Abnormality detection device, abnormality detection method, and recording medium

Abstract

Provided is an abnormality detection device, an abnormality detection method, and a recording medium that enable high-precision abnormality detection processing. The abnormality detection device includes: an acquisition unit that acquires a time-series data group measured in association with processing of an object in a predetermined processing unit of a manufacturing process; and a learning unit that includes a plurality of network units that process the acquired time-series data group, a connection unit that synthesizes each output data output by processing using the plurality of network units, and performs machine learning on the plurality of network units and the connection unit in a manner that causes a synthesis result output by the connection unit to approximate information representing an abnormality degree acquired when the object is processed in the predetermined processing unit of the manufacturing process.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to an abnormality detection device, an abnormality detection method, and a recording medium. BACKGROUND

[0002] Conventionally, an abnormality detection technique is known, which detects an abnormality occurring in various manufacturing processes, based on measurement data (a data set of various time series data, hereinafter referred to as "time series data group") measured in processing of an object.

[0003] For example, in a semiconductor manufacturing process, it is determined whether or not there is an abnormality or the degree of abnormality, by monitoring a time series data group measured in processing of a wafer, using an abnormality detection model generated in advance.

[0004] <Related Art>

[0005] <Patent Literature>

[0006] Patent Literature 1: Japanese Patent Application Publication No. 2011-100211 SUMMARY

[0007] <Problems to be Solved by the Invention>

[0008] The present disclosure provides an abnormality detection device, an abnormality detection method, and a recording medium, which can perform high-precision abnormality detection processing.

[0009] <Means for Solving the Problems>

[0010] The abnormality detection device according to one embodiment of the present disclosure, for example, has the following configuration. That is, it includes: an acquisition unit that acquires a time series data group measured in processing of an object in a predetermined processing unit of a manufacturing process; and a learning unit that includes a plurality of network units that process the acquired time series data group, and a connection unit that synthesizes each output data output by processing using the plurality of network units, and performs machine learning for the plurality of network units and the connection unit in such a manner that a result of the synthesis output by the connection unit approaches information indicating a degree of abnormality acquired when the object is processed in the predetermined processing unit of the manufacturing process.

[0011] <EFFECTS OF THE INVENTION>

[0012] According to the present disclosure, it is possible to provide an abnormality detection device, an abnormality detection method, and a recording medium, which can perform high-precision abnormality detection processing. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1FIG. 1 is a first drawing showing one example of the overall configuration of a system including a semiconductor manufacturing process and an abnormality detection device.

[0014] Figure 2 FIG. 1 is a first drawing showing one example of a predetermined processing unit of a semiconductor manufacturing process.

[0015] Figure 3 FIG. 2 is a second drawing showing one example of a predetermined processing unit of a semiconductor manufacturing process.

[0016] Figure 4 FIG. 3 is a drawing showing one example of the hardware configuration of an abnormality detection device.

[0017] Figure 5 FIG. 4 is a first drawing showing one example of learning data.

[0018] Figure 6 FIG. 5 is a drawing showing one example of a time series data group.

[0019] Figure 7 FIG. 6 is a first drawing showing one example of the functional configuration of a learning unit.

[0020] Figure 8 FIG. 7 is a first drawing showing a specific example of the processing of a branching unit.

[0021] Figure 9 FIG. 8 is a second drawing showing a specific example of the processing of a branching unit.

[0022] Figure 10 FIG. 9 is a third drawing showing a specific example of the processing of a branching unit.

[0023] Figure 11 FIG. 10 is a drawing showing a specific example of the processing of a normalization unit included in each network unit.

[0024] Figure 12 FIG. 11 is a fourth drawing showing a specific example of the processing of a branching unit.

[0025] Figure 13 FIG. 12 is a first drawing showing one example of the functional configuration of a speculation unit.

[0026] Figure 14 FIG. 13 is a first flowchart showing the flow of an abnormality detection process.

[0027] Figure 15 FIG. 14 is a second drawing showing one example of the overall configuration of a system including a semiconductor manufacturing process and an abnormality detection device.

[0028] Figure 16 FIG. 15 is a second drawing showing one example of learning data.

[0029] Figure 17is a diagram showing one example of OES data.

[0030] Figure 18 is a diagram showing a specific example of the processing of the normalization section included in each network section into which the input OES data is input.

[0031] Figure 19 is a diagram showing a specific example of the processing of each normalization section.

[0032] Figure 20 is a diagram showing a specific example of the processing of the pooling section.

[0033] Figure 21 is a second diagram showing one example of the functional configuration of the speculation section.

[0034] Figure 22 is a second flowchart showing the flow of the abnormality detection processing. DETAILED DESCRIPTION

[0035] Hereinafter, each embodiment will be described with reference to the drawings. Note that in the present specification and the drawings, the same reference signs are assigned to the same constituent elements having substantially the same functional configuration, so that redundant description is omitted.

[0036] [First Embodiment]

[0037] Overall Configuration of System Including Semiconductor Manufacturing Process and Abnormality Detection Device

[0038] First, the overall configuration of a system including a manufacturing process (here, a semiconductor manufacturing process) and an abnormality detection device will be described. Figure 1 is a first diagram showing one example of the overall configuration of a system including a semiconductor manufacturing process and an abnormality detection device. As shown in Figure 1 , the system 100 includes a semiconductor manufacturing process, time series data acquisition devices 140_1 to 140_n, and an abnormality detection device 160.

[0039] In the semiconductor manufacturing process, an object (pre-process wafer 110) is processed in a predetermined processing unit 120, and a resultant object (post-process wafer 130) is generated. Note that the processing unit 120 referred to here is an abstract concept, and details thereof will be described later. In addition, the pre-process wafer 110 refers to a wafer (substrate) before being processed in the processing unit 120, and the post-process wafer 130 refers to a wafer (substrate) after being processed in the processing unit 120.

[0040] The time-series data acquisition devices 140_1 to 140_n acquire time-series data measured in association with the processing of the pre-process wafer 110 in the processing unit 120, respectively. The time-series data acquisition devices 140_1 to 140_n perform measurement with respect to mutually different kinds of measurement items. Note that the number of measurement items measured by the time-series data acquisition devices 140_1 to 140_n, respectively, can be one or a plurality. In addition, among the time-series data measured in association with the processing of the pre-process wafer 110, time-series data measured during pre-processing and post-processing performed before and after the processing of the pre-process wafer 110 is included in addition to time-series data measured during the processing of the pre-process wafer 110. The processing can include pre-processing and post-processing performed in a state in which there is no wafer (substrate).

[0041] The time-series data groups acquired by the time-series data acquisition devices 140_1 to 140_n are stored in the learning data storage section 163 of the abnormality detection device 160 as learning data (input data).

[0042] Note that information indicating the degree of abnormality acquired in the processing unit 120 during the processing of the pre-process wafer 110 is associated with the time-series data groups, and is stored in the learning data storage section 163 of the abnormality detection device 160 as learning data (correct answer data).

[0043] An abnormality detection program is installed in the abnormality detection device 160, and the abnormality detection device 160 functions as the learning section 161 and the estimation section 162 by executing the program.

[0044] The learning section 161 performs machine learning using the learning data (the time-series data groups acquired by the time-series data acquisition devices 140_1 to 140_n, and the information indicating the degree of abnormality associated with the time-series data groups). Specifically, the learning section 161 performs machine learning with respect to a plurality of network sections by processing the time-series data groups as input data using the plurality of network sections, and in a manner in which the resultant of each output data outputted approaches the information indicating the degree of abnormality as correct answer data.

[0045] The estimation section 162 inputs the time-series data groups acquired by the time-series data acquisition devices 140_1 to 140_n in association with the processing of a new object (pre-process wafer) in the processing unit 120 to the plurality of network sections in which machine learning has been performed. Thereby, the estimation section 162 estimates information indicating the degree of abnormality based on the time-series data acquired in association with the processing of the new pre-process wafer.

[0046] Thus, by being configured to process the time series data sets measured in association with the processing of the objects in the predetermined processing units 120 of the semiconductor manufacturing process using the plurality of network sections, the time series data sets of the predetermined processing units can be analyzed in various ways. Therefore, by the abnormality detection device 160, a model (estimation section 162) for realizing high-precision abnormality detection processing can be generated compared to a configuration in which processing is performed using one network section.

[0047] <the predetermined processing unit of the semiconductor manufacturing process>

[0048] Next, the predetermined processing unit 120 of the semiconductor manufacturing process will be described. Figure 2 is a first drawing showing one example of a predetermined processing unit of a semiconductor manufacturing process. As shown in Figure 2 , the semiconductor manufacturing device 200, which is one example of a substrate processing device, has a plurality of chambers (one example of a plurality of processing spaces. In Figure 2 , the chambers are "chamber A" to "chamber C"), and wafers are processed in each of the chambers.

[0049] Among them, Figure 2 (a) shows a case in which the plurality of chambers are defined as the processing units 120. In this case, the pre-processing wafer 110 refers to a wafer before being processed in the chamber A, and the post-processing wafer 130 refers to a wafer after being processed in the chamber C.

[0050] In addition, the time series data sets measured in association with the processing of the pre-processing wafer 110 in the processing units 120 of Figure 2 (a) include:

[0051] • the time series data sets output in association with the processing of the wafer in the chamber A (the first processing space);

[0052] • the time series data sets output in association with the processing of the wafer in the chamber B (the second processing space); and

[0053] • the time series data sets output in association with the processing of the wafer in the chamber C (the third processing space).

[0054] On the other hand, Figure 2 (b) shows a case in which one chamber (in the example of Figure 2 (b), the chamber B) is defined as the processing unit 120. In this case, the pre-processing wafer 110 refers to a wafer before being processed in the chamber B (a wafer after being processed in the chamber A), and the post-processing wafer 130 refers to a wafer after being processed in the chamber B (a wafer before being processed in the chamber C).

[0055] In addition, the time series data sets measured in association with the processing of the pre-processing wafer 110 in the processing units 120 ofFigure 2 (b) The time series data set measured in the processing unit 120 accompanying the processing of the pre-processing wafer 110 includes the time series data set measured in chamber B accompanying the processing of the pre-processing wafer 110.

[0056] Figure 3 Figure 2 is an example of a predetermined processing unit in a semiconductor manufacturing process. (Compared to...) Figure 2 Similarly, the semiconductor manufacturing apparatus 200 has multiple chambers, and the wafer is processed in each chamber through multiple processing contents.

[0057] in, Figure 3 (a) illustrates a case where processing unit 120 is defined as the processing contents in chamber B other than preprocessing and postprocessing (referred to as “wafer processing”). In this case, preprocessing wafer 110 refers to the wafer before wafer processing (the wafer after preprocessing), and postprocessing wafer 130 refers to the wafer after wafer processing (the wafer before postprocessing).

[0058] In addition, Figure 3 (a) The time series data set measured in the processing unit 120 accompanying the processing of the pre-processing wafer 110 includes the time series data set measured in chamber B accompanying the wafer processing of the pre-processing wafer 110.

[0059] It should be noted that, in Figure 3 In example (a), wafer processing is shown as a processing unit 120 in which preprocessing, wafer processing (this processing), and post-processing are performed in the same chamber (chamber B). However, for example, in cases where each processing is performed in a different chamber, such as preprocessing in chamber A, wafer processing in chamber B, and post-processing in chamber C, each processing in each chamber can be used as a processing unit 120.

[0060] on the other hand, Figure 3 (b) illustrates a process formulation (in) the wafer processing within the processing contents of chamber B. Figure 3 In example (b), the processing is defined as "Process Recipe III" in the case of processing unit 120. In this case, the pre-processing wafer 110 refers to the wafer before processing with Process Recipe III (the wafer after processing with Process Recipe II). Furthermore, the post-processing wafer 130 refers to the wafer after processing with Process Recipe III (the wafer before processing with Process Recipe IV (not shown)).

[0061] In addition, Figure 3The time-series data set measured in the processing unit 120 of (b) in conjunction with the processing of the wafer before processing 110 includes a time-series data set measured in the chamber B in conjunction with the processing by the process recipe III.

[0062] <Hardware configuration of abnormality detection device>

[0063] Next, the hardware configuration of the abnormality detection device 160 will be described. Figure 4 is a diagram illustrating one example of the hardware configuration of the abnormality detection device. As shown in Figure 4 the abnormality detection device 160 has a CPU (Central Processing Unit) 401, a ROM (Read Only Memory) 402, and a RAM (Random Access Memory) 403. In addition, the abnormality detection device 160 has a GPU (Graphics Processing Unit) 404. Note that the processor (processing circuit, Processing Circuit, Processing Circuitry) such as the CPU 401 and the GPU 404, and the memory such as the ROM 402 and the RAM 403 form a so-called computer.

[0064] In addition, the abnormality detection device 160 has an auxiliary storage device 405, a display device 406, an operation device 407, an I / F (Interface) device 408, and a driver device 409. Note that each hardware of the abnormality detection device 160 is connected to each other via a bus 410.

[0065] The CPU 401 is an arithmetic device for executing various programs (for example, an abnormality detection program and the like) installed in the auxiliary storage device 405.

[0066] The ROM 402 is a nonvolatile memory, and functions as a main storage device. The ROM 402 stores various programs, data, and the like required for the CPU 401 to execute various programs installed in the auxiliary storage device 405. Specifically, the ROM 402 stores a boot program such as a BIOS (Basic Input / Output System) or an EFI (Extensible Firmware Interface), and the like.

[0067] The RAM 403 is a volatile memory such as a DRAM (Dynamic Random Access Memory) or an SRAM (Static Random Access Memory), and functions as a main storage device. The RAM 403 provides a work area that expands when various programs installed in the auxiliary storage device 405 are executed by the CPU 401.

[0068] The GPU 404 is an arithmetic device for image processing, and performs a high-speed operation by parallel processing with respect to various image data (time series data sets in the present embodiment) when an anomaly detection program is executed by the CPU 401. Note that the GPU 404 is equipped with an internal memory (GPU memory), and temporarily stores information required when various image data are processed in parallel.

[0069] The auxiliary storage device 405 stores various programs, or various data used when the various programs are executed by the CPU 401, and the like. For example, the learning data storage 163 is realized in the auxiliary storage device 405.

[0070] The display device 406 is a display device for displaying the internal state of the anomaly detection device 160. The operation device 407 is an input device used when various instructions are input to the anomaly detection device 160 by the manager of the anomaly detection device 160. The I / F device 408 is a connection device for connecting to and communicating with a network not shown.

[0071] The drive device 409 is a device for setting a recording medium 420. The recording medium 420 referred to herein includes a medium that records information optically, electrically, or magnetically, such as a CD-ROM, a floppy disk, an optical magnetic disk, and the like. In addition, the recording medium 420 can include a semiconductor memory that records information electrically, such as a ROM, a flash memory, and the like.

[0072] Note that the various programs installed in the auxiliary storage device 405 are installed, for example, by setting a distributed recording medium 420 in the drive device 409, and reading the various programs recorded in the recording medium 420 by the drive device 409. Alternatively, the various programs installed in the auxiliary storage device 405 can be installed by downloading via a network not shown.

[0073] <Specific Example of Learning Data>

[0074] Next, the learning data read from the learning data storage 163 when machine learning is performed by the learning unit 161 is described. Figure 5Fig. 1 is a diagram showing one example of learning data. As shown in Figure 5 The learning data 500 includes "device", "recipe kind", "time series data group", "abnormality degree" as information items. Note that, here, a case where the predetermined processing unit 120 is processing of one process recipe is explained.

[0075] In the "device", an identifier of a semiconductor manufacturing device (for example, the semiconductor manufacturing device 200) as an object of monitoring whether or not an abnormality has occurred is stored. In the "recipe kind", an identifier (for example, the process recipe I) of a kind of process recipe that is executed at the time of measurement of the time series data group among process recipes that are executed in the corresponding semiconductor manufacturing device (for example, the EqA) is stored.

[0076] In the "time series data group", a time series data group that is measured by the time series data acquisition device 140_1 to 140_n at the time of processing of the process recipe determined by the "recipe kind" in the semiconductor manufacturing device determined by the "device" is stored.

[0077] In the "abnormality degree", information (information that is obtained by quantifying the abnormality degree) that indicates the abnormality degree that is acquired at the time of measurement of the corresponding time series data group (for example, the time series data group 1) by the time series data acquisition device 140_1 to 140_n is stored. Note that, in the example of Figure 5 In the example of Fig. 1, as the information that indicates the abnormality degree, the abnormality degree 0 to the abnormality degree 3 are stored.

[0078] <Specific Example of Time Series Data Group>

[0079] Next, a specific example of the time series data group that is measured by the time series data acquisition device 140_1 to 140_n is explained. Figure 6 Fig. 2 is a diagram showing one example of a time series data group. Note that, in the example of Figure 6 In the example of Fig. 2, in order to simplify the explanation, although it is assumed that one-dimensional data is measured by the time series data acquisition device 140_1 to 140_n respectively, two-dimensional data (a data set of a plurality of one-dimensional data) can be measured by one time series data acquisition device.

[0080] In the example of Fig. 2, 6a indicates that the time series data group is measured by the time series data acquisition device 140_1 to 140_n respectively. Figure 2 (b), Figure 3 (a), Figure 3Any one of (b) defines the time series data group in the case where the processing units 120 are defined by (a). In this case, the time series data acquisition devices 140_1 to 140_n acquire time series data measured in the chamber B accompanying the processing of the wafer 110 before the processing, respectively. In addition, the time series data acquisition devices 140_1 to 140_n acquire time series data measured in the same time range as the time series data group.

[0081] On the other hand, 6b shows a case where the time series data group in the case where the processing units 120 are defined by (a) is defined by (b). In this case, the time series data acquisition devices 140_1 to 140_n acquire time series data measured in the chamber B accompanying the processing of the wafer 110 before the processing, respectively. In addition, the time series data acquisition devices 140_1 to 140_n acquire time series data measured in the same time range as the time series data group. Figure 2 (a) defines the time series data group in the case where the processing units 120 are defined by (a). In this case, the time series data acquisition devices 140_1 to 140_3 acquire the time series data group 1 measured in the chamber A accompanying the processing of the wafer before the processing, for example. In addition, the time series data acquisition device 140_n-2 acquires the time series data group 2 measured in the chamber B accompanying the processing of the wafer, for example. In addition, the time series data acquisition devices 140_n-1 to 140_n acquire the time series data group 3 measured in the chamber C accompanying the processing of the wafer, for example.

[0082] Note that, in 6a, a case is shown where the time series data acquisition devices 140_1 to 140_n acquire time series data measured in the chamber B accompanying the processing of the wafer before the processing as the time series data group in the same time range. However, the time series data acquisition devices 140_1 to 140_n can acquire time series data measured in the chamber B accompanying the processing of the wafer before the processing as the time series data group in different time ranges.

[0083] Specifically, the time series data acquisition devices 140_1 to 140_n can acquire a plurality of time series data measured during the execution of the pre-processing as the time series data group 1. In addition, the time series data acquisition devices 140_1 to 140_n can acquire a plurality of time series data measured during the execution of the wafer processing as the time series data group 2. Furthermore, the time series data acquisition devices 140_1 to 140_n can acquire a plurality of time series data measured during the execution of the post-processing as the time series data group 3.

[0084] Likewise, the time-series data acquisition device 140_1 to 140_n can acquire a plurality of time-series data measured during execution of the process recipe I as a time-series data group 1. Further, the time-series data acquisition device 140_1 to 140_n can acquire a plurality of time-series data measured during execution of the process recipe II as a time-series data group 2. Furthermore, the time-series data acquisition device 140_1 to 140_n can acquire a plurality of time-series data measured during execution of the process recipe III as a time-series data group 3.

[0085] <Function configuration of learning unit>

[0086] Next, the function configuration of the learning unit 161 will be described. Figure 7 Fig. 1 is a diagram illustrating one example of the function configuration of the learning unit. The learning unit 161 has a branching unit 710, a first network unit 720_1 to an Mth network unit 720_M, a connection unit 730, and a comparison unit 740.

[0087] The branching unit 710 is one example of an acquisition unit, and is configured to read out a time-series data group from the learning data storage unit 163. Further, the branching unit 710 processes the time-series data group so that the time-series data group is processed by a plurality of network units from the first network unit 720_1 to the Mth network unit 720_M.

[0088] The first network unit 720_1 to the Mth network unit 720_M are configured on the basis of a convolutional neural network (CNN), and have a plurality of layers.

[0089] Specifically, the first network unit 720_1 has a first layer 720_11 to an Nth layer 720_1N. Likewise, the second network unit 720_2 has a first layer 720_21 to an Nth layer 720_2N. Hereinafter, the Mth network unit 720_M has a first layer 720_M1 to an Nth layer 720_MN, which have the same configuration.

[0090] In each of the first layer 720_11 to the Nth layer 720_1N of the first network unit 720_1, various processes such as normalization processing, convolution processing, activation processing, or pooling processing are performed. Likewise, the same various processes are performed in each of the first network unit 720_2 to the Mth network unit 720_M.

[0091] The connection unit 730 synthesizes each of the output data from the output data outputted from the Nth layer 720_1N of the first network unit 720_1 to the output data outputted from the Nth layer 720_MN of the Mth network unit 720_M, and outputs the synthesis result to the comparison unit 740.

[0092] The comparison section 740 compares the synthesis result output from the connection section 730 with the information (correct answer data) indicating the degree of abnormality read out from the learning data storage section 163, and calculates an error. In the learning section 161, the error is back-propagated in such a manner that the error calculated by the comparison section 740 satisfies a predetermined condition, and machine learning is performed with respect to the 1st network section 720_1 to the Mth network section 720_M and the connection section 730.

[0093] Thereby, the model parameters of the 1st layer to the Nth layer of the 1st network section 720_1 to the Mth network section 720_M and the model parameters of the connection section 730 are optimized.

[0094] <Details of the processing of each part of the learning section>

[0095] Next, the details of the processing of each part of the learning section 161 (in particular, the branching section here) will be described with a specific example.

[0096] (1) Details 1 of the processing of the branching section

[0097] First, the details of the processing of the branching section 710 will be described. Figure 8 is a first view showing a specific example of the processing of the branching section. In the case of Figure 8 , the branching section 710 generates a time series data group 1 (a first time series data group) by processing the time series data groups measured by the time series data acquisition devices 140_1 to 140_n according to a first reference, and inputs it to the 1st network section 720_1.

[0098] In addition, the branching section 710 generates a time series data group 2 (a second time series data group) by processing the time series data groups measured by the time series data acquisition devices 140_1 to 140_n according to a second reference, and inputs it to the 2nd network section 720_2.

[0099] In this way, by performing machine learning on the basis of being configured to process the time series data groups according to different references and divide them into respective different network sections, it is possible to analyze the processing unit 120 in various ways. Therefore, compared to the case of being configured to process the time series data groups with one network section, it is possible to generate a model (the estimation section 162) for achieving higher estimation accuracy.

[0100] Note that, although in the case of Figure 8The example shows the case where two time series data sets are generated by processing them according to two benchmarks, but it is also possible to generate three or more time series data sets by processing them according to three or more benchmarks.

[0101] (2) Details of the processing carried out by the branch office

[0102] Next, details of the other processing of branch 710 will be explained. Figure 9 Figure 2 shows a specific example of the processing of a branch. Figure 9 In this case, branch unit 710 groups the time series data sets measured by time series data acquisition devices 140_1 to 140_n according to data type, thereby generating time series data set 1 (first time series data set) and time series data set 2 (second time series data set). Furthermore, branch unit 710 inputs the generated time series data set 1 to third network unit 720_3 and the generated time series data set 2 to fourth network unit 720_4.

[0103] In this way, by performing machine learning based on the configuration of dividing time series data into multiple groups according to data type and processing them using different network units, the processing unit 120 can be analyzed from multiple perspectives. Therefore, compared to the case where time series data groups are input into a single network unit for machine learning, a model (prediction unit 162) that achieves higher prediction accuracy can be generated.

[0104] It should be noted that, although in Figure 9 In the example, the time series data sets are grouped according to the differences in the types of data based on the differences in the time series data acquisition devices 140_1 to 140_n. However, the time series data sets can also be grouped according to the time range used to acquire the data. For example, if the time series data sets are time series data sets (time series data sets 1 to 3) measured in conjunction with processing performed through multiple process formulations (process formulations I to III), the time series data sets can be divided into 3 groups according to the time range of each process formulation.

[0105] (3) Details of the processing carried out by the branch 3

[0106] Next, details of other processing performed by branch 710 will be explained. Figure 10 Figure 3 shows a specific example of how the branch is processed. Figure 10In this case, the branch unit 710 inputs the time-series data group acquired by the time-series data acquisition units 140_1 to 140_n to both the 5th network unit 720_5 and the 6th network unit 720_6. Then, the same time-series data group is subjected to different processing (normalization processing) in the 5th network unit 720_5 and the 6th network unit 720_6.

[0107] Figure 11 is a diagram illustrating a specific example of the processing of the normalization unit included in each network unit. As shown in Figure 11 In each layer of the 5th network unit 720_5, a normalization unit, a convolution unit, an activation function unit, and a pooling unit are included.

[0108] Figure 11 The example of the 1st layer 720_51 among the layers included in the 5th network unit 720_5 includes a normalization unit 1101, a convolution unit 1102, an activation function unit 1103, and a pooling unit 1104.

[0109] In the normalization unit 1101, first normalization processing is performed on the time-series data group inputted by the branch unit 710, and a normalized time-series data group 1 (first time-series data group) is generated.

[0110] Similarly, Figure 11 The example of the 1st layer 720_61 among the layers included in the 6th network unit 720_6 includes a normalization unit 1111, a convolution unit 1112, an activation function unit 1113, and a pooling unit 1114.

[0111] In the normalization unit 1111, second normalization processing is performed on the time-series data group inputted by the branch unit 710, and a normalized time-series data group 2 (second time-series data group) is generated.

[0112] In this way, by performing machine learning on the basis of processing of the time-series data group by a plurality of network units configured to include a normalization unit that performs normalization processing by a different method in each network unit, the processing unit 120 can be analyzed in various ways. Therefore, compared to a case where the network unit is configured to perform one kind of normalization processing on the time-series data group, a model (estimation unit 162) for achieving higher estimation accuracy can be generated.

[0113] (4) Details 4 of the processing by the branch unit

[0114] Next, details of other processing by the branch unit 710 will be described. Figure 12 is a fourth diagram illustrating a specific example of the processing of the branch unit. In Figure 12In the case of time series data, the branch 710 inputs the time series data group 1 (first time series data group) that is measured by the time series data acquisition devices 140_1 to 140_n and is measured during the processing of the wafer in chamber A to the 7th network unit 720_7.

[0115] Additionally, branch 710 inputs time series data group 2 (second time series data group) that is measured during the processing of the wafer in chamber B from the time series data groups measured by time series data acquisition devices 140_1 to 140_n to the 8th network unit 720_8.

[0116] In this way, by performing machine learning on each time series data set measured by processing in different chambers (first processing space, second processing space) using different network units, the processing unit 120 can be analyzed from multiple perspectives. Therefore, compared to the case where each time series data set is processed using a single network unit, a model (prediction unit 162) with higher prediction accuracy can be generated.

[0117] <Functional Composition of the Speculation Department>

[0118] Next, the functional structure of the speculation unit 162 will be explained. Figure 13 Figure 1 is an example illustrating the functional structure of the speculation unit. (See Figure 1.) Figure 13 As shown, the speculation unit 162 has a branch unit 1310, a first network unit 1320_1 to an Mth network unit 1320_M, and a connection unit 1330.

[0119] Branch unit 1310 acquires newly measured time series data sets from time series data acquisition devices 140_1 to 140_N. Furthermore, branch unit 1310 controls the processing of the acquired time series data sets in the first network unit 1320_1 to the Mth network unit 1320_M.

[0120] The first network section 1320_1 to the Mth network section 1320_M are formed by using the learning unit 161 to perform machine learning and optimizing the model parameters of each layer of the first network section 720_1 to the Mth network section 720_M.

[0121] The connection section 1330 is formed by the connection section 730 that performs machine learning using the learning section 161 and optimizes the model parameters. The connection section 1330 synthesizes each of the output data from the output data output by the Nth layer 1320_1N of the 1st network section 1320_1 to the output data output by the Nth layer 1320_MN of the Mth network section 1320_M. Thereby, in the connection section 1330, information indicating the degree of abnormality is output.

[0122] In this way, the estimation section 162 is generated by machine learning using the learning section 161 that analyzes the time series data sets of the predetermined processing units 120 in a variety of ways. Therefore, the estimation section 162 can also be applied to different process recipes, different chambers, and different apparatuses. Alternatively, the estimation section 162 can also be applied before and after maintenance of the same chamber. In other words, by the estimation section 162 according to the present embodiment, for example, it is not necessary to maintain or re-learn the model as the chamber is maintained, unlike the conventional technique.

[0123] <Flow of abnormality detection processing>

[0124] Next, the flow of the entire abnormality detection processing performed by the abnormality detection apparatus 160 will be described. Figure 14 is a 1st flowchart illustrating the flow of the abnormality detection processing.

[0125] In step S1401, the learning section 161 acquires the time series data sets and the information indicating the degree of abnormality as learning data.

[0126] In step S1402, the learning section 161 performs machine learning with the time series data sets among the acquired learning data as input data and with the information indicating the degree of abnormality as correct answer data.

[0127] In step S1403, the learning section 161 determines whether to continue the machine learning. In a case where further learning data is acquired and the machine learning is continued (in a case where the determination in step S1403 is YES), the processing returns to step S1401. On the other hand, in a case where the machine learning is ended (in a case where the determination in step S1403 is NO), the processing proceeds to step S1404.

[0128] In step S1404, the estimation section 162 generates the 1st network section 1320_1 to the Mth network section 1320_2 by reflecting the model parameters optimized by the machine learning.

[0129] In step S1405, the estimation section 162 estimates the information indicating the degree of abnormality by inputting the time series data sets measured in association with processing of a new pre-processing wafer.

[0130] In step S1406, the estimation unit 162 outputs the estimation result together with an identifier indicating the corresponding semiconductor manufacturing apparatus or an identifier indicating the kind of the corresponding process recipe, and the like.

[0131] <Summary>

[0132] As described above, it is clear that the abnormality detection device according to the first embodiment has the following features:

[0133] • acquires a time series data group measured in association with processing of an object in a predetermined processing unit of a manufacturing process;

[0134] • for the acquired time series data group,

[0135] • generates a first time series data group and a second time series data group by processing according to a first reference and a second reference, or

[0136] • groups according to a data kind or a time range,

[0137] and synthesizes each output data output by processing using a plurality of network units; or

[0138] • inputs the acquired time series data group to a plurality of network units that normalize using different methods, respectively, and synthesizes each output data output by processing using the plurality of network units;

[0139] • machine learns the plurality of network units in a manner such that a synthesis result obtained by synthesizing each output data approaches information indicating a degree of abnormality acquired when processing the object in the predetermined processing unit of the manufacturing process;

[0140] • processes a time series data group measured by the time series acquisition device in association with processing of a new object using the plurality of network units that have been machine learned, and estimates a synthesis result of each output data output by the plurality of network units as information indicating a degree of abnormality.

[0141] In this way, by being configured to input a time series data group to a plurality of network units and machine learning, it is possible to analyze a time series data group of a predetermined processing unit of a semiconductor manufacturing process in various ways. Therefore, compared to a case where a time series data group is input to one network unit and machine learning is performed, it is possible to generate a model for realizing high-precision abnormality detection processing.

[0142] In other words, according to the first embodiment, it is possible to provide an abnormality detection device capable of performing high-precision abnormality detection processing.

[0143] [Embodiment 2]

[0144] In the abnormality detection device 160 according to Embodiment 1, four kinds of configurations are shown as configurations in which the plurality of network sections processes the acquired time series data group. In this regard, in Embodiment 2, the configurations in which the plurality of network sections each including a normalization section that performs normalization processing using a different method processes the time series data group among the four kinds of configurations are explained in further detail. Note that, in the explanation, the same reference numerals are given to the same components as those in Embodiment 1, and the explanation thereof is omitted.

[0145] • The time series data acquisition device is an emission spectroscopic analysis device,

[0146] • The time series data group is OES (Optical Emission Spectroscopy) data (a data set of time series data of the number of emission intensities corresponding to the kind of wavelength) is assumed. Hereinafter, Embodiment 2 is explained focusing on the difference from Embodiment 1 described above.

[0147] <Overall configuration of the system including the semiconductor manufacturing process and the abnormality detection device>

[0148] First, the overall configuration of the system including the semiconductor manufacturing process and the abnormality detection device in the case where the time series data acquisition device is an emission spectroscopic analysis device is explained. Figure 15 is a second drawing showing one example of the overall configuration of the system including the semiconductor manufacturing process and the abnormality detection device. As shown in Figure 15 , the system 1500 includes a semiconductor manufacturing process, an emission spectroscopic analysis device 1501, and an abnormality detection device 160.

[0149] In the system 1500 shown in Figure 15 , the emission spectroscopic analysis device 1501 measures OES data as a time series data group using an emission spectroscopic analysis technique in association with the processing of the wafer 110 before the processing in the processing unit 120. Part of the OES data measured by the emission spectroscopic analysis device 1501 is stored in the learning data storage section 163 of the abnormality detection device 160 as learning data (input data) when machine learning is performed.

[0150] <Specific example of learning data>

[0151] Next, the learning data read from the learning data storage section 163 when machine learning is performed by the learning section 161 is explained. Figure 16 is a second drawing showing one example of learning data. As shown in Figure 16 , the learning data 1600 includes a plurality of pieces of data 1601 to 1604 each including a plurality of pieces of time series data 1611 to 1614.Figure 5 The information items of the learning data 500 shown are the same information items. And Figure 5 the difference from that is that it includes "OES data" as an information item instead of "time series data group", and stores the OES data measured by the emission spectroscopic analysis device 1501.

[0152] <Specific example of OES data>

[0153] Next, a specific example of the OES data measured in the emission spectroscopic analysis device 1501 will be described. Figure 17 is a diagram showing an example of the OES data.

[0154] In Figure 17 it, the graph 1710 is a graph showing the characteristics of the OES data as a time series data group measured in the emission spectroscopic analysis device 1501, and the horizontal axis shows the wafer identification number for identifying each pre-process wafer 110 processed in the processing unit 120. In addition, the vertical axis shows the time length of the OES data measured in the emission spectroscopic analysis device 1501 along with the processing of each pre-process wafer 110.

[0155] As shown in the graph 1710, the time length of the OES data measured in the emission spectroscopic analysis device 1501 is different for each wafer to be processed.

[0156] In Figure 17 the example, for example, the OES data 1720 shows the OES data measured along with the processing of the pre-process wafer with the wafer identification number = "770". The data size in the vertical direction of the OES data 1720 depends on the wavelength range measured in the emission spectroscopic analysis device 1501. In the second embodiment, since the emission spectroscopic analysis device 1501 measures the emission intensity within a predetermined wavelength range, the data size in the vertical direction of the OES data 1720 is, for example, the number of wavelengths "N λ " included in the predetermined wavelength range.

[0157] On the other hand, the data size in the horizontal direction of the OES data 1720 depends on the time length during measurement in the emission spectroscopic analysis device 1501. In Figure 17 the example, the data size in the horizontal direction of the OES data 1720 is "LT".

[0158] In this way, it can be said that the OES data 1720 is a time series data group that aggregates one-dimensional time series data with a predetermined number of wavelengths and a predetermined time length for each wavelength.

[0159] Note that in the branch section 710, when the OES data 1720 is input to the 5th network section 720_5 and the 6th network section 720_6, resizing processing is performed in mini-batch units in such a manner that the data size is made the same as that of the OES data of the other wafer identification numbers.

[0160] <Specific example of processing by normalization section>

[0161] Next, a specific example of processing by the normalization section of the 5th network section 720_5 and the 6th network section 720_6 to which the OES data 1720 is input from the branch section 710 will be described.

[0162] Figure 18 is a diagram showing a specific example of processing by the normalization section included in each network section to which OES data is input. As shown in Figure 18 the 1st layer 720_51 among the layers included in the 5th network section 720_5 has a normalization section 1101. In addition, the normalization section 1101 generates normalized data (normalized OES data 1810) by normalizing the OES data 1720 using the first method (using the average value and the standard deviation of the emission intensity for the entire wavelength).

[0163] In addition, as shown in Figure 18 the 1st layer 720_61 among the layers included in the 6th network section 720_6 has a normalization section 1111. In addition, the normalization section 1111 generates normalized data (normalized OES data 1820) by normalizing the OES data 1720 using the second method (using the average value and the standard deviation of the emission intensity for each wavelength).

[0164] Figure 19 is a diagram showing a specific example of processing by each normalization section. As shown in 19a, in the normalization section 1101, normalization is performed using the average value and the standard deviation of the emission intensity for the entire wavelength. On the other hand, as shown in 19b, in the normalization section 1111, normalization is performed using the average value and the standard deviation of the emission intensity for each wavelength.

[0165] Thus, depending on what is taken as a reference to observe the change in the emission intensity (in other words, depending on the analysis method), the information that can be observed changes even for the same OES data 1720. In the abnormality detection device 160 according to the second embodiment, the same OES data 1720 is processed by the respective different network sections for different normalization processing. By thus combining a plurality of normalization processing, the OES data 1720 in the processing unit 120 can be analyzed in various aspects. Thus, compared to a case where the OES data 1720 is subjected to one kind of normalization processing by one network section, a model (the estimation section 162) for achieving higher estimation accuracy can be generated.

[0166] Note that, although in the above specific example, the case where normalization is performed using the average value and the standard deviation of the emission intensity is described, the statistical value used when performing normalization is not limited thereto. For example, normalization can be performed using the maximum value and the standard deviation of the emission intensity, or normalization can be performed using other statistical values. Further, it can be configured so that which statistical value is used for normalization can be selected.

[0167] <Specific example of processing by pooling section>

[0168] Next, a specific example of the processing by the pooling section included in the final layer of the fifth network section 720_5 and the sixth network section 720_6 will be described. Figure 20 is a view illustrating a specific example of the processing by the pooling section.

[0169] Since the data size differs between the mini-batches, in the pooling sections 1104, 1114 included in the final layer of the fifth network section 720_5 and the sixth network section 720_6, the pooling processing is performed in such a manner that data of a fixed length is output between the mini-batches.

[0170] Figure 20 is a view illustrating a specific example of the processing by the pooling section. As Figure 20 indicated, in the pooling sections 1104, 1114, GAP (Global Average Pooling) processing is performed on the feature data output from the activation function sections 1103, 1113.

[0171] In Figure 20 , the feature data 2011_1 to 2011_m is feature data input to the pooling section 1104 of the Nth layer 720_5N of the fifth network section 720_5, and indicates feature data generated based on the OES data belonging to the mini-batch 1. The feature data 2011_1 to 2011_m respectively indicates one channel portion of the feature data.

[0172] Further, the feature data 2012_1 to 2012_m is feature data input to the pooling unit 1104 of the Nth layer 720_5N of the 5th network unit 720_5, and indicates feature data generated based on the OES data belonging to the mini-batch 2. The feature data 2012_1 to 2012_m indicates one channel portion of feature data, respectively.

[0173] Further, the feature data 2031_1 to 2031_m, 2032_1 to 2032_m are also the same as the feature data 2011_1 to 2011_m, 2012_1 to 2012_m (however, N λ channel portions of feature data), respectively.

[0174] Here, in the pooling units 1104, 1114, by calculating the average value in a channel unit for each feature amount value included in the input feature data, output data of a fixed length is output. By this, it is possible to make the data output from the pooling units 1104, 1114 have the same data size between the mini-batches.

[0175] <Function configuration of the inference unit>

[0176] Next, the function configuration of the inference unit 162 will be described. Figure 21 is a second drawing illustrating one example of the function configuration of the inference unit. As Figure 21 indicated, the inference unit 162 has a branching unit 1310, a 5th network unit 1320_5 and a 6th network unit 1320_6, and a connection unit 1330.

[0177] The branching unit 1310 acquires the OES data newly measured by the spectrometer spectrometer analysis device 1501. Further, the branching unit 1310 branches the acquired OES data in a manner that the acquired OES data is processed in the 5th network unit 1320_5 and the 6th network unit 1320_6.

[0178] The 5th network unit 1320_5 and the 6th network unit 1320_6 are formed by performing machine learning using the learning unit 161 and optimizing the model parameters of each layer of the 5th network unit 720_5 and the 6th network unit 720_6.

[0179] The connection unit 1330 is formed by performing machine learning using the learning unit 161 and optimizing the model parameters. The connection unit 1330 synthesizes the output data output by the Nth layer 1320_5N of the 5th network unit 1320_5 and the output data output by the Nth layer 1320_6N of the 6th network unit 1320_6, and outputs information indicating the degree of abnormality.

[0180] In this way, the estimation unit 162 is generated by machine learning using the learning unit 161 that analyzes the OES data of the predetermined processing unit 120 in various aspects. Therefore, the estimation unit 162 can also be applied to different process recipes, different chambers, different apparatuses. Alternatively, the estimation unit 162 can also be applied before and after maintenance of the same chamber. In other words, with the estimation unit 162 according to the present embodiment, it is not necessary to maintain or re-learn the model as with the conventional technology, for example, along with maintenance of the chamber.

[0181] <Flow of abnormality detection processing>

[0182] Next, the flow of the entire abnormality detection processing by the abnormality detection apparatus 160 will be described. Figure 22 is a second flowchart showing the flow of the abnormality detection processing.

[0183] In step S2201, the learning unit 161 acquires the OES data and the information indicating the degree of abnormality as learning data.

[0184] In step S2202, the learning unit 161 performs machine learning with the OES data among the acquired learning data as input data and with the information indicating the degree of abnormality as correct answer data.

[0185] In step S2203, the learning unit 161 determines whether to continue the machine learning. In a case where further learning data is acquired and the machine learning is continued (in a case where the determination in step S2203 is YES), the processing returns to step S2201. On the other hand, in a case where the machine learning is ended (in a case where the determination in step S2203 is NO), the processing proceeds to step S2204.

[0186] In step S2204, the estimation unit 162 generates the fifth network unit 1320_5 and the sixth network unit 1320_6 by reflecting the model parameters optimized by the machine learning.

[0187] In step S2205, the estimation unit 162 estimates the information indicating the degree of abnormality by inputting the OES data measured by the emission spectrometric analysis apparatus 1501 along with processing of a new pre-processing wafer.

[0188] In step S2206, the estimation unit 162 outputs the estimation result together with an identifier indicating the corresponding semiconductor manufacturing apparatus or an identifier indicating the kind of the corresponding process recipe, and the like.

[0189] <Summary>

[0190] As can be clearly seen from the above, the abnormality detection apparatus according to the second embodiment has the following features:

[0191] • Acquiring OES data measured by the emission spectrometric analysis device accompanying processing of the object in a predetermined processing unit of the manufacturing process;

[0192] • Inputting the acquired OES data to two network sections which are normalized by different methods, respectively, and synthesizing each output data outputted from the two network sections;

[0193] • Performing machine learning for the two network sections in a manner that a result of the synthesis obtained by synthesizing each output data approaches information indicating a degree of abnormality acquired when the object is processed in the predetermined processing unit of the manufacturing process;

[0194] • Processing OES data measured by the emission spectrometric analysis device accompanying processing of a new object by the two network sections which have performed the machine learning, and inferring a result of the synthesis of each output data outputted from the two network sections as information indicating a degree of abnormality.

[0195] In this way, by being configured to input OES data to two network sections and perform machine learning, it is possible to analyze OES data of a predetermined processing unit of a semiconductor manufacturing process in various ways. Therefore, compared to a case where OES data is input to one network section and machine learning is performed, it is possible to generate a model for realizing high-precision abnormality detection processing.

[0196] In other words, according to the second embodiment, it is possible to provide an abnormality detection device capable of performing high-precision abnormality detection processing.

[0197] [Third Embodiment]

[0198] Although, in the above-described second embodiment, the emission spectrometric analysis device is cited as one example of the time series data acquisition device, the time series data acquisition device described in the first embodiment is not limited to the emission spectrometric analysis device.

[0199] For example, in the time series data acquisition device described in the first embodiment, a process data acquisition device for acquiring various process data such as temperature data, pressure data, flow rate data of a gas, and the like as one-dimensional time series data can be included. Alternatively, in the time series data acquisition device described in the first embodiment, a high-frequency power source device for plasma for acquiring various RF data such as RF power source voltage data as one-dimensional time series data can be included.

[0200] In addition, although the type of abnormality is not particularly mentioned in the above-described first and second embodiments, the type of abnormality that occurs in the processing unit 120 includes any type of abnormality. Note that the type of abnormality that occurs in the processing unit 120 includes, for example, an abnormality that occurs due to a change over time and an abnormality that occurs suddenly.

[0201] The abnormality that occurs due to a change over time includes, for example, an abnormality that occurs in a result object such as a film thickness of a wafer or an etching rate. In addition, the abnormality that occurs due to a change over time includes a change over time of a part in the apparatus or a device connected to the apparatus itself (wear of a part, wear of an electrode, degradation of a device, deposition of a film on a part in the apparatus, and the like). In addition, the abnormality that occurs due to a change over time includes an abnormality caused by a change over time of a part in the apparatus or a device connected to the apparatus (fluctuation in gas flow, temperature abnormality, and the like).

[0202] On the other hand, the abnormality that occurs suddenly includes, for example, abnormal discharge, liquid droplet (a phenomenon in which a large number of micron-sized particles are attached), unstable operation of discharge, air leakage or helium gas leakage, and the like.

[0203] In addition, in the above-described first and second embodiments, a case in which information indicating the degree of abnormality is stored as learning data (correct answer data) is described. However, instead of storing information indicating the degree of abnormality, it is also possible to store a comparison result (presence or absence of abnormality) on the basis of a comparison between information indicating the degree of abnormality and a predetermined threshold value as learning data (correct answer data). In this case, in the learning unit 161, machine learning is performed in such a manner that the learning data coincides with the stored presence or absence of abnormality, and in the estimation unit 162, the presence or absence of abnormality is estimated.

[0204] Alternatively, it is possible to configure such that, in the learning unit 161, the same machine learning as in the first and second embodiments is performed, and on the basis of a conversion of information indicating the degree of abnormality output from the connection unit 730 into information indicating the presence or absence of abnormality by the estimation unit 162, the information is output as an estimation result.

[0205] In the conversion from information indicating the degree of abnormality into information indicating the presence or absence of abnormality, for example, a sigmoid function can be used. Specifically, in a case where an output when information indicating the degree of abnormality is input to the sigmoid function is 0.5 or more, the estimation unit 162 outputs "abnormal" as information indicating the presence or absence of abnormality. On the other hand, in a case where the output when information indicating the degree of abnormality is input to the sigmoid function is less than 0.5, the estimation unit 162 outputs "normal" as information indicating the presence or absence of abnormality.

[0206] In addition, in the above-described first and second embodiments, the learning unit performs machine learning using the same learning data regardless of the type of abnormality. However, the method of machine learning performed by the learning unit is not limited thereto, and can be configured to perform machine learning using different learning data according to the type of abnormality. Specifically, for example, the learning unit can be configured to set a plurality of network units that perform machine learning in such a manner that the network units output information indicating the presence or absence of an abnormality, and perform machine learning using different learning data.

[0207] • a plurality of network units that perform machine learning in such a manner that the network units output information indicating the presence or absence of an abnormality, and perform machine learning using different learning data.

[0208] • a plurality of network units that perform machine learning in such a manner that the network units output information indicating the presence or absence of an abnormality, and perform machine learning using different learning data.

[0209] In addition, in the above-described first and second embodiments, the learning unit performs machine learning using the same learning data regardless of the type of abnormality. However, the method of machine learning performed by the learning unit is not limited thereto, and can be configured to perform machine learning using different learning data according to the type of abnormality. Specifically, for example, the learning unit can be configured to set a plurality of network units that perform machine learning in such a manner that the network units output information indicating the presence or absence of an abnormality, and perform machine learning using different learning data.

[0210] However, the method of machine learning performed by the learning unit is not limited thereto, and can be configured to perform machine learning in such a manner that the learning unit outputs information indicating the type of abnormality in addition to information indicating the degree of abnormality. Specifically, the learning unit can be configured to set a number of output units corresponding to the number of types of abnormality in the connection unit, and output the reliability of the occurrence of an abnormality for each type.

[0211] [Other Embodiments]

[0212] In the above-described first to third embodiments, the machine learning algorithm of each network unit of the learning unit 161 is configured based on a convolutional neural network. However, the machine learning algorithm of each network unit of the learning unit 161 is not limited to a convolutional neural network, and can be configured based on other machine learning algorithms.

[0213] In addition, in the above-described first to third embodiments, the abnormality detection device 160 functions as the learning unit 161 and the estimation unit 162. However, the device functioning as the learning unit 161 and the device functioning as the estimation unit 162 do not need to be integrated, and can be configured separately. In other words, the abnormality detection device 160 can function as the learning unit 161 without the estimation unit 162, or can function as the estimation unit 162 without the learning unit 161.

[0214] Note that other elements can be combined with the configurations described above, and the present disclosure is not limited to the configurations shown here. Changes can be made in the range not departing from the gist of the present application, and can be appropriately determined according to the application mode.

Claims

1. An abnormality detection device comprising: an acquisition unit that acquires a time series data group measured in association with processing of an object in a predetermined processing unit of a manufacturing process; and a learning unit that includes a plurality of network units that respectively process the time series data group acquired by the acquisition unit and subjected to size adjustment processing according to the amount of output data for each mini batch, that is, the time series data group subjected to branching processing according to processing to be performed by a corresponding network unit, and a connection unit that synthesizes each output data output by processing using the plurality of network units, and performs machine learning for the plurality of network units and the connection unit in a manner such that a synthesis result output by the connection unit approaches information indicating a degree of abnormality acquired when the object is processed in the predetermined processing unit of the manufacturing process, wherein output data of the same data size as the data size corresponding to the data size of each mini batch is output to the connection unit in a manner such that the data size of the output data from the final layer of each of the plurality of network units corresponds to the data size of each mini batch. 2.The abnormality detection device according to claim 1, further comprising: a presumption unit that presumes the synthesis result output by the connection unit in which machine learning is performed as information indicating a degree of abnormality when a new object is processed by processing a time series data group acquired for the new object using the plurality of network units in which machine learning is performed. 3.The abnormality detection device according to claim 2, wherein the presumption unit outputs information indicating the presence or absence of abnormality based on the presumed information indicating the degree of abnormality. 4.The abnormality detection device according to claim 1, wherein the learning unit generates a first time series data group and a second time series data group by processing the acquired time series data group according to a first reference and a second reference, respectively, and performs machine learning for different network units and the connection unit in a manner such that a synthesis result output by the connection unit approaches information indicating a degree of abnormality acquired when the object is processed in the predetermined processing unit of the manufacturing process by processing each time series data group using the different network units. 5.The abnormality detection device according to claim 4, further comprising: a presumption unit that generates a first time series data group and a second time series data group by processing a time series data group acquired for a new object according to the first reference and the second reference, respectively, and presumes a synthesis result output by the connection unit in which machine learning is performed as information indicating a degree of abnormality when the new object is processed by processing each time series data group using the different network units in which machine learning is performed. 6.The abnormality detection device according to claim 1, wherein The learning unit performs machine learning with respect to the different network units and the connection unit in a manner that the synthesis result output from the connection unit approaches information indicating a degree of abnormality acquired when the object is processed in the predetermined processing unit of the manufacturing process, by grouping the acquired time-series data sets according to data categories or time ranges and processing each group by using the different network units.

7. The abnormality detection apparatus according to claim 6, further comprising: a presumption unit that presumes the synthesis result output from the connection unit in which machine learning is performed as information indicating a degree of abnormality when a new object is processed, by processing time-series data sets acquired for the new object according to data categories or time ranges and processing each group by using the different network units in which machine learning is performed.

8. The abnormality detection apparatus according to claim 1, wherein the learning unit performs machine learning with respect to the different network units and the connection unit in a manner that the synthesis result output from the connection unit approaches information indicating a degree of abnormality acquired when the object is processed in the predetermined processing unit of the manufacturing process, by processing the acquired time-series data sets by using different network units each including a normalization unit that normalizes by using a different method.

9. The abnormality detection apparatus according to claim 8, further comprising: a presumption unit that presumes the synthesis result output from the connection unit in which machine learning is performed as information indicating a degree of abnormality when a new object is processed, by processing time-series data sets acquired for the new object by using the different network units in which machine learning is performed.

10. The abnormality detection apparatus according to claim 1, wherein the learning unit performs machine learning with respect to the different network units and the connection unit in a manner that the synthesis result output from the connection unit approaches information indicating a degree of abnormality acquired when the object is processed in the predetermined processing unit of the manufacturing process, by processing a first time-series data set measured in association with processing of the object in a first processing space within the predetermined processing unit and a second time-series data set measured in association with processing of the object in a second processing space within the predetermined processing unit by using different network units.

11. The abnormality detection apparatus according to claim 10, further comprising: a presumption unit that presumes the synthesis result output from the connection unit in which machine learning is performed as information indicating a degree of abnormality when a new object is processed, by processing a first time-series data set measured in association with processing of the new object in the first processing space within the predetermined processing unit and a second time-series data set measured in association with processing of the new object in the second processing space within the predetermined processing unit by using the different network units in which machine learning is performed.

12. The abnormality detection apparatus according to claim 1, wherein the time-series data group is data measured in association with processing in a substrate processing apparatus.

13. The abnormality detection apparatus according to claim 1, wherein the learning unit performs machine learning on the plurality of network units and the connection unit in such a manner that the synthesis result output from the connection unit approaches information indicating a degree of abnormality and information indicating a kind of abnormality, which are acquired when the object is processed in the predetermined processing unit of the manufacturing process, by processing the acquired time-series data group using the plurality of network units.

14. The abnormality detection apparatus according to claim 8, wherein the time-series data group is data measured by an emission spectroscopic analysis apparatus in association with processing in a substrate processing apparatus, and is data indicating an emission intensity of each wavelength measured at each time.

15. The abnormality detection apparatus according to claim 14, wherein the normalization unit included in a first network unit among the different network units normalizes with respect to the entire wavelengths using a statistical value of the emission intensity.

16. The abnormality detection apparatus according to claim 14, wherein the normalization unit included in a second network unit among the different network units normalizes with respect to each wavelength using a statistical value of the emission intensity.

17. The abnormality detection apparatus according to claim 8, wherein a pooling unit included in a final layer among the layers of the different network units performs GAP processing.

18. An abnormality detection apparatus comprising: an acquisition unit that acquires a time-series data group measured in association with processing of an object in a predetermined processing unit of a manufacturing process; and a presumption unit that includes a plurality of network units that respectively process the time-series data group, which is acquired by the acquisition unit and which is subjected to size adjustment processing according to the number of output data for each mini-batch, i.e., which is subjected to branching processing according to processing to be performed by a corresponding network unit, and a connection unit that synthesizes each output data output by processing using the plurality of network units, and presumes a synthesis result output from the connection unit as information indicating a degree of abnormality when the object is processed, wherein the plurality of network units and the connection unit are subjected to machine learning in such a manner that the synthesis result output from the connection unit approaches information indicating a degree of abnormality and information indicating a kind of abnormality, which are acquired when the object is processed in the predetermined processing unit of the manufacturing process, by processing a time-series data group acquired in advance using the plurality of network units, and output output data from a final layer of each of the plurality of network units to the connection unit in such a manner that a data size of the output data is a data size corresponding to a data size of each mini-batch.

19. An abnormality detection method comprising: an acquisition process of acquiring a time-series data group measured in association with processing of an object in a predetermined processing unit of a manufacturing process; and The learning step is performed on the plurality of network units and the connection unit in a learning unit including a plurality of network units that respectively process the time-series data group obtained in the obtaining step and subjected to size adjustment processing according to the number of output data for each mini-batch, that is, the time-series data group subjected to branch processing according to processing to be performed by the corresponding network unit, and a connection unit that synthesizes each output data output by processing using the plurality of network units, in a manner such that the synthesis result output by the connection unit approximates information indicating the degree of abnormality obtained when the object is processed in the predetermined processing unit of the manufacturing process, wherein the output data of the same data size is output to the connection unit in a manner such that the data size of the output data from the final layer of each of the plurality of network units corresponds to the data size of each mini-batch.

20. A recording medium in which an abnormality detection program for causing a computer to execute the following steps is recorded: an obtaining step of obtaining a time-series data group measured in association with processing of an object in a predetermined processing unit of a manufacturing process; and a learning step of performing machine learning on a plurality of network units and a connection unit in a learning unit including the plurality of network units that respectively process the time-series data group obtained in the obtaining step and subjected to size adjustment processing according to the number of output data for each mini-batch, that is, the time-series data group subjected to branch processing according to processing to be performed by the corresponding network unit, and the connection unit that synthesizes each output data output by processing using the plurality of network units, in a manner such that the synthesis result output by the connection unit approximates information indicating the degree of abnormality obtained when the object is processed in the predetermined processing unit of the manufacturing process, wherein, wherein the output data of the same data size is output to the connection unit in a manner such that the data size of the output data from the final layer of each of the plurality of network units corresponds to the data size of each mini-batch.

Images