Process information presentation system, process error estimation system

Abstract

An object of the present invention is to provide a system capable of estimating a cause of a process error and prompting a correction candidate for the process error. A process information prompting system or a process error estimation system of the present invention causes a learner to learn a correspondence between a process and an error caused by the process, and acquires an estimation result for whether an error occurs when a new process is used from the learner (see FIG. 1).

Description

Technical Field

[0001] This invention relates to a system for providing corrective information on the manufacturing process of a semiconductor inspection device, and a system for predicting process errors. Background Technology

[0002] Semiconductor measurement and inspection devices perform measurement and inspection actions according to set parameters called process parameters. These process parameters include multiple items, which engineers typically optimize manually based on the properties of the object being measured / inspected and the characteristics of the device. However, if the characteristics of the device change over time, the process parameters may not match the actual object being measured / inspected, potentially leading to errors in the measurement and inspection actions. These errors, caused by the process parameters themselves, are called process errors.

[0003] Previously, to reduce process errors, service engineers typically entered the semiconductor factory and manually downloaded internal data from semiconductor measurement and inspection devices. They then graphically represented this data to estimate the process items requiring correction. This correction process relied heavily on the engineer's experience, thus depending heavily on their skills.

[0004] Patent Document 1 described below discloses a technique for correcting the causes of errors in a manufacturing process. In this document, the cause of a process error is deduced based on the process having the largest standardized number of errors in the error log (see claim 1 of this document).

[0005] Patent Document 2 below describes a technique for predicting process inspection time by having a neural network learn the relationship between process conditions and operation time (see claim 1 of that document).

[0006] Existing technical documents

[0007] Patent documents

[0008] Patent Document 1: Japanese Patent No. 4398441

[0009] Patent Document 2: Japanese Patent No. 5933985 Summary of the Invention

[0010] The problem that the invention aims to solve

[0011] With the miniaturization and diversification of semiconductors, the complexity and number of processes in semiconductor measurement and inspection equipment have increased. Consequently, the reduced equipment uptime due to process errors has become a challenge. Furthermore, the shortage of engineers caused by the rapid opening of semiconductor factories, particularly in China, has made the desire for more efficient process correction even more urgent. However, the resolution time and robustness of process correction are significantly affected by the skills of the engineers.

[0012] In the technology described in Patent Document 1, the cause of the error is inferred based on the error log, but it is deemed unnecessary to specifically consider how to correct the process. This is because the document does not utilize information about the normal process where no error occurred, and therefore, it is impossible to compare the time when the error occurred with the normal time.

[0013] In the technology described in Patent Document 2, a neural network learns the relationship between the process and the operation time. Therefore, it may be possible to determine the process in the event of an abnormal operation time. However, it does not indicate whether the estimated operation time is incorrect. Therefore, it is considered difficult to deduce the cause of the process error using the technology described in this document.

[0014] The present invention was made in view of the aforementioned problems, and its object is to provide a system capable of deduce the cause of a process error and suggest corrective measures for the process error.

[0015] Methods for solving problems

[0016] The process information prompting system or process error estimation system of the present invention enables a learner to learn the correspondence between a process and an error caused by that process, and obtains from the learner an estimation result regarding whether an error has occurred when a new process is used.

[0017] Invention Effects

[0018] According to the process information prompting system or process error estimation system of the present invention, the cause of the process error can be estimated and the correction candidates for the process error can be suggested. Attached Figure Description

[0019] Figure 1 This is a structural diagram of the process information prompting system and process error estimation system of Implementation Method 1.

[0020] Figure 2 This is a flowchart illustrating the process of inspecting semiconductor wafers using the semiconductor inspection apparatus 11.

[0021] Figure 3A This is a diagram illustrating the contents of three data points obtained by the computer system 100 from the semiconductor inspection device 11.

[0022] Figure 3BIt is a diagram that schematically represents the correspondence between the input and output of the learner 121.

[0023] Figure 4 This is a flowchart illustrating the process of generating learner 121 and parsing it.

[0024] Figure 5 This diagram schematically illustrates the process by which the learner 121 infers whether an error has occurred when the semiconductor inspection device 11 uses a new process 112.

[0025] Figure 6A This is a schematic diagram showing the result of the parser 122 sorting the process parameters.

[0026] Figure 6B This diagram schematically illustrates the process by which the correction calculation unit 123 calculates the correction amount for the process parameters.

[0027] Figure 7A This is an example of a GUI provided to operators by computer system 100.

[0028] Figure 7B This is an example of a GUI provided to operators by computer system 100.

[0029] Figure 7C This is an example of a GUI provided to operators by computer system 100.

[0030] Figure 8 This is an explanation of the action flow for the Machine Learning Department 120 to implement relearning.

[0031] Figure 9 This is a structural diagram of the process information prompting system and process error estimation system of implementation method 4. Detailed Implementation

[0032] In the embodiments described below, "semiconductor inspection apparatus" refers to an apparatus for measuring the size of a pattern formed on a semiconductor wafer, an apparatus for inspecting whether there are defects in a pattern formed on a semiconductor wafer, or an apparatus for inspecting whether there are defects in a bare wafer without a pattern, and also includes composite apparatuses that combine these multiple apparatuses.

[0033] Additionally, in the embodiments described below, "inspection" is used in the sense of measuring or checking, "inspection action" is used in the sense of measuring or checking, and "inspection object" refers to the wafer that is the object of measurement or inspection, or the area of ​​the wafer that is the object of measurement or inspection.

[0034] <Implementation Method 1>

[0035] Figure 1 This is a structural diagram of the process information prompting system and process error estimation system according to Embodiment 1 of the present invention. This system is used to estimate the cause of an error in a process for which an inspection operation is performed by a semiconductor inspection apparatus, and to prompt corrective information for that process (hereinafter referred to as the "process information prompting system" or "process error estimation system"). Here, the system based on semiconductor inspection apparatuses 11 to 13 is illustrated. Since each inspection apparatus has the same structure, in the following description, semiconductor inspection apparatus 11 will be used without distinguishing between the various inspection apparatuses.

[0036] Semiconductor inspection apparatus 11 can be configured, for example, by a scanning electron microscope (SEM), an optical inspection apparatus, etc. Hereinafter, as an example, an SEM is used to measure the dimensions of patterns formed on a semiconductor wafer. Semiconductor inspection apparatuses 11 to 13 are sometimes the same apparatus, and sometimes different models or other devices.

[0037] The process error estimation system 1 includes a computer system 100. The computer system 100 includes a database 110 and a machine learning unit 120. The machine learning unit 120 also includes a learner 121 and a parser 122 (and a correction calculation unit 123 described later). The database 110 can be configured as a storage device for storing data. The machine learning unit 120 can be configured as hardware such as circuitry with its functions (i.e., the functions of the learner 121 and parser 122 described later, and the functions of the correction calculation unit 123 described later), or it can be configured by a computing device executing software with those functions installed.

[0038] Computer system 100 acquires the three data sets (described later) from semiconductor inspection devices 11-13 and stores them in database 110. Learner 121 uses this data as supervised data to perform machine learning. Details of the learning process will be described later. The operation of parser 122 will also be described later.

[0039] Figure 2This is a flowchart illustrating the process of inspecting a semiconductor wafer using a semiconductor inspection apparatus 11. The semiconductor inspection apparatus 11 begins this process by loading the semiconductor wafer to be inspected. To perform the inspection, the coordinate system on the semiconductor wafer needs to be aligned with the coordinate system inside the semiconductor inspection apparatus 11. This step is called alignment (S201). Since the field of view of the semiconductor inspection apparatus 11 only covers a portion of the semiconductor wafer surface, it is necessary to align the field of view of the semiconductor inspection apparatus 11 with the area to be inspected. This step is called addressing (S202). The semiconductor inspection apparatus 11 measures the length, etc., of the semiconductor pattern formed within the field of view (S203). S202 to S203 are repeated until measurements are completed for all measurement points (S204). Upon completion of the measurements, the semiconductor wafer is unloaded.

[0040] Figure 3A This is a diagram illustrating the contents of three data points obtained by the computer system 100 from the semiconductor inspection apparatus 11. The computer system 100 obtains apparatus data 111, measurement process (hereinafter sometimes simply referred to as "process") 112, and measurement result 113 from the semiconductor inspection apparatus 11.

[0041] Device data 111 includes (a) device inherent parameters, (b) device error correction data, and (c) observation condition parameters. Device inherent parameters are correction parameters used to ensure that the semiconductor inspection apparatus 11 operates according to specified specifications. Since there may be cases where the actual operation of the apparatus differs from the specified specifications, device inherent parameters are used to ensure that the actual operation matches the specified specifications. Device error correction data are parameters used to correct for errors between semiconductor inspection apparatuses 11 to 13. Observation condition parameters are, for example, parameters specifying the observation conditions of SEM, such as the accelerating voltage of the electron optical system.

[0042] Process 112 includes (a) a wafer pattern, (b) alignment parameters, (c) addressing parameters, and (d) length measurement parameters as process parameters. The wafer pattern is a coordinate diagram (e.g., the coordinates of a pattern) on a semiconductor wafer. Alignment parameters are parameters used to implement S201, such as parameters used to correct the deviation between the coordinate system on the semiconductor wafer and the coordinate system inside the semiconductor inspection apparatus 11. Addressing parameters are parameters used to implement S202, such as parameters that determine information (feature quantities, etc.) about characteristic patterns existing within the inspection target area in the pattern formed on the semiconductor wafer. Length measurement parameters are parameters that describe the conditions for measuring length, such as parameters that specify which part of the pattern's length to measure.

[0043] Measurement result 113 includes (a) length measurement result, (b) image data, (c) error parameters, and (d) action log. The length measurement result describes the measurement of the length of the pattern on the semiconductor wafer. The image data is an observed image of the semiconductor wafer. The error parameters are parameters that describe the error content in the event of an error occurring in any of S201 to S203. The action log records the implementation... Figure 2 The data includes the internal state of the semiconductor inspection device 11 at each step. For example, the operating voltage of each component and the coordinates of the observation field are listed.

[0044] Figure 3B This is a diagram that schematically represents the correspondence between the input and output of learner 121. Learner 121 uses device data 111, measurement process 112, and measurement result 113 as supervised data to perform machine learning, generating learner 121 with device data 111 and measurement process 112 as input and measurement result (error information) 113 as output.

[0045] After the learner 121 completes machine learning, when a new device data 502 and a new measurement process 501 are input to the learner 121, the learner 121 uses the device data 502 and the measurement process 501 to output a measurement result 113 as whether an error occurred when the semiconductor inspection device 11 corresponding to the device data 502 performed the inspection in the measurement process 501. That is, it estimates whether an error occurred. Furthermore, as described later, the machine learning unit 120 can prompt the learner 121, the parser 122, and the correction calculation unit 123 with the correction amount for the process parameters that caused the error, and the corrected process.

[0046] Figure 4 This is a flowchart illustrating the generation of learner 121 and the parsing process of the parser. The following is a description of... Figure 4 The steps are explained below.

[0047] ( Figure 4 Step S401)

[0048] Computer system 100 acquires device data 111, process data 112, and measurement results 113 from semiconductor inspection devices 11-13, respectively, and stores them in database 110. Computer system 100 accumulates this data over a certain period of time (e.g., several weeks, several months, etc.).

[0049] ( Figure 4 Step S402)

[0050] Computer system 100 uses the data accumulated in database 110 as supervised data to perform machine learning, thereby generating learner 121. The correspondence between the inputs to learner 121 and the outputs from learner 121 in this machine learning step is as follows: Figure 3B As shown.

[0051] ( Figure 4 Step S403)

[0052] After the learner 121 completes its machine learning, the parser 122 uses the results learned by the learner 121, or the learned results and process 112, to analyze the impact of each process parameter on the learner's prediction results, thereby ranking the process parameters. For example, ranking can be performed by calculating the SHAP value of each process parameter. Ranking can also be performed using evaluation values ​​other than SHAP values. An example of using SHAP values ​​is explained below.

[0053] ( Figure 4 Step S403: Supplement)

[0054] The SHAP value is a numerical representation of the degree of influence (i.e., contribution to the target variable) that each feature has on the target variable. Process 112 includes multiple process parameters ( Figure 3A The process parameters described herein can be treated as characteristic quantities, and their SHAP values ​​can be calculated. The process parameters are ranked according to their contribution, from highest to lowest SHAP value. As target variables, the probability of error generation (or whether an error occurs) can be used. Examples of ranking and how to use the ranking results will be described later.

[0055] Figure 5 This diagram schematically illustrates the process by which a learner 121 infers whether an error has occurred and indicates the process parameters to be corrected and the amount of correction when a new process 501 is used in a specified semiconductor inspection apparatus. It assumes that the learner 121 completes the process... Figure 4 The machine learning described in the text. For simplicity, the new process 112 has three process parameters A to C. The following is a description of... Figure 5 The steps are explained below.

[0056] The learner 121 is input with a new process 501 and device data 502 of a specified semiconductor inspection device. The learner 121 uses the learning results to determine whether the semiconductor inspection device corresponding to the device data 502 has generated an error when using the new process 501. Here, we assume that the learner 121 determines that an error has occurred, and the following explanation continues.

[0057] Parser 122 sorts the process parameters by calculating the contribution (e.g., SHAP value) of the estimated results for each process parameter within the new process 501. An example of the sorting will be described later.

[0058] The machine learning unit 120 includes a correction calculation unit 123. The correction calculation unit 123 corrects each process parameter within the new process 501 in a way that does not produce errors. That is, it calculates the correction amount for each process parameter to achieve an error-free value. It is not necessary to calculate the correction amount for all process parameters; for example, the correction amount can be calculated sequentially from highest to lowest contribution. A specific example of the correction amount calculation process will be described later.

[0059] The learner 121 re-evaluates whether an error has occurred due to the process parameters reflecting the correction amount calculated by the correction amount calculation unit 123. If an error has occurred, the above process is repeated. If no error has occurred, the process parameters reflecting the correction amount are output from the learner 121 as corrected process parameters. At this time, the sorting result of the parser 122 may also be output. Figure 5 The example shown is where process parameter A is ranked highest.

[0060] The corrected process parameters can be reflected in the new process 501 either manually by the operator or automatically by the machine learning unit 120. Additionally, as described later, an interface for displaying the correction results to the operator can be provided on the computer system 100. The machine learning unit 120 (or correction calculation unit 123) functions as a "process correction suggestion unit" that recommends process corrections.

[0061] Figure 6A This is a schematic diagram showing the result after the parser 122 sorts the process parameters. The parser 122 can calculate the contribution of each process parameter to the estimation result of the learner 121 based on the SHAP value. The contribution can be regarded as high to low according to the order of SHAP values. Here, an example is shown where process parameter A has the highest priority and process parameters BC have the same priority.

[0062] Figure 6B This diagram schematically illustrates the process by which the correction calculation unit 123 calculates the correction amount for process parameters. The parser 122 calculates statistical values ​​for normal process parameters that have not previously produced errors. For example, it can calculate the distribution representing the relationship between the values ​​and frequencies of normal process parameters as statistical values. Furthermore, it can also use the upper and lower limits, average values, and median values ​​of past normal process parameters as statistical values. Figure 6B From the perspective of visual ease of understanding, an example of frequency distribution is shown.

[0063] The correction calculation unit 123 obtains the new process parameter A within process 501. Figure 6B The example shown illustrates a new process parameter A with a value that slightly deviates from the frequency distribution of the past normal process parameter A. The correction calculation unit 123 can calculate the correction amount by reducing the difference between the new process parameter A and past statistical values. For example, the difference between the new process parameter A and the average value of the past normal process parameter A can be used as the correction amount. When other statistical measures are used, the difference between the new process parameter A and the past statistical measures can also be reduced using other appropriate methods. Furthermore, process parameter correction can be performed simultaneously on multiple parameters for error generation determination, or parameters can be corrected one by one for error generation determination sequentially.

[0064] The correction calculation unit 123 also calculates the correction amount for process parameter BC. If the new process parameter does not deviate significantly from past statistical values, correction may not be necessary. Figure 6B In the example, the new process parameter B converges within the past frequency distribution, so no correction is necessary. Alternatively, correction can be made by further reducing the difference between the parameter and the statistical value.

[0065] Figure 7A This is an example of a GUI (Graphical User Interface) provided by the computer system 100 to the operator. Here, an example of an error display screen 701 is shown. Error display screen 701 displays the content of an error generated in the semiconductor inspection device 11. For example, it can display the content of the measurement result 113.

[0066] Figure 7B This is an example of a GUI provided by the computer system 100 to the operator. Here, an example of a process display screen 702 is shown. The process display screen 702 is a screen that displays the contents of process 112. In addition, the contents of device data 111 can also be displayed at the same time.

[0067] Figure 7C This is an example of a GUI provided by computer system 100 to the operator. Here, an example of a process correction display screen 703 is shown. The process correction display screen 703 is used to... Figure 5 The revised process 501 described in the text provides a visual confirmation of whether errors have been reduced. The revised process parameters can also be displayed simultaneously. Operators can confirm the effectiveness of the revised process via the revised process display screen 703 and decide whether to reflect the revision results, etc.

[0068] therefore, Figures 7A to 7CThe GUI shown can display information such as the following: (a) Displaying the error details in the semiconductor inspection device along with the error status of the time series. (b) Displaying the details of the process used by the semiconductor inspection device along with the execution results of that process. (c) Prompting for process correction and displaying the effect of the correction. (d) Displaying process or process-related information.

[0069] <Implementation Method 1: Summary>

[0070] In the process error estimation system of this embodiment 1, the learner 121 learns the correspondence between device data 111 / process 112 / measurement result 113 and estimates whether an error occurs when the semiconductor inspection device 11 uses a new process 501. Thus, the operator can determine whether to adopt the new process 501 without relying on personal judgment.

[0071] In the process error estimation system of this embodiment 1, the correction calculation unit 123 calculates the correction amount for the new process parameters based on the difference between the process parameters in the past normal process and the process parameters in the new process. Therefore, the operator can determine the process parameters that are the cause of the error without relying on personal judgment. Furthermore, it eliminates the need for repeated corrections and re-checks of process parameters, thus making process correction operations more efficient.

[0072] In the process error estimation system of this embodiment 1, the parser 122 sorts the process parameters in the new process according to their contribution to the estimation result. As a result, the operator can reflect the correction amount sequentially from the process parameters with the highest priority, thus making the process correction operation more efficient.

[0073] The process error estimation system of this embodiment 1, when calculating the correction amount of a new process parameter, can also estimate the process parameter that is the cause of the error by identifying the process parameter that is a candidate for correction. Therefore, the cause of the error can be determined quickly.

[0074] <Implementation Method 2>

[0075] There are three types of errors that can occur in a semiconductor inspection apparatus: (a) apparatus-related errors caused by the state of the semiconductor inspection apparatus; (b) wafer (or process)-related errors caused by the state of the semiconductor wafer; and (c) process-related errors caused by the values ​​of process parameters. To correct apparatus-related errors, the apparatus needs to be repaired; therefore, process 112 is not corrected. Wafer-related errors can sometimes be addressed through process correction. Therefore, in Embodiment 1, by learning the correspondence between apparatus data 111, process 112, and measurement result 113, errors that can be corrected based on process parameters are determined.

[0076] On the other hand, it is considered useful for error correction to speculate on which of the above (a) to (c) the error originated from. Therefore, in Embodiment 2 of the present invention, a method for speculating on the category of error cause will be described. The structure of the process error estimation system is the same as in Embodiment 1.

[0077] (Example 1 of the categories for presumed error causes)

[0078] When the computer system 100 uses the same process 112 in each of the semiconductor inspection devices 11-13, and an error occurs only in one specific semiconductor inspection device, it can be presumed that the error originated from the state of that specific semiconductor inspection device. Even if the process 112 is not strictly the same, when the semiconductor inspection devices use processes 112 with similar content (e.g., close distance between feature vectors), if an error occurs only in a specific semiconductor inspection device, it can be presumed that the error originated from that specific semiconductor inspection device. This presumption can be implemented by (a) the computing device of the computer system 100 according to the above presumption rules (as in Examples 2 and 3 described below), or by the learner 121 learning the error parameters that occur only in a specific semiconductor inspection device using the same process 112. In the latter case, the learner 121 presumes whether the error is caused by a device based on the difference between the device data 111 of each semiconductor inspection device.

[0079] (Example 2 of the categories for presumed error causes)

[0080] Computer system 100 acquires measurement results 113 from a semiconductor inspection apparatus using a specific process 112 along a time series. If an error occurs only during a specific period, it can infer that the error originated from the state of the semiconductor wafer. Similarly, even if the processes 112 are not strictly identical, the same applies if an error occurs only during a specific period in the time series history of processes 112 with similar content. This inference can be implemented, for example, by a learner 121 learning the periods during which errors occur using the same process 112.

[0081] (Example 3 of the categories for presumed error causes)

[0082] When the computer system 100 uses the same process 112 in all semiconductor inspection devices 11-13, it can presume that the error originated from process 112 if errors occur in all semiconductor inspection devices. Similarly, this applies even if the process 112 is not strictly identical, but uses a process 112 with similar content, and errors occur in all semiconductor inspection devices. Furthermore, this also applies even if errors do not occur in all semiconductor inspection devices, but errors occur in more than half of the semiconductor inspection devices (e.g., above a certain threshold). This presumption can be implemented, for example, by learning, through the learner 121, whether errors have occurred in multiple semiconductor inspection devices using the same process 112.

[0083] <Implementation Method 3>

[0084] The learning model obtained by learner 121 corresponds to the supervision data used in creating the learning model. Therefore, if the content of the supervision data changes, the content of the learning model may also change. For example, if a semiconductor inspection device is used for a certain period of time, the device state and the semiconductor wafer state may change significantly from the start of use, thereby causing the learning model to become outdated. Therefore, in Embodiment 3 of the present invention, an example of relearning using new supervision data after learner 121 has completed learning will be described. The structure of the process error estimation system is the same as in Embodiment 1.

[0085] Figure 8 This is a flowchart illustrating the process of relearning implemented by the Machine Learning Unit 120. This flowchart can be implemented either each time new supervised data (additional data) is acquired, or at a point in time when a certain amount of additional data has accumulated in the database 110. In this flowchart, the Machine Learning Unit 120 acts as the "relearning implementation unit."

[0086] Computer system 100 stores the additional data in a database (S801). Machine learning unit 120 causes learner 121 to learn from the additional data again, thereby generating a learning model (S802). Machine learning unit 120 determines whether the newly generated learning model should be treated as a different model from the existing learning model based on whether the difference between the newly generated learning model and the existing learning model is above a threshold (S803). For example, by comparing the difference in distances in the feature space of the parameters described by the learning model with the threshold, it can be determined whether it is another model. If it is another model, machine learning unit 120 replaces the existing learning model with the learning model generated in S802 (S804). If it is not necessary to treat it as another model, this flowchart ends.

[0087] <Implementation Method 4>

[0088] Figure 9 This is a structural diagram of the process error estimation system according to Embodiment 4 of the present invention. In this Embodiment 4, the process error estimation system estimates the cause of errors in the semiconductor inspection devices of multiple manufacturing sites (Fabric or Fab). Here, sites 910 (Fab1) and 920 (Fab2) are shown as examples. Each site is connected to the computer system 100 via a network. The structure of the computer system 100 is the same as in Embodiments 1 to 3.

[0089] Location 910 has two production lines 911 and 912. Each production line is connected to a different type (Type A, B, C) of semiconductor inspection equipment. Location 920 can also have the same structure. Computer system 100 is connected to the semiconductor inspection equipment of each production line at each location, and obtains equipment data 111 / process 112 / measurement results 113 respectively. Subsequent processing is the same as in embodiments 1 to 3.

[0090] Computer system 100 can generate different learning models by location or by production line, or it can generate a single learning model that integrates any one or all of these learning models. In the former case, it can generate a learning model corresponding to the characteristics of each location or each production line. In the latter case, for example, when using similar inspection devices at each location or production line, it can improve estimation accuracy by increasing the amount of supervisory data.

[0091] <Regarding variations of the present invention>

[0092] This invention is not limited to the described embodiments and includes various modifications. For example, the embodiments described above are those that have been explained in detail for the purpose of easily understanding the invention and are not limited to having all the described structures. Furthermore, a portion of the structure of one embodiment can be replaced with the structure of another embodiment, and it is also possible to add structures of other embodiments to the structure of one embodiment. Additionally, for a portion of the structure of each embodiment, other structures can be added, deleted, or replaced.

[0093] In the above embodiments, an example is shown in which the learner 121 learns the correspondence between (a) device data, (b) process, and (c) measurement results. However, in order to determine process errors, it is sufficient to at least learn the correspondence between the process and the error parameters. Therefore, such a structure is also the subject of this invention.

[0094] In the above embodiments, an example of a process error in a semiconductor inspection device has been described. However, by having the learner 121 learn the parameters of the operation of the specified device and whether an error occurs when using those parameters, the present invention can also be applied to other devices.

[0095] Symbol Explanation

[0096] 11-13: Semiconductor inspection device

[0097] 100: Computer System

[0098] 110: Database

[0099] 120: Machine Learning Department

[0100] 121: Learning Device

[0101] 122: Parser

[0102] 123: Correction Calculation Department.

Claims

1. A process information presentation system that presents a process that specifies an inspection action performed by a semiconductor inspection device or information related to the process, characterized by comprising: a computer system that estimates the process that is a cause of an error generated in the semiconductor inspection device, the computer system having a learner that learns, by machine learning, a correspondence between the process used when the inspection action is performed by the semiconductor inspection device and the error generated due to the process, the computer system, after the machine learning by the learner is completed, feeding a new process to the learner, thereby obtaining, as an output of the learner, an estimation result for whether the error occurs when the new process is used in the semiconductor inspection device, the computer system, in a case where the estimation result that the error is generated by the new process is obtained from the learner, repeatedly performing correction of the new process and re-feeding to the learner until an estimation result that the error is not generated is obtained, the computer system presenting, as the process or the information related to the process, the process or the information obtained by the repetition as a process or information to be used by the semiconductor inspection device, the process describing one or more process parameters that specify the inspection action, the computer system having a process correction suggestion section that presents a correction suggestion for the process that generates the error by performing the repetition, the process correction suggestion section presenting, as the correction suggestion, the process parameter to be used by the semiconductor inspection device as a correction candidate parameter, and presenting a correction amount of the correction candidate parameter, the learner further having an analyzer that calculates a contribution degree of each of the process parameters to the error, the process correction suggestion section calculating the correction amount in descending order of the contribution degree for some or all of the process parameters.

2. The process information presentation system according to claim 1, characterized in that the process describes one or more process parameters that specify the inspection action, the learner learns a correspondence between a combination of the process parameters used by the semiconductor inspection device when the error is generated and an error parameter that describes a content of the error, and the computer system obtains, as the estimation result, the error parameter corresponding to the process parameter by feeding the process parameter described by the new process to the learner.

3. The process information presentation system according to claim 1, characterized in that the computer system further has a correction amount calculation section that calculates the correction amount, the correction amount calculation section obtains a value distribution or a statistical value of a normal process parameter in a past normal process in which the error is not generated, and the correction amount calculation section calculates the correction amount in accordance with a difference between the correction candidate parameter and the value distribution or the statistical value of the normal process parameter corresponding to the correction candidate parameter in the normal process.

4. The process information presentation system according to claim 1, characterized in that ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ The computer system acquires device data that describes a state parameter indicating a state of the semiconductor inspection apparatus, The learner learns a correspondence between the process, the state parameter, and an error parameter that describes a content of the error, by the machine learning, After the learner completes the machine learning, the computer system inputs the state parameter together with the new process to the learner, whereby a result of estimation of whether the error is generated when the semiconductor inspection apparatus uses the new process and the semiconductor inspection apparatus is in a state indicated by the state parameter is acquired as an output of the learner.

5. The process information prompting system according to claim 4, wherein The semiconductor inspection apparatus includes a first inspection apparatus and a second inspection apparatus, The computer system acquires a first device parameter indicating a state of the first inspection apparatus and a second device parameter indicating a state of the second inspection apparatus as the state parameter, The learner learns a correspondence between the process, the first device parameter, the second device parameter, and the error, by the machine learning, The learner estimates whether the error is generated due to at least either the first device parameter or the second device parameter, in accordance with a difference between the first device parameter and the second device parameter.

6. The process information prompting system according to claim 1, wherein The learner determines a time point at which the error is generated and a time point at which the error is not generated by using the same process or similar process groups, respectively, in a time series history of the correspondence learned by the machine learning, whereby whether the error is generated due to a state of an inspection object of the semiconductor inspection apparatus is estimated.

7. The process information prompting system according to claim 1, wherein The semiconductor inspection apparatus includes a first inspection apparatus and a second inspection apparatus, The learner estimates whether the error is generated due to the process, in accordance with whether the error is generated in the first inspection apparatus and the second inspection apparatus by using the same process or similar process groups, respectively.

8. The process information prompting system according to claim 1, wherein The computer system further has a re-learning implementation unit that causes the learner to re-implement the machine learning after the machine learning is completed, The re-learning implementation unit acquires a new correspondence between a new process and a new error after the machine learning is completed, inputs the new correspondence to the learner, whereby the learner is caused to re-implement the machine learning, The re-learning implementation unit replaces a re-learning data model that describes a result of the re-implementation to a learning data model that describes a result of the machine learning, in a case where a difference between the learning data model and the re-learning data model is above a determination threshold.

9. The process information prompting system according to claim 1, wherein The computer system also has a user interface that presents parameters representing a state of the semiconductor inspection device when the error occurred in the semiconductor inspection device together with the content of the error.

10. The process information presentation system according to claim 4, wherein The device data describes at least any one of the following parameters as the state parameter: a device-specific parameter that describes an operation correction parameter for causing the semiconductor inspection device to operate in accordance with a prescribed specification; a device machine difference correction parameter that describes an operation correction parameter for correcting a machine difference of the semiconductor inspection device; and an observation condition parameter that describes an observation condition of the semiconductor inspection device.

11. The process information presentation system according to claim 2, wherein The process describes at least any one of the following as a parameter that prescribes the inspection operation: a coordinate map on a semiconductor wafer for the semiconductor inspection device to perform inspection; an alignment parameter for aligning a coordinate system of the semiconductor wafer with a coordinate system of the semiconductor inspection device; an addressing parameter for aligning a position at which the semiconductor inspection device performs inspection with an inspection position on the semiconductor wafer; and a length measurement parameter for specifying a portion of a pattern formed on the semiconductor wafer at which a length should be measured at the inspection position.

12. The process information presentation system according to claim 2, wherein The error parameter describes the content of the error and at least any one of the following: a length measurement result that describes a result of determining a length of a pattern formed on a semiconductor wafer inspected by the semiconductor inspection device; an observation image that is an image obtained by the semiconductor inspection device observing the semiconductor wafer; and an operation log of the semiconductor inspection device.

13. The process information presentation system according to claim 1, wherein The computer system also has a user interface that presents the content of the error in the semiconductor inspection device together with an error condition of a time series of the error.

14. The process information presentation system according to claim 1, wherein The computer system also has a user interface that presents the content of the process in the semiconductor inspection device together with an execution result when the process is executed.

15. The process information presentation system according to claim 1, wherein The computer system also has a user interface that presents a result of a correction of the process in the semiconductor inspection device, thereby enabling an effect of the corrected process to be confirmed.

16. The process information presentation system according to claim 9, wherein The computer system also presents the process or information related to the process to the user interface.

17. A process error presumption system that presumes a process that is a cause of an error generated in a semiconductor inspection device, in a process that prescribes an inspection operation performed by the semiconductor inspection device, characterized by ​ The process error presumption system has a computer system that presumes the process that is the cause of the error, The computer system has a learner that learns, by machine learning, a correspondence between the process used when the semiconductor inspection device implements the inspection action and the error that occurs due to the process, The computer system, after the learner completes the machine learning, inputs a new process to the learner, thereby obtaining, as an output of the learner, a presumption result for whether the error occurs when the semiconductor inspection device uses the new process, and presumes the process that is the cause of the error from the presumption result, The process description specifies one or more process parameters for the inspection action, The computer system has a process correction suggestion section that, by implementing iteration, suggests a correction suggestion for the process that generates the error, As the correction suggestion, the process correction suggestion section suggests, as a correction candidate parameter, the process parameter that the semiconductor inspection device should use, and suggests a correction amount of the correction candidate parameter, The learner further has an analyzer that calculates a contribution degree of each of the process parameters to the error, For some or all of the process parameters, the process correction suggestion section calculates a correction amount in descending order of the contribution degree.

Images