designated apparatus 4100 (S4602). The computer system 4120 calculates the scan area in each designated apparatus 4100 (S4603). The computer system 4120 checks in which apparatus 4100 the overlapping part of the scan positions is generated (S4604). The computer system 4120 presents the check results in the Figure 25 screen described later (S4605 - S4606).

[0153] Figure 25 is an example of a screen showing the execution result of the scan overlap test. The functions related to this screen are processed by the scan overlap test result processing component 4213.

[0154] In the scan information table 4701, based on the setting information of the IDS file and the IDW file read from the scan overlap test screen, the test results in each alignment point information 4405 and measurement point / inspection point information 4404 are displayed row by row.

[0155] When the user clicks on a row in the scan information table 4701 that specifies the details of the test results they wish to confirm and then presses the Show button 4705, the detailed test results are displayed as thumbnails in the scan image thumbnail display section 4709. In Fig. 47, as the two rows of test results specified for No. 002 and No. 003 in the scan information table 4701, the result of pressing the Show button 4705 is that the test results at these two points (the test results of the scan area) are displayed. Furthermore, when the user specifies an area 4717 in the scan image thumbnail display section 4709 that they wish to magnify and view from the input / output device, the details of this area 4717 are magnified and displayed in the scan image display section 4730.

[0156] The scan image thumbnail display section 4709 can zoom in and out of the screen by operating the zoom bar 4719. In addition, the scan image display section 4720 can zoom in and out of the screen by operating the zoom bar 4719 or by specifying a magnification factor from the magnification setting section 4718.

[0157] The user can perform various viewing operations on the test results via the viewing operation section 4723. When the user clicks on a given row of test results in the scan information table 4701 and then presses the Hide button 4706, the test results corresponding to the specified row become non-displayed in the scan image thumbnail display section 4709 and the scan image display section 4730. If the Jump button 4702 is pressed, the input screen (not shown) for the sequential number 4724 is launched, and the row of test results corresponding to the input sequential number can be specified. If the Bring to Front button 4703 is pressed, the top row of the scan information table 4701 is specified, and if the Send to Back button 4704 is pressed, the last row of the scan information table 4701 is specified. If the Next Overlap button 4707 is pressed, the results of the next overlap area are specified and displayed in the scan information table 4701, the scan image thumbnail display section 4709, and the scan image display section 4730. If the Prev.Overlap button 4708 is pressed, the results of the previous overlap area are specified and displayed in the scan information table 4701, the scan image thumbnail display section 4709, and the scan image display section 4730.

[0158] Next, the content of the test results will be described. The sequential number 4724 of the test results, the scan information 4725 (detailed information of the alignment point or measurement point / check point that is the starting point of the scan), the X-direction magnification factor 4726, and the Y-direction magnification factor 4727 are displayed in the scan information table 4701. In Figure 25In the example, the row of No.002 corresponds to the test result in the alignment point (alignment pattern) 4711, and the row of No.003 corresponds to the test result in the alignment point (alignment pattern) 4714. In addition, the test result of the row of No.002 is displayed as the scan area 4710 and the scan area 4716 on the scan image thumbnail display section 4709, and the test result of the row of No.003 is displayed as the scan area 4713 and the scan area 4724 on the scan image thumbnail display section 4709. In addition, the length measurement points (length measurement patterns) 4712 and 4715 each represent the length measurement points (length measurement patterns) that become the measurement objects after being addressed by the alignment points (alignment patterns) 4711 and 4714.

[0159] The scan areas 4710, 4713, 4716, and 4724 are calculated using the scan overlap parameters of each device saved in the stage position accuracy analysis screen ( Figure 21 ). That is, the scan overlap test result processing component 4213 follows various parameters of the scan overlap parameter setting section 4322 in each device ( Figure 21 ) and the scan overlap setting section 4417 of the scan overlap test screen ( Figure 22 ) to calculate the scan areas 4710, 4713, 4716, and 4724, and displays the calculation result on the scan overlap test execution result screen ( Figure 25 ).

[0160] The scan areas 4710 and 4713 shown by solid lines represent the maximum scan areas in the ideal case where there is no deviation of the X-Y stage 4104 in each device (deviation = 0), and the scan areas 4716 and 4724 shown by dotted lines represent the maximum scan areas in the case where there is a deviation of the X-Y stage 4104 in each device (deviation ≠ 0) and it changes over time. In Figure 25 's example, in the ideal state without deviation (deviation = 0), there is no overlapping area (overlap zone) between the scan area 4710 of No.002 and the scan area 4713 of No.003. On the other hand, in the state where there is a deviation and it changes over time, there is an overlapping area between the scan area 4716 of No.002 and the scan area 4724 of No.003. This indicates that in the devices 4200-1 to 4200-3 of the same model, when measurements are performed using the same measurement recipe (the recipe composed of the IDS file and the IDW file), in the actual state where there is a deviation in each device and it changes over time, an overlapping area will be generated during scanning.

[0161] When the user wishes to change the scan overlap parameter settings, specify the target device for which the parameter settings are to be changed from the stage analysis menu 4722 of the scan overlap test execution result screen. By this specification, the stage position accuracy analysis screen ( Figure 21)Start, which can change the desired parameter settings in the target device. Furthermore, after changing the parameter settings, the scan overlap test can be executed again from the re-inspection menu 4721.

[0162] In the disclosure related to the above-described embodiments, a system, a method, and a non-transitory computer-readable medium storing a program are described as follows: A first inspection device that inspects a plurality of inspection points of a first sample by scanning a first inspection beam records first tracking information of the movement locus of a first movement mechanism that moves the first sample during the inspection of the first sample, and a second inspection device that inspects a plurality of inspection points of a second sample by scanning a second inspection beam records second tracking information of the movement locus of a second movement mechanism that moves the second sample during the inspection of the second sample, and determines or adjusts the inspection beam scanning area of an inspection recipe used in both the first inspection device and the second inspection device.

[0163] According to the present disclosure related to the above-described embodiments, since the parameters of the measurement recipe can be set on the basis of taking into account the deviation amount of the stage position accuracy different for each device and its change over time, a measurement recipe common (the same) to each device in which the electron beam does not repeatedly irradiate (scan) the same area of the sample can be created. That is, in the case where there are multiple CD-SEMs of the same model, the same measurement recipe can be used without change in the same measurement inspection process. In addition, if the created same measurement recipe is used in multiple devices, even in the case where there are slight changes in the working range (movable range) of the stage due to changes over time and variations in the stop position accuracy are different for each device, the effect that the measurement recipe operates without problems for each device is achieved.

[0164] <Regarding a modification example of the present disclosure>

[0165] The present disclosure is not limited to the foregoing embodiments and includes various modification examples. For example, the above-described embodiments have been described in detail for easy understanding of the present invention, but are not necessarily limited to having all the structures described. In addition, a part of the structure of a certain embodiment can be replaced with the structure of another embodiment, and in addition, the structure of another embodiment can be added to the structure of a certain embodiment. In addition, addition, deletion, and replacement of other structures can be performed on a part of the structure of each embodiment.

[0166] For example, the respective data described as learning data in Embodiment 1 and the respective data described as learning data in Embodiment 2 can be used in combination. Or, learning can also be implemented by using only a part of the respective data described as learning data in Embodiments 1 to 2.

[0167] In Embodiment 2, it was described that four learning units were generated, but instead, only one learner may be generated, and in this one learner, clustering is performed for any one of the four classifications corresponding to the first to fourth learning units.

[0168] Description of Reference Numerals

[0169] 10: Measurement system

[0170] 100 to 102: CD-SEM

[0171] 103: Computer system

[0172] 105: Measurement recipe storage medium

[0173] 106: Sample information storage medium

[0174] 113: Log information storage medium

[0175] 114: Design data storage medium

[0176] 1901: Learning model

[0177] 4100: Device

[0178] 4120: Computer system.

Claims

1. A diagnostic system for diagnosing the state of a device for measuring or inspecting a sample, wherein the diagnostic system is characterized in that, the diagnostic system includes: a computer system that infers the cause of an error generated in the device, the computer system includes: a learner that learns, by machine learning, the correspondence between at least any one of a recipe that defines the operation of the device, log data that describes the state of the device, or sample data that describes the characteristics of the sample, and the type of the error, after the learner has performed the machine learning, when an error occurs in the device, the computer system inputs at least any one of the recipe used by the device when the error occurred, the log data at the time of the error occurrence, or the sample data at the time of the error occurrence to the learner, obtains a presumption result of the cause of the error that is attributable to at least any one of the recipe, the state of the device, or the characteristics of the sample as an output of the learner, and outputs the presumption result, the learner further learns the correspondence between at least any one of the amount of change in the recipe that defines the operation of the device, the amount of change in the log data that describes the state of the device, or the amount of change in the sample data that describes the characteristics of the sample, and the amount of change in the inspection accuracy of the device, the computer system inputs at least any one of the new recipe, the new log data, or the new sample data to the learner to obtain a presumption result of the amount of change in the inspection accuracy of the device as an output of the learner, and outputs the presumption result.

2. A diagnostic system for diagnosing the state of a device for measuring or inspecting a sample, wherein the diagnostic system is characterized in that, the diagnostic system includes: a computer system that infers the cause of an error generated in the device, the computer system includes: a learner that learns, by machine learning, the correspondence between at least any one of a recipe that defines the operation of the device, log data that describes the state of the device, or sample data that describes the characteristics of the sample, and the type of the error, after the learner has performed the machine learning, when an error occurs in the device, the computer system inputs at least any one of the recipe used by the device when the error occurred, the log data at the time of the error occurrence, or the sample data at the time of the error occurrence to the learner, obtains a presumption result of the cause of the error that is attributable to at least any one of the recipe, the state of the device, or the characteristics of the sample as an output of the learner, and outputs the presumption result, the learner further learns at least any one of the recipe, the log data, or the sample data when the error has not occurred together with an indication that the error has not occurred, the learner further learns the error generation probability together with the type of the error, The learner further learns the non-occurrence probability of the error together with the meaning indicating that the error does not occur. After the learner implements the machine learning, the computer system inputs at least any one of the recipe, the log data, or the sample data to the learner to obtain the occurrence probability and the non-occurrence probability as the output of the learner. The computer system diagnoses the degree of the omen of the occurrence of the error in accordance with the occurrence probability and the non-occurrence probability, and outputs the diagnosis result.

3. The diagnostic system according to claim 2, wherein the diagnostic system further includes a storage unit that stores at least any one of the change over time of the recipe, the change over time of the log data, or the change over time of the sample data. The learner learns the correspondence relationship between at least any one of the parts of the change over time of the recipe where the change amount over time is equal to or greater than the threshold, the parts of the change over time of the log data where the change amount over time is equal to or greater than the threshold, or the parts of the change over time of the sample data where the change amount over time is equal to or greater than the threshold, and at least any one of the occurrence probability or the non-occurrence probability. The computer system inputs at least any one of the change over time of the recipe, the change over time of the log data, or the change over time of the sample data to the learner to obtain the occurrence probability and the non-occurrence probability as the output of the learner.

4. A diagnostic system for diagnosing the state of a device for measuring or inspecting a sample, wherein the diagnostic system includes a computer system that infers the cause of an error occurring in the device. The computer system includes a learner that learns, through machine learning, the correspondence relationship between at least any one of a recipe that defines the operation of the device, log data that describes the state of the device, or sample data that describes the characteristics of the sample, and the type of the error. After the learner implements the machine learning, when the error occurs in the device, the computer system inputs at least any one of the recipe used by the device when the error occurs, the log data when the error occurs, or the sample data when the error occurs to the learner, obtains a presumption result of the cause of the error due to at least any one of the recipe, the state of the device, or the characteristics of the sample as the output of the learner, and outputs the presumption result. When the presumption result of the cause of the error is a presumption result due to a known cause, the computer system obtains a re-inspection result from the device after correcting at least any one of the recipe or the state of the device. When the re - inspection result is normal, the computer system re - learns at least any one of the recipe used by the device when the re - inspection result is obtained, the log data when the re - inspection result is obtained, or the sample data when the re - inspection result is obtained, together with the meaning indicating that the error has not occurred.

5. The diagnostic system according to claim 4, wherein when the re - inspection result is the error and the presumed result of the cause of the error is not the presumed result due to a known cause, the computer system, on the basis of implementing error repair processing, re - obtains the re - inspection result from the device. When the re - obtained re - inspection result is normal, the computer system causes the learner to perform the re - learning.

6. A diagnostic system for diagnosing the state of a device for measuring or inspecting a sample, characterized in that the diagnostic system includes: a computer system that presumes the cause of an error occurring in the device. The computer system includes: a learner that learns, through machine learning, the correspondence between at least any one of a recipe that defines the operation of the device, log data that describes the state of the device, or sample data that describes the characteristics of the sample, and the type of the error. After the learner performs the machine learning, when an error occurs in the device, the computer system inputs at least any one of the recipe used by the device when the error occurs, the log data when the error occurs, or the sample data when the error occurs to the learner, obtains a presumed result of the cause of the error that is attributable to at least any one of the recipe, the state of the device, or the characteristics of the sample as the output of the learner, and outputs the presumed result. The learner includes: a first learning unit that learns the correspondence between the recipe and the type of the error; a second learning unit that learns the correspondence between the log data and the type of the error; a third learning unit that learns the correspondence between the sample data and the type of the error; and a fourth learning unit that learns the correspondence between the presumed result of the first learning unit, the presumed result of the second learning unit, the presumed result of the third learning unit, and the type of the error. The learner outputs the presumed results of the first learning unit, the second learning unit, the third learning unit, and the fourth learning unit respectively as the output of the learner.

7. A diagnostic system for diagnosing the state of a device for measuring or inspecting a sample, characterized in that the diagnostic system includes: a computer system that presumes the cause of an error occurring in the device. The computer system includes: a learner that learns, through machine learning, the correspondence between at least any one of a recipe that defines the operation of the device, log data that describes the state of the device, or sample data that describes the characteristics of the sample, and the type of the error. After the computer system performs the machine learning on the learner and an error occurs in the device, by inputting at least any one of the recipe used by the device when the error occurs, the log data when the error occurs, or the sample data when the error occurs into the learner, a presumption result regarding at least any one of the reasons for the error attributable to at least any one of the recipe, the state of the device, or the characteristics of the sample is obtained as the output of the learner, and this presumption result is output. The device is an image acquisition device that acquires an image of the sample. The learner learns at least any one of the number of checkpoints on the sample, the coordinates of the checkpoints on the sample, the image of the sample acquired in advance by the device before inspecting the sample, the inspection conditions of the device, and the inspection sequence of the device as the recipe. The learner learns at least any one of the coordinates of the addressing pattern that serves as a reference when determining the checkpoints, the required time for autofocus, the gain value and bias value when performing automatic brightness contrast adjustment, the shape and size of the pattern formed on the sample, the output value of the sensor provided in the device, the voltage value and current value supplied to the components provided in the device, and the control signal for the DA converter as the state of the device. The learner learns at least any one of the manufacturing conditions of the sample, the type of exposure machine used to form the resist pattern when forming the pattern on the sample, the exposure conditions of the exposure machine, the material of the resist pattern, the film thickness of the resist pattern, and the shape and size of the resist pattern as the characteristics of the sample.

8. A diagnostic system for diagnosing the state of a device that measures or inspects a sample, characterized in that the diagnostic system includes: a computer system that presumes the cause of an error occurring in the device. The computer system includes: a learner that clusters a combination of one or more of the processing content data describing the content of the inspection process performed by the device, the inspection conditions of the device, and the attribute data describing the attributes of the device through machine learning. After the learner performs the machine learning and an error occurs in the device, the computer system inputs a first combination of one or more of the processing content data describing the content of the inspection process performed by the device when the error occurs, the inspection conditions when the error occurs, and the attribute data into the learner, and classifies the first combination into any one of the clusters obtained by performing the clustering. The computer system obtains a presumption result regarding whether the error occurs according to the result of the classification and outputs this presumption result. The device is an image acquisition device that acquires an image of the sample. The computer system generates a first learning unit by causing the learner to learn the first combination when no error occurs and the first combination is clustered into any one of the classifications and the image is normal, and the first learning unit clusters the first combination when no error occurs, the content of the inspection process, the inspection conditions, and the attributes are all normal, and the image is normal. When the computer system uses a new first combination in accordance with whether the new first combination is classified into any one of the clustering results of the first learning unit, the computer system obtains a presumption result from the first learning unit as to whether no error occurs, the content of the inspection process, the inspection conditions, and the attributes are all normal, and the image becomes normal.

9. A diagnostic system for diagnosing the state of a device for measuring or inspecting a sample, characterized in that the diagnostic system comprises a computer system that presumes the cause of an error occurring in the device. The computer system comprises a learner that clusters, by machine learning, a combination of one or more of the process content data describing the content of the inspection process performed by the device, the inspection conditions of the device, and the attribute data describing the attributes of the device. After the learner has performed the machine learning, when an error occurs in the device, the computer system classifies a first combination of one or more of the process content data describing the content of the inspection process performed by the device when the error occurred, the inspection conditions when the error occurred, and the attribute data into any one of the clusters obtained by performing the clustering by inputting the first combination to the learner. The computer system obtains a presumption result regarding whether the error has occurred in accordance with the result of the classification and outputs the presumption result. The device is an image acquisition device that acquires an image of the sample. The computer system generates a third learning unit by causing the learner to learn the first combination when no error occurs and the first combination is clustered into any one of the classifications and the image is abnormal, and the third learning unit clusters the first combination when no error occurs, the content of the inspection process, the inspection conditions, and the attributes are all normal, and the image is abnormal. When the computer system uses a new first combination in accordance with whether the new first combination is classified into any one of the clustering results of the third learning unit, the computer system obtains a presumption result from the third learning unit as to whether no error occurs, the content of the inspection process, the inspection conditions, and the attributes are all normal, and the image becomes abnormal.

10. A diagnostic system for diagnosing the state of a device for measuring or inspecting a sample, characterized in that the diagnostic system comprises a computer system that presumes the cause of an error occurring in the device. The computer system includes: a learner that clusters, by machine learning, a combination of one or more of processing content data describing the content of the inspection process performed by the device, the inspection conditions of the device, and attribute data describing the attributes of the device. After the learner performs the machine learning, when an error occurs in the device, the computer system inputs a first combination of one or more of the processing content data describing the content of the inspection process performed by the device when the error occurred, the inspection conditions at the time of the error occurrence, and the attribute data to the learner, and classifies the first combination into any one of the clusters obtained by performing the clustering. The computer system obtains a presumption result regarding whether the error has occurred in accordance with the result of the classification, and outputs the presumption result. When the error has not occurred and the first combination is not clustered into any of the classifications, the computer system causes the device to perform a recovery process specified in advance to avoid the error. The computer system generates a second learning unit by causing the learner to learn the first combination when the error has not occurred, the first combination is not clustered into any of the classifications, and the error is avoided by the recovery process. The second learning unit clusters the first combination when at least any one of the content of the inspection process, the inspection conditions, and the attributes is abnormal when the error has not occurred, and the content of the inspection process, the inspection conditions, and the attributes used in the recovery process are normal. When the computer system uses the new first combination in accordance with whether the new first combination is classified into any of the clustering results of the second learning unit, the computer system obtains, from the second learning unit, a presumption result that at least any one of the content of the inspection process, the inspection conditions, and the attributes is abnormal when the error has not occurred, and the content of the inspection process, the inspection conditions, and the attributes used in the recovery process become normal.

11. A diagnostic system for diagnosing the state of a device for measuring or inspecting a sample, wherein the diagnostic system is characterized in that the diagnostic system includes: a computer system that presumes the cause of an error occurring in the device. The computer system includes: a learner that clusters, by machine learning, a combination of one or more of processing content data describing the content of the inspection process performed by the device, the inspection conditions of the device, and attribute data describing the attributes of the device. After the learner performs the machine learning, when an error occurs in the device, the computer system inputs a first combination of one or more of the processing content data describing the content of the inspection process performed by the device when the error occurred, the inspection conditions at the time of the error occurrence, and the attribute data to the learner, and classifies the first combination into any one of the clusters obtained by performing the clustering. The computer system obtains a presumption result regarding whether the error has occurred in accordance with the result of the classification, and outputs the presumption result. In the case where the error has not occurred and the first combination does not cluster into any of the classifications, the computer system causes the device to perform a recovery process that has been predefined to avoid the error. The computer system generates a fourth learning unit by causing the learning unit to learn the first combination when the error has not occurred, the first combination does not cluster into any of the classifications, and the error is not avoided by the recovery process. The fourth learning unit clusters the first combination in the case where at least any one of the content of the inspection process, the inspection conditions, and the attributes is abnormal when the error has not occurred and at least any one of the content of the inspection process, the inspection conditions, and the attributes used in the recovery process is abnormal. When the computer system uses a new first combination in accordance with whether the new first combination is classified into any of the clustering results of the fourth learning unit, the computer system obtains from the fourth learning unit a presumption result as to whether at least any one of the content of the inspection process, the inspection conditions, and the attributes is abnormal and at least any one of the content of the inspection process, the inspection conditions, and the attributes used in the recovery process becomes abnormal when the error has not occurred.

12. A diagnostic system for diagnosing the state of a device that measures or inspects a sample, the diagnostic system being characterized in that the diagnostic system includes: a computer system that presumes the cause of an error occurring in the device, the computer system includes: a learning unit that clusters, by machine learning, a combination of one or more of processing content data that describes the content of an inspection process performed by the device, the inspection conditions of the device, and attribute data that describes the attributes of the device, After the learning unit has performed the machine learning, in the case where an error has occurred in the device, the computer system classifies the first combination, which is one or more of the processing content data that describes the content of the inspection process performed by the device when the error occurred, the inspection conditions when the error occurred, and the attribute data, into any of the clusters obtained by performing the clustering by inputting the first combination to the learning unit. The computer system obtains a presumption result regarding whether the error has occurred in accordance with the result of the classification, and outputs the presumption result. The device is an image acquisition device that acquires an image of the sample. The learning unit learns at least any one of alignment for aligning the coordinate system of the stage on which the sample is placed and the coordinate system of the device, addressing for moving an inspection position to an inspection point on the sample, adjustment of an autofocus mechanism, automatic astigmatism correction, and automatic brightness contrast adjustment as the content of the inspection process. The learner learns at least any one of the field of view size obtained when learning the addressing, the number of images obtained during the adjustment of the autofocus mechanism, the frame accumulation number of the images, the distance between the shape pattern on the sample and the inspection point used during the adjustment of the autofocus mechanism, the relative direction between the shape pattern on the sample and the inspection point used during the adjustment of the autofocus mechanism, and the lens conditions as the inspection conditions. The learner learns at least any one of the identifier of the device and the characteristics of the environment where the device is set as the attribute of the device.

Images