Information processing device, abnormality detection method and semiconductor manufacturing system

Abstract

To provide a technology for retrieving univariate abnormality and correlation abnormality of a candidate for an abnormality factor that occurs in a process from sensor values of a plurality of sensors installed in a semiconductor manufacturing device.SOLUTION: An information processing device performs abnormality detection of a semiconductor manufacturing device. The information processing device includes an acquisition part for acquiring a plurality of sensor values outputted from a plurality of sensors installed in the semiconductor manufacturing device, an inference part for inferring an abnormality degree of a process from the plurality of sensor values acquired by the acquisition part by using an abnormality detection model that has learned a correspondence relation between the plurality of sensor values and the abnormality degree of the process by learning data, abnormality detection part for detecting abnormality that occurs in the process on the basis of the inferred abnormality degree of the process, an abnormality factor retrieval part for retrieving univariate abnormality and correlation abnormality to be a candidate for the abnormality factor that occurs in the process by using a plurality of abnormality factor retrieval methods, and an abnormality detection result output part for outputting the detected abnormality and the retrieved univariate abnormality and correlation abnormality as abnormality detection results.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present disclosure relates to an information processing device, an anomaly detection method, and a semiconductor manufacturing system. [Background technology]

[0002] For example, one of the conventional methods for detecting an anomaly in a semiconductor manufacturing device is to generate a multivariate analysis model formula for evaluating the state of a processing device or predicting the processing result by multivariate analysis (see, for example, Patent Document 1).

[0003] In Patent Document 1, the correlation between the detection data detected from multiple sensors of the processing device when operating based on the setting data and the setting data is obtained for each of multiple processing devices by multivariate analysis, and a multivariate analysis model formula is generated for evaluating the device status of the processing device or predicting the processing results. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 4224454 Summary of the Invention [Problem to be solved by the invention]

[0005] The present disclosure provides a technique for searching for univariate anomalies and correlated anomalies that are candidates for anomaly causes occurring in a process from sensor values ​​of multiple sensors installed in semiconductor manufacturing equipment. [Means for solving the problem]

[0006] One aspect of the present disclosure is an information processing device that detects anomalies in semiconductor manufacturing equipment, the information processing device having: an acquisition unit that acquires multiple sensor values ​​output from multiple sensors installed in the semiconductor manufacturing equipment while a process is being executed; an inference unit that infers the degree of anomaly of the process from the multiple sensor values ​​acquired by the acquisition unit using an anomaly detection model that has been learned from learning data to determine a correspondence between the multiple sensor values ​​and the degree of anomaly of the process; an anomaly detection unit that detects an anomaly that has occurred in the process based on the inferred degree of anomaly of the process; an anomaly factor search unit that uses multiple anomaly factor search methods to search for univariate anomalies and correlated anomalies that are candidates for anomaly factors that have occurred in the process; and an anomaly determination result output unit that outputs the detected anomaly and the searched univariate anomalies and correlated anomalies as an anomaly determination result. Effect of the Invention

[0007] According to the present disclosure, it is possible to provide a technique for searching for univariate anomalies and correlated anomalies that are candidates for anomaly causes occurring in a process from sensor values ​​of multiple sensors installed in semiconductor manufacturing equipment. [Brief description of the drawings]

[0008] [Figure 1] 1 is a configuration diagram of an example of a semiconductor manufacturing system according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a diagram illustrating a hardware configuration of an example of a computer. [Diagram 3] FIG. 2 is a functional block diagram of an example of a semiconductor manufacturing apparatus according to the present embodiment. [Figure 4] FIG. 11 is an explanatory diagram showing an example of a procedure for creating a trained anomaly detection model. [Diagram 5] 11 is a flowchart showing an example of an anomaly detection procedure using a trained anomaly detection model according to the present embodiment. [Figure 6] FIG. 11 is an explanatory diagram showing an example of a processing procedure for detecting an abnormality that occurs in a process. [Figure 7] FIG. 11 is an explanatory diagram showing an example of a processing procedure for searching for an abnormality cause; [Figure 8]FIG. 11 is an explanatory diagram of an example of an anomaly detection model created for each section of a process. [Figure 9] FIG. 11 is an explanatory diagram of an example of a table displaying the degree of abnormality for each run and each section. [Figure 10] FIG. 13 is a diagram for explaining an example of calculation of the maximum deviation degree. [Figure 11] 11 is an explanatory diagram of a specific example of searching for candidates for abnormality causes in the semiconductor manufacturing system according to the embodiment; FIG. [Figure 12] 11 is an explanatory diagram of a specific example of searching for candidates for abnormality causes in the semiconductor manufacturing system according to the embodiment; FIG. [Figure 13] 11 is an explanatory diagram of a specific example of searching for candidates for abnormality causes in the semiconductor manufacturing system according to the embodiment; FIG. [Figure 14] 11 is an explanatory diagram of a specific example of searching for candidates for abnormality causes in the semiconductor manufacturing system according to the embodiment; FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0010] <System configuration> Fig. 1 is a configuration diagram of an example of a semiconductor manufacturing system according to this embodiment. The semiconductor manufacturing system 1 in Fig. 1 includes semiconductor manufacturing equipment 10, an equipment controller 12, a server device 14, and an operator terminal 16. The semiconductor manufacturing equipment 10 and the equipment controller 12 are installed in a manufacturing factory 2. The server device 14 and the operator terminal 16 may be installed in the manufacturing factory 2 or may be installed outside the manufacturing factory 2. The operator terminal 16 is operated by an operator such as an equipment technician or an analyst of the semiconductor manufacturing equipment 10 installed in the manufacturing factory 2.

[0011] The semiconductor manufacturing equipment 10, the equipment controller 12, the server device 14, and the operator terminal 16 are communicatively connected via networks 18 and 20, such as the Internet and a LAN (Local Area Network).

[0012] The semiconductor manufacturing equipment 10 is an equipment that performs processes such as film formation, etching, and ashing, and is an equipment that processes semiconductor wafers or an equipment that processes glass substrates for flat panel displays. The semiconductor manufacturing equipment 10 is, for example, a substrate processing equipment, a heat treatment equipment, or a film formation equipment. The semiconductor manufacturing equipment 10 receives a control command according to, for example, a recipe from an equipment controller 12 and executes a process. The semiconductor manufacturing equipment 10 is equipped with a plurality of sensors.

[0013] 1, the semiconductor manufacturing equipment 10 may be equipped with an equipment controller 12, or may not be equipped with the equipment controller 12 as long as it is communicably connected. The equipment controller 12 has a computer that controls the semiconductor manufacturing equipment 10. The equipment controller 12 outputs control commands to control control components of the semiconductor manufacturing equipment 10 according to a recipe, thereby causing the semiconductor manufacturing equipment 10 to execute a process according to the recipe.

[0014] The equipment controller 12 receives instructions for the semiconductor manufacturing equipment 10 from an operator, and also functions as a man-machine interface that provides the operator with information related to the semiconductor manufacturing equipment 10. The equipment controller 12 receives sensor values ​​output from a plurality of sensors installed in the semiconductor manufacturing equipment 10. The sensors are temperature sensors, pressure sensors, flow rate sensors, etc. The equipment controller 12 is an example of an information processing device that detects abnormalities in the semiconductor manufacturing equipment 10.

[0015] The server device 14 may receive sensor values ​​output from a plurality of sensors installed in the semiconductor manufacturing equipment 10 and store them as a process log. The server device 14, like the equipment controller 12, is an example of an information processing device that detects an abnormality in the semiconductor manufacturing equipment 10.

[0016] Furthermore, the equipment controller 12 or the server device 14 may display the result of the abnormality detection on the display of the equipment controller 12, the display of the server device 14, or the display of the operator terminal 16, etc., so as to notify the operator of the abnormality detection result. When the equipment controller 12 or the server device 14 detects an abnormality in the semiconductor manufacturing equipment 10, it may notify the operator by using email or the like.

[0017] The operator terminal 16 is a PC (Personal Computer) or a smartphone operated by an operator such as an equipment operator or an analysis operator of the semiconductor manufacturing equipment 10 installed in the manufacturing factory 2.

[0018] 1 is just one example, and it goes without saying that there are various system configurations depending on the application and purpose. For example, the semiconductor manufacturing system 1 can have various configurations, such as a configuration in which the equipment controller 12 of each semiconductor manufacturing equipment 10 is integrated into an equipment controller for multiple semiconductor manufacturing equipment 10, or a configuration in which it is further divided.

[0019] <Hardware configuration> The equipment controller 12, server device 14, and operator terminal 16 of the semiconductor manufacturing system 1 shown in Fig. 1 are realized by, for example, a computer (information processing device) having a hardware configuration shown in Fig. 2. Fig. 2 is a hardware configuration diagram of an example of the computer.

[0020] 2 includes an input device 501, an output device 502, an external I / F (interface) 503, a RAM (Random Access Memory) 504, a ROM (Read Only Memory) 505, a CPU (Central Processing Unit) 506, a communication I / F 507, and a HDD (Hard Disk Drive) 508, all of which are connected to each other via a bus B. The input device 501 and the output device 502 may be connected and used when necessary.

[0021] The input device 501 is a keyboard, mouse, touch panel, etc., and is used by an operator to input various operation signals. The output device 502 is a display, etc., and displays the results of processing by the computer 500. The communication I / F 507 is an interface that connects the computer 500 to the network 18 or 20. The HDD 508 is an example of a non-volatile storage device that stores programs and data.

[0022] The external I / F 503 is an interface with an external device. The computer 500 can read and / or write data from and to a recording medium 503a such as a Secure Digital (SD) memory card via the external I / F 503. The ROM 505 is an example of a non-volatile semiconductor memory (storage device) in which programs and data are stored. The RAM 504 is an example of a volatile semiconductor memory (storage device) that temporarily holds programs and data.

[0023] The CPU 506 is a calculation device that reads out programs and data from storage devices such as the ROM 505 and the HDD 508 onto the RAM 504 and executes processes, thereby realizing overall control and functions of the computer 500 .

[0024] The equipment controller 12, the server device 14, and the operator terminal 16 in FIG. 1 can realize various functions described below by executing programs on a computer 500 having the hardware configuration shown in FIG.

[0025] <Functional configuration> The semiconductor manufacturing apparatus 10 of the semiconductor manufacturing system 1 according to this embodiment is realized by, for example, functional blocks as shown in Fig. 3. Fig. 3 is a functional block diagram of an example of the semiconductor manufacturing apparatus according to this embodiment. Note that the functional block diagram of Fig. 3 omits illustration of configurations that are not necessary for explaining this embodiment.

[0026] The equipment controller 12 of the semiconductor manufacturing equipment 10 in FIG. 3 executes a program for the equipment controller 12, thereby realizing an acquisition unit 40, an anomaly detection model memory unit 42, a process log memory unit 44, an inference unit 46, an anomaly detection unit 48, an anomaly cause searching unit 50, an anomaly judgment result output unit 52, a learning data receiving unit 60, and a learning unit 62.

[0027] The multiple sensors 30 are installed in the semiconductor manufacturing equipment 10 and output multiple sensor values ​​during execution of a process. The multiple sensors 30 are, for example, temperature sensors, pressure sensors, flow rate sensors, etc. An acquisition unit 40 of the equipment controller 12 acquires the multiple sensor values ​​output from the multiple sensors 30 during execution of a process. A process log storage unit 44 stores the multiple sensor values ​​output from the multiple sensors 30 during execution of a process as a process log for each run of the process.

[0028] The inference unit 46 infers the degree of anomaly of the process from the multiple sensor values ​​of the multiple sensors 30 included in the process log, using an anomaly detection model that has been trained on the correspondence relationship between the multiple sensor values ​​of the multiple sensors 30 and the degree of anomaly of the process from training data. The anomaly detection model is stored in the anomaly detection model storage unit 42.

[0029] The learning data receiving unit 60 receives learning data used for learning the anomaly detection model by the learning unit 62. The learning data receiving unit 60 may receive the learning data from the server device 14, for example, or from another computer 500.

[0030] The learning unit 62 uses the learning data to make the anomaly detection model learn the correspondence between the multiple sensor values ​​of the multiple sensors 30 and the degree of anomaly in the process. The learning of the anomaly detection model by the learning unit 62 will be described in detail later.

[0031] The anomaly detection unit 48 detects an anomaly that has occurred in the process based on the anomaly degree of the process inferred by the inference unit 46. The anomaly detection unit 48 detects an anomaly that has occurred in the process when the anomaly degree of the process inferred by the inference unit 46 is included in an anomaly range described below. Details of the detection of anomalies by the anomaly detection unit 48 will be described later.

[0032] The anomaly cause search unit 50 searches for univariate anomalies and correlated anomalies that are candidates for the anomaly cause that has occurred in a process in which the anomaly detection unit 48 has detected an anomaly, using multiple anomaly cause search methods as described below.

[0033] Furthermore, the anomaly judgment result output unit 52 outputs the anomaly detected by the anomaly detection unit 48 and the univariate anomaly and correlation anomaly searched for by the anomaly cause search unit 50 as an anomaly judgment result. The anomaly judgment result output unit 52 issues a warning or issues an anomaly report to notify an operator of the anomaly judgment result. The operator may be notified by displaying the result on a display or the like, by email, or by printing out the result.

[0034] 3 is an example, and the learning data receiving unit 60 and the learning unit 62 may be provided in a computer 500 other than the device controller 12. For example, the learning data receiving unit 60 and the learning unit 62 may be provided in the server device 14.

[0035] The acquisition unit 40, anomaly detection model storage unit 42, process log storage unit 44, inference unit 46, anomaly detection unit 48, anomaly cause search unit 50, anomaly determination result output unit 52, learning data receiving unit 60, and learning unit 62 in the functional block diagram of Fig. 3 may be provided in the server device 14. The functional block diagram of Fig. 3 shows an example in which the equipment controller 12 is mounted on the semiconductor manufacturing equipment 10, but it does not have to be mounted thereon as long as it is connected so as to be able to communicate with the equipment.

[0036] In order to minimize damage to products in the semiconductor manufacturing equipment 10, it is necessary to detect anomalies as soon as they occur. In addition, the semiconductor manufacturing equipment 10 has many sensors 30, and these sensors 30 need to be monitored comprehensively. Therefore, the semiconductor manufacturing equipment 10 detects anomalies using a multivariate anomaly detection method. The multivariate anomaly detection method uses machine learning to calculate an anomaly level (objective variable) from the sensor values ​​(explanatory variables) of the multiple sensors 30, and considers an anomaly to have occurred when the anomaly level exceeds a threshold value.

[0037] When an abnormality occurs in the semiconductor manufacturing equipment 10, in order to quickly restore the semiconductor manufacturing equipment 10, an operator needs to quickly identify where and what kind of abnormality occurred in the semiconductor manufacturing equipment 10. However, since the multivariate anomaly detection method calculates the degree of anomaly by taking into account changes in the relationships between the sensor values ​​of multiple sensors 30, it is difficult to interpret what caused the abnormality (consider the cause of the abnormality).

[0038] Therefore, in this embodiment, multivariate anomaly detection using an anomaly detection model is combined with an anomaly cause search method to present and visualize individual sensors 30 that are candidates for anomaly causes and correlations between multiple sensors 30 as described below.

[0039] <Processing> The semiconductor manufacturing system 1 of this embodiment creates a trained anomaly detection model, for example, in the procedure shown in Fig. 4. Fig. 4 is an explanatory diagram showing an example of the procedure for creating a trained anomaly detection model. The trained anomaly detection model may be created by the equipment controller 12, or may be created by the server device 14 and registered in the equipment controller 12 of the semiconductor manufacturing equipment 10. Here, an example performed by the equipment controller 12 shown in Fig. 3 will be described.

[0040] The learning data receiving unit 60 of the equipment controller 12 receives learning data 100 that the learning unit 62 uses to learn the anomaly detection model. The learning data 100 associates multiple sensors 30 installed in the semiconductor manufacturing equipment 10 with sensor values ​​for each run of the process output from the multiple sensors 30 while the process is being executed. "Run1" and "Run2" in the learning data 100 are identification information for the run of the process. "Sensor-1" and "Sensor-2" in the learning data 100 are identification information for the multiple sensors 30.

[0041] The learning unit 62 creates a learned anomaly detection model by having a known anomaly detection algorithm learn normal learning data 100. The normal learning data 100 is sensor values ​​output from the multiple sensors 30 while the semiconductor manufacturing equipment 10 is executing a process when no anomaly is occurring.

[0042] The learning unit 62 can set an abnormality range 102 for the process anomaly degree based on the process anomaly degree for each process execution (hereinafter referred to as Run) output when normal learning data 100 is input to a trained anomaly detection model.

[0043] In the inference described below, if the degree of anomaly of the process output from the trained anomaly detection model is included in the anomaly range 102, the anomaly detection unit 48 can detect an anomaly that has occurred in the process.

[0044] The semiconductor manufacturing system 1 of this embodiment performs anomaly detection for the semiconductor manufacturing equipment 10, for example, in a procedure as shown in Fig. 5. Fig. 5 is a flowchart showing an example of an anomaly detection procedure using the trained anomaly detection model according to this embodiment.

[0045] In step S20, the acquisition unit 40 of the equipment controller 12 acquires, for each Run, a plurality of sensor values ​​output from the plurality of sensors 30 during execution of the process as a process log. The process log associates the plurality of sensors 30 installed in the semiconductor manufacturing equipment 10 with the sensor values ​​for each Run output from the plurality of sensors 30 during execution of the process.

[0046] In step S22, the inference unit 46 uses the trained anomaly detection model to infer the degree of anomaly of the process for each Run from the multiple sensor values ​​for each Run of the multiple sensors 30 included in the process log. Also, in step S24, the anomaly detection unit 48 detects an anomaly that has occurred in the process based on the degree of anomaly of the process inferred by the inference unit 46.

[0047] The processing in steps S20 to S24 is performed, for example, according to the procedure shown in Fig. 6. Fig. 6 is an explanatory diagram showing an example of a processing procedure for detecting an abnormality that occurs in a process.

[0048] In step S20, for example, when a new Run is executed, the acquisition unit 40 acquires new data 200, which is a process log of the new Run. In step S22, the inference unit 46 infers the degree of anomaly of the new data 200 from the sensor values ​​of the multiple sensors 30 included in the new data 200, using the trained anomaly detection model.

[0049] The graph shown in the lower part of Fig. 6 plots the degree of process abnormality for each run number (Run No.). Moreover, plot 204 in the graph shows the degree of process abnormality of the inferred new data 200. Note that the graph in Fig. 6 shows an example in which plot 204 of new data 200 is included in abnormal range 202.

[0050] If the anomaly level of the process inferred by the inference unit 46 is included in the anomaly range 202, the anomaly factor searching unit 50 proceeds from step S26 to step S28. If the anomaly level of the process inferred by the inference unit 46 is not included in the anomaly range 202, the anomaly factor searching unit 50 proceeds from step S26 to step S30.

[0051] In step S28, the abnormality factor search unit 50 uses a plurality of abnormality factor search methods to search for univariate anomalies and correlated anomalies that are candidates for the abnormality factor that has occurred in the process in which the abnormality detection unit 48 has detected an abnormality.

[0052] Fig. 7 is an explanatory diagram showing an example of a processing procedure for abnormality factor search. The graph shown on the left side of Fig. 7 shows an example in which a plot 302 showing the degree of abnormality of the inferred process of new data 200 is included in an abnormality range 300. The abnormality factor search unit 50 searches for univariate anomalies and correlation anomalies of the sensor 30 as abnormality factors for the runs of the plot 302 in which an abnormality is detected, using a plurality of abnormality factor search methods, and outputs, for example, a ranking and a graph.

[0053] "Method 1. Univariate Anomaly" shown in FIG. 7 is an example in which univariate anomalies found for runs of plot 302 in which anomalies were detected are output in the form of a ranking 304 and graph 306 of sensors 30. For the runs of plot 302 in which anomalies were detected, the ranking 304 finds the degree of deviation from normal sensor values ​​for each of the multiple sensors 30, and displays identification information such as the names of the sensors 30 in descending order of deviation. In the ranking 304 of FIG. 7, "Sensor-1" is output as the sensor 30 with the largest deviation. In addition, the graph 306 outputs the change in deviation for each run number (Run No.) of the sensor 30 with the largest deviation. A plot 308 of the graph 306 shows the deviation of "Sensor-1" for the runs in which anomalies were detected. In this way, the graph 306 can display a graph of sensors 30 with large deviations. Note that the graph 306 may display a plurality of graphs ranked high in the deviation degree ranking, or may display a graph selected by the user from the ranking 304 .

[0054] According to "Method 1. Univariate Anomaly" shown in Fig. 7, the change in the sensor value of the sensor 30 having the largest deviation from the normal sensor value for each run can be visually output, making it easy for the operator to search for the cause of the anomaly. In addition, the change in the sensor value of the sensor 30 having the highest ranking in deviation from the normal sensor value or the sensor value of the sensor 30 selected by the user from the ranking 304 can also be visually output, making it easy for the operator to search for the cause of the anomaly.

[0055] Moreover, "Method 2. Correlated Anomaly" shown in Fig. 7 is an example in which correlated anomalies found for runs of plot 302 in which an anomaly was detected are output as a ranking 310 and graph 312 of correlations between sensors 30. For the ranking 310, the relationship between the sensor values ​​of each of the two sensors 30 is found for the runs of plot 302 in which an anomaly was detected, and identification information such as the names of the two sensors 30 is displayed in order of the degree of deviation from the normal relationship. For the ranking 310 in Fig. 7, "Sensor-1 vs. Sensor-2" is output as the two sensors 30 with the greatest degree of deviation from the normal relationship.

[0056] The graph 312 outputs, for each Run, the relationship between the sensor values ​​of "Sensor-1" and "Sensor-2", which are the two sensors 30 with the largest deviation from the normal relationship. Note that a plot 314 of the graph 312 shows the relationship between the sensor values ​​of "Sensor-1" and "Sensor-2" for a Run in which an abnormality was detected. In this manner, the graph 312 can display a graph of the relationship between two sensors 30 with a large deviation. Note that the graph 312 may display multiple graphs of the relationship between two sensors 30 ranked high in the deviation degree ranking, or may display a graph of the relationship between two sensors 30 selected by the user from the ranking 310.

[0057] According to "Method 2. Correlation Anomaly" shown in Fig. 7, it is possible to visually output the change for each run in the correlation between the two sensors 30 that have the largest deviation from the relationship between the sensors 30 in a normal run, making it easy for the operator to search for the cause of the anomaly. In addition, it is also possible to visually output the change for each run in the correlation between the two sensors 30 that are ranked high in the degree of deviation from the relationship between the sensors 30 in a normal run, or the correlation between two sensors 30 selected by the user from the ranking 310, making it easy for the operator to search for the cause of the anomaly.

[0058] In this way, the abnormality cause search unit 50 uses a plurality of abnormality cause search methods to search for univariate anomalies and correlated anomalies that are candidates for abnormality causes that have occurred in a process in which an abnormality has been detected by the abnormality detection unit 48, and can search for abnormality causes not only in changes in the sensor value of a single sensor 30, but also in changes in the correlation between the sensor values ​​of the sensors 30.

[0059] Therefore, according to the semiconductor manufacturing system 1 of this embodiment, candidate causes of an abnormality are searched for by taking into consideration not only the change in the sensor value of a single sensor 30 but also the change in the correlation between the sensor values ​​of the two sensors 30, making it easier to search for abnormalities caused by multiple factors and enabling early recovery of the semiconductor manufacturing equipment 10 when an abnormality occurs.

[0060] 5, when the process proceeds from step S26 to step S30, the abnormality determination result output unit 52 outputs an abnormality determination result indicating that no abnormality has occurred in the process of the new data 200, based on the result of the abnormality detection process in step S24. When the process proceeds from step S28 to step S30, the abnormality determination result output unit 52 outputs an abnormality determination result indicating that an abnormality has occurred in the process of the new data 200, based on the result of the abnormality detection process in step S24 and the result of the abnormality cause search process in step S28.

[0061] The semiconductor manufacturing system 1 according to this embodiment may divide the process into a plurality of sections and create an anomaly detection model for each section, as shown in Fig. 8. Fig. 8 is an explanatory diagram of an example of an anomaly detection model created for each process section.

[0062] Fig. 8 shows an example in which a process is divided into three sections, "Section 1" to "Section 3", with different processing contents, and an "anomaly detection model for Section 1", "anomaly detection model for Section 2", or "anomaly detection model for Section 3" is created for each section. Note that the sections may also be called steps.

[0063] The "anomaly detection model for section 1" in FIG. 8 performs machine learning to determine the correspondence between the sensor values ​​of the multiple sensors 30 and the anomaly degree of the process in "section 1" in FIG. 8. The "anomaly detection model for section 2" in FIG. 8 performs machine learning to determine the correspondence between the sensor values ​​of the multiple sensors 30 and the anomaly degree of the process in "section 2" in FIG. 8. The "anomaly detection model for section 3" in FIG. 8 performs machine learning to determine the correspondence between the sensor values ​​of the multiple sensors 30 and the anomaly degree of the process in "section 3" in FIG. 8.

[0064] The inference unit 46 infers the degree of anomaly of the process from the sensor values ​​of the multiple sensors 30 in "Section 1" included in the process log, using an anomaly detection model that has learned the correspondence relationship between the sensor values ​​of the multiple sensors 30 in "Section 1" of Fig. 8 and the degree of anomaly of the process. The inference unit 46 infers the degree of anomaly of the process from the sensor values ​​of the multiple sensors 30 in "Section 2" included in the process log, using an anomaly detection model that has learned the correspondence relationship between the sensor values ​​of the multiple sensors 30 in "Section 2" of Fig. 8 and the degree of anomaly of the process. The inference unit 46 infers the degree of anomaly of the process from the sensor values ​​of the multiple sensors 30 in "Section 3" included in the process log, using an anomaly detection model that has learned the correspondence relationship between the sensor values ​​of the multiple sensors 30 in "Section 3" of Fig. 8 and the degree of anomaly of the process.

[0065] The anomaly detection unit 48 detects anomalies that have occurred in "Section 1" to "Section 3" of the process based on the anomaly degrees of "Section 1" to "Section 3" of the process inferred by the inference unit 46. The anomaly cause search unit 50 searches for univariate anomalies and correlated anomalies that are candidates for anomaly causes that have occurred in the process section, using a plurality of anomaly cause search methods, for the process section in which the anomaly detection unit 48 has detected an anomaly.

[0066] When a process is divided into a plurality of sections, an anomaly detection model is created for each section, the degree of anomaly in the process section is inferred, and an anomaly that has occurred in the process section is detected, the anomaly determination result output unit 52 can display the degree of anomaly for each run and section, for example, as shown in Fig. 9. Fig. 9 is an explanatory diagram of an example of a table displaying the degree of anomaly for each run and section.

[0067] The table shown in Fig. 9 displays the degree of anomaly, maximum deviation, and abnormal section rate [%] of the process section for each run number. For example, in Fig. 9, the range of the Warning and Alarm anomaly levels is specified based on the magnitude of the degree of anomaly of the process section, and the display of the section whose degree of anomaly falls within the range of Warning and Alarm anomaly levels is visually changed from the display of the section whose degree of anomaly falls outside the range of Warning and Alarm anomaly levels.

[0068] The maximum deviation is the maximum deviation among all the deviations of the process section, which is a numerical value indicating how abnormal the process is compared to the learning data (normal data) under normal conditions. The maximum deviation is calculated, for example, as shown in Figure 10.

[0069] Fig. 10 is a diagram for explaining an example of calculating the maximum deviation. In step S50, the anomaly detection unit 48 calculates the average and standard deviation of the anomaly degree for each section of the process from the anomaly degree of the learning data in the normal state. In Fig. 10, for example, the average anomaly degree of "1.08" and the standard deviation of "0.130" for "Section 1" are calculated from the learning data in the normal state for "Section 1".

[0070] In step S52, the anomaly detection unit 48 standardizes the degree of anomaly for each section of the new Run, which is the verification data, using the average and standard deviation of the degrees of anomaly for each section calculated in step S50. In Fig. 10, for example, the degree of anomaly "1.4" for "section 1" is standardized to a deviation of "2.454" using the average of the degrees of anomaly for each section, "1.08", and the standard deviation, "0.130", calculated in step S50.

[0071] In step S54, the anomaly detection unit 48 selects the maximum deviation for each run from the deviations for all sections obtained in step S52 as the maximum deviation. For example, in FIG. 10, the maximum deviation "2.454" is selected from the deviations "2.454, 2.454, 0.592" for all sections of the run "Run number 8."

[0072] It should be noted that the calculation example shown in FIG. 10 is merely an example, and the maximum deviation may be calculated using another method.

[0073] Returning to Figure 9, the abnormal section rate [%] indicates the percentage of sections in which an abnormality was detected among all sections in a run. For example, if a process is divided into five sections and an abnormality was detected in three sections in a run, the abnormal section rate is 60%.

[0074] According to the table shown in Fig. 9, the operator can easily grasp in which section of which run the abnormality was detected. Also, according to the table shown in Fig. 9, the operator can easily grasp the tendency of the abnormality based on the maximum deviation and the abnormal section rate, as described later.

[0075] <Example> Fig. 11 is an explanatory diagram of a specific example of searching for candidates for anomaly causes in the semiconductor manufacturing system according to this embodiment. Table 1000 shown in Fig. 11 shows an example in which the maximum deviation of run number "3" in which an anomaly was detected in "section 3" of the process is high and the rate of abnormal sections is low. In the example of table 1000, it is considered that an anomaly has occurred in the sensor value of sensor 30 which performs special processing in a specific section.

[0076] The abnormality cause searching unit 50 outputs a ranking 1002 of the sensors 30 causing the abnormality, using the abnormality cause searching method of "Method 1. Single variable abnormality" shown in Fig. 7 for "Section 3" of the run with run number "3" in which the abnormality was detected. According to the ranking 1002 of the sensors 30 causing the abnormality, it is possible to visually output the change in the sensor value of the sensor 30 for "MFC flow rate" which has the greatest deviation from the normal sensor value.

[0077] For example, if a mass flow controller (MFC) has a zero point deviation, there will be no difference in the displayed flow rate when gas is flowing, and so the operator will not notice a change in the actual flow rate.

[0078] Graph 1004 in Fig. 11 shows the change in the sensor value of sensor 30 for "MFC flow rate" for each section, and there is no difference in the displayed flow rate due to the influence of zero point shift in "Section 1," "Section 2," "Section 4," and "Section 5" where gas is flowing. On the other hand, graph 1004 in Fig. 11 shows a difference in the displayed flow rate due to the influence of zero point shift in "Section 3" where the gas flow rate is set to zero.

[0079] In this way, for example, if an MFC (mass flow controller) is experiencing a zero point deviation, the degree of abnormality will be high only in certain sections, so by checking the maximum deviation and abnormal section rate, it is possible to grasp the trend of the cause of the abnormality that has occurred.

[0080] Fig. 12 is an explanatory diagram of a specific example of searching for candidates for anomaly causes in the semiconductor manufacturing system according to this embodiment. Table 1100 shown in Fig. 12 shows an example in which the maximum deviation of run number "3" in which an anomaly was detected in "section 3" of the process is high and the rate of abnormal sections is low. In the example of table 1100, it is considered that an anomaly has occurred in the sensor value of sensor 30 which performs special processing in a specific section.

[0081] The abnormality cause searching unit 50 outputs a ranking 1102 of the sensors 30 causing abnormalities, using the abnormality cause searching method of "Method 1. Single variable abnormality" shown in Fig. 7 for "Section 3" of the run with run number "3" in which an abnormality was detected. According to the ranking 1102 of the sensors 30 causing abnormalities, it is possible to visually output the change in the sensor value of the "pressure" sensor 30 and the change in the sensor value of the "MFC opening" sensor 30, which have the largest deviation from the sensor value under normal conditions.

[0082] For example, if an MFC has a zero point deviation, there will be no difference in the displayed flow rate when gas is flowing, so the operator will not notice a change in the actual flow rate.

[0083] If the process does not include a section where the MFC flow rate setting is zero, such as "Section 3" shown in graph 1004 in Fig. 11, an abnormality cannot be detected from the displayed flow rate. Even in this case, as shown in area 1104 in Fig. 12, changes occur in the sensor values ​​of the pressure gauge and MFC opening sensor 30 on the same line as the target MFC.

[0084] In this way, for example, even if an MFC is experiencing a zero point deviation and the process does not include a section where the MFC flow rate setting value becomes zero, changes will occur in the sensor values ​​of the pressure gauge and MFC opening sensor 30 on the same line as the target MFC, and by checking the maximum deviation and abnormal section rate, the tendency of the cause of the abnormality that has occurred can be grasped.

[0085] Fig. 13 is an explanatory diagram of a specific example of searching for candidates for abnormality factors in the semiconductor manufacturing system according to this embodiment. Table 1200 shown in Fig. 13 shows an example in which the maximum deviation of Run number "3" is low and the abnormal section rate is high. In the example of table 1200, it is considered that an abnormality has occurred in the sensor value of sensor 30 that does not depend on the processing content of a specific section.

[0086] The abnormality factor searching unit 50 outputs a ranking 1202 of the sensors 30 that are abnormal factors, using the abnormality factor searching method "Method 1. Single variable abnormality" shown in Fig. 7 for "Section 5" of the run with run number "3" in which an abnormality was detected. According to the ranking 1202 of the sensors 30 that are abnormal factors, it is possible to visually output the change in the sensor value of the sensor 30 for "heater power" that has the largest deviation from the normal sensor value.

[0087] For example, heater power may deteriorate due to poor heater installation, deterioration over time, etc. Graph 1204 in Fig. 13 shows the change in the sensor value of the sensor 30 for "heater power" for each section for each run. Graph 1204 in Fig. 13 shows a change in the heater power for run number "3". In this way, when the heater power has deteriorated, the abnormal section rate becomes very high, so by checking the maximum deviation and the abnormal section rate, the tendency of the cause of the abnormality that has occurred can be understood.

[0088] Fig. 14 is an explanatory diagram of a specific example of searching for candidates for abnormality factors in the semiconductor manufacturing system according to this embodiment. As shown in area 1306 in Fig. 14, the heater power of the piping heater is controlled so that the temperature of the sensor value read by the sensor 30 such as TC becomes the desired temperature.

[0089] If the TC sensor 30 is not installed correctly due to poor pipe construction, the temperature in the pipes may be abnormal even if the displayed temperature is normal. As described above, the heater power is affected by the pipe construction state, and so there is a large variation, making it difficult to detect an abnormality based only on the sensor value of the heater power sensor 30.

[0090] The abnormality factor searching unit 50 outputs a ranking 1302 of the relationships between the sensors 30 that are abnormal factors, using the abnormality factor searching method of "Method 2. Correlation abnormality" shown in Fig. 7 for "Section 5" of the run with run number "3" in which an abnormality was detected. According to the ranking 1302 of the relationships between the sensors 30 that are abnormal factors, it is possible to visually output changes in the correlation between the sensors 30, such as "heater power Av vs. heater power B" and "heater power B vs. heater power C", which have a large change from the correlation between the sensors 30 under normal conditions.

[0091] Graph 1304 in Fig. 14 shows the correlation between the sensors 30 for "heater power" for each section for each run. In graph 1304 in Fig. 14, plots 1308 and 1310 indicate a collapsed correlation. If a sensor 30 such as TC is not installed correctly due to poor piping construction, the relationship between the sensors 30 changes, so by checking the correlation anomaly searched for in the process section where the anomaly was detected, the tendency of the cause of the anomaly that occurred can be understood.

[0092] According to this embodiment, since not only the change in the sensor value of a single sensor 30 but also the change in the correlation between the sensors 30 is taken into consideration, it is possible to deal with anomalies caused by multiple factors. Furthermore, according to this embodiment, the process is divided into sections, multivariate anomaly detection and anomaly cause search are performed for each section, and anomaly determination results can be output for each section, making it easier to consider the anomaly causes.

[0093] Although the preferred embodiments of the present invention have been described in detail above, the present invention is not limited to the above-described embodiments, and various modifications and substitutions can be made to the above-described embodiments without departing from the scope of the present invention. [Explanation of symbols]

[0094] 1. Semiconductor manufacturing systems 10. Semiconductor manufacturing equipment 12 Equipment Controller 14 Server equipment 16 Worker terminal 18, 20 Network 30 Sensors 40 Acquisition Department 42 Anomaly detection model memory unit 44 Process Log Storage 46 Reasoning part 48 Anomaly detection unit 50 Abnormality Cause Investigation Department 52 Abnormality judgment result output section 60 Learning data receiver 62 Learning Department

Claims

1. An information processing device that detects an abnormality in a semiconductor manufacturing device, an acquisition unit that acquires a plurality of sensor values ​​output from a plurality of sensors installed in the semiconductor manufacturing apparatus during execution of a process; an inference unit that infers an anomaly degree of the process from the sensor values ​​acquired by the acquisition unit, using an anomaly detection model that has been trained on a correspondence relationship between the sensor values ​​and an anomaly degree of the process using training data; an anomaly detection unit that detects an anomaly that has occurred in the process based on the inferred anomaly degree of the process; an anomaly factor searching unit that searches for univariate anomalies and correlation anomalies that are candidates for an anomaly factor occurring in the process by using a plurality of anomaly factor searching methods; an anomaly determination result output unit that outputs the detected anomaly and the searched univariate anomaly and correlation anomaly as an anomaly determination result; An information processing device having the above configuration.

2. The process is divided into a number of sections, the anomaly detection model has learned a correspondence relationship between the plurality of sensor values ​​and anomaly degrees of the process for each of the sections; the inference unit infers an anomaly level of the process section from the plurality of sensor values ​​output during execution of the process section using the anomaly detection model; The anomaly detection unit detects an anomaly that occurs in the process section based on the inferred degree of anomaly of the process section.

2. The information processing apparatus according to claim 1,

3. The abnormality determination result output unit displays, for each run of the process, the degree of abnormality of the process section, a maximum deviation degree of the degree of abnormality of the process section that is the maximum deviation degree from the degree of abnormality in a normal state, and an abnormal section rate that is a ratio of the section in which an abnormality is detected among the sections of the process.

3. The information processing apparatus according to claim 2,

4. The abnormality determination result output unit displays the plurality of sensors according to a difference between the sensor value in a normal state and the sensor value of the process in which the abnormality is detected, as a univariate abnormality that is a candidate for an abnormality cause that has occurred in the process, for the execution of the process in which the abnormality is detected.

4. The information processing apparatus according to claim 3,

5. The abnormality determination result output unit displays, for the execution of the process in which the abnormality is detected, a correlation between the plurality of sensors according to a difference from a correlation between the plurality of sensors in a normal state as a correlation abnormality that is a candidate for an abnormality cause that has occurred in the process.

5. The information processing device according to claim 3 or 4.

6. The abnormality determination result output unit visually changes and displays a process section having an abnormal value equal to or greater than a threshold value from a process section having an abnormal value not equal to or greater than the threshold value.

5. The information processing device according to claim 2, wherein:

7. An anomaly detection method for detecting an anomaly in a semiconductor manufacturing device, comprising: acquiring a plurality of sensor values ​​outputted from a plurality of sensors installed in the semiconductor manufacturing apparatus during execution of a process; inferring an anomaly degree of the process from the acquired sensor values ​​by using an anomaly detection model that has been trained on a correspondence relationship between the sensor values ​​and an anomaly degree of the process using training data; Detecting an anomaly that has occurred in the process based on the inferred anomaly degree of the process; searching for univariate anomalies and correlated anomalies that are candidates for anomaly factors occurring in the process using a plurality of anomaly factor searching methods; outputting the detected anomaly and the searched univariate anomaly and correlation anomaly as an anomaly determination result; The anomaly detection method includes the steps of:

8. A semiconductor manufacturing system having a semiconductor manufacturing apparatus and an information processing apparatus that detects an abnormality in the semiconductor manufacturing apparatus, an acquisition unit that acquires a plurality of sensor values ​​output from a plurality of sensors installed in the semiconductor manufacturing apparatus during execution of a process; an inference unit that infers an anomaly degree of the process from the sensor values ​​acquired by the acquisition unit by using an anomaly detection model that has been trained on a correspondence relationship between the sensor values ​​and an anomaly degree of the process using training data; an anomaly detection unit that detects an anomaly that has occurred in the process based on the inferred anomaly degree of the process; an anomaly factor searching unit that searches for univariate anomalies and correlation anomalies that are candidates for an anomaly factor occurring in the process by using a plurality of anomaly factor searching methods; an anomaly determination result output unit that outputs the detected anomaly and the searched univariate anomaly and correlation anomaly as an anomaly determination result; A semiconductor manufacturing system having

Images