Anomaly detection device, anomaly detection method, and anomaly detection program

Abstract

To execute an abnormality detection processing with high accuracy.SOLUTION: An abnormality detection device includes: an acquisition part that acquires a time series data group measured in accordance with a processing of an object; a plurality of network parts that is the time series data group acquired by the acquisition part and implanting a resize processing in accordance with the number of output data in each mini batch, and processes each of time series data groups branched in accordance with a processing performed in a network part corresponded; and a learning part that generates the plurality of network parts to which a machine learning has been performed and a coupling part to which the machine learning has been performed by performing a machine learning for the coupling part for synthesizing each output data processed by using the plurality of network parts. An output data of the same data size is output to the coupling part so that the data size of the output data from a final layer of each of the plurality of network parts becomes the data size corresponded to the data size of each mini batch.SELECTED DRAWING: Figure 6

Description

[Technical field]

[0001] The present disclosure relates to an anomaly detection device, an anomaly detection method, and an anomaly detection program. [Background technology]

[0002] Conventionally, in the field of various manufacturing processes, anomaly detection techniques have been known that detect anomalies that occur in various manufacturing processes from measurement data (a data set of multiple types of time series data, hereinafter referred to as a time series data group) measured during the processing of an object.

[0003] For example, in a semiconductor manufacturing process, a pre-generated anomaly detection model is used to monitor a time-series data group measured during wafer processing to determine the presence or absence of anomalies, the degree of anomaly, etc. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2011-100211 A Summary of the Invention [Problem to be solved by the invention]

[0005] The present disclosure provides an anomaly detection device, an anomaly detection method, and an anomaly detection program capable of performing highly accurate anomaly detection processing. [Means for solving the problem]

[0006] An anomaly detection device according to an aspect of the present disclosure has, for example, the following configuration. an acquisition unit that acquires a group of time series data measured in association with processing of an object in a predetermined processing unit of a manufacturing process; the time-series data group acquired by the acquisition unit and resized according to the number of output data for each mini-batch, the time-series data group being branched according to processing performed in a corresponding network unit; and a linking unit that synthesizes each output data output by processing using the plurality of network units, and a learning unit that performs machine learning on the plurality of network units and the linking unit to generate the plurality of machine-learned network units and the linking unit so that the synthesis result output by the linking unit approaches information indicating a degree of abnormality acquired when the target object is processed in the predetermined processing unit of the manufacturing process; Output data of the same data size is output to the connection section so that the data size of the output data from the final layer of each of the multiple network sections corresponds to the data size of each mini-batch. Effect of the Invention

[0007] According to the present disclosure, it is possible to provide an anomaly detection device, an anomaly detection method, and an anomaly detection program capable of performing highly accurate anomaly detection processing. [Brief description of the drawings]

[0008] [Figure 1] FIG. 1 is a first diagram showing an example of an overall configuration of a system including a semiconductor manufacturing process and an anomaly detection device. [Diagram 2] FIG. 1 is a first diagram showing an example of a predetermined processing unit in a semiconductor manufacturing process. [Diagram 3] FIG. 2 is a second diagram showing an example of a predetermined processing unit in a semiconductor manufacturing process. [Figure 4] FIG. 2 illustrates an example of a hardware configuration of the anomaly detection device. [Diagram 5] FIG. 1 is a first diagram showing an example of learning data. [Figure 6] FIG. 2 is a diagram illustrating an example of a time-series data group. [Figure 7] FIG. 11 is a first diagram illustrating an example of a functional configuration of a learning unit. [Figure 8] FIG. 11 is a first diagram showing a specific example of processing in a branching section. [Figure 9] FIG. 2 is a second diagram showing a specific example of processing in a branching section. [Figure 10] FIG. 11 is a third diagram showing a specific example of processing at a branching section. [Figure 11] 11A and 11B are diagrams illustrating a specific example of processing by a normalization unit included in each network unit. [Figure 12] FIG. 4 is a fourth diagram showing a specific example of processing at a branching section. [Figure 13] FIG. 11 is a first diagram illustrating an example of a functional configuration of an inference unit. [Figure 14] 1 is a first flowchart showing the flow of an abnormality detection process. [Figure 15] FIG. 2 is a second diagram showing an example of the overall configuration of a system including a semiconductor manufacturing process and an anomaly detection device. [Figure 16] FIG. 2 is a second diagram showing an example of learning data. [Figure 17] FIG. 13 is a diagram showing an example of OES data. [Figure 18] 11A and 11B are diagrams showing a specific example of processing by a normalization unit included in each network unit to which OES data is input. [Figure 19] 11A to 11C are diagrams illustrating specific examples of processing by each normalization unit. [Figure 20] 11A and 11B are diagrams illustrating a specific example of processing by a pooling unit. [Figure 21] FIG. 2 is a second diagram showing an example of the functional configuration of the inference unit. [Figure 22] 11 is a second flowchart showing the flow of the abnormality detection process. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Hereinafter, each embodiment will be described with reference to the accompanying drawings. In this specification and the drawings, components having substantially the same functional configurations are denoted by the same reference numerals, and redundant description will be omitted.

[0010] [First embodiment] <Overall configuration of a system including a semiconductor manufacturing process and an anomaly detection device> First, the overall configuration of a system including a manufacturing process (here, a semiconductor manufacturing process) and an anomaly detection device will be described. Fig. 1 is a first diagram showing an example of the overall configuration of a system including a semiconductor manufacturing process and an anomaly detection device. As shown in Fig. 1, a system 100 includes a semiconductor manufacturing process, time-series data acquisition devices 140_1 to 140_n, and an anomaly detection device 160.

[0011] In the semiconductor manufacturing process, a target object (unprocessed wafer 110) is processed in a predetermined processing unit 120 to generate a result (processed wafer 130). Note that the processing unit 120 here is an abstract concept, and details will be described later. Furthermore, the unprocessed wafer 110 refers to a wafer (substrate) before being processed in the processing unit 120, and the processed wafer 130 refers to a wafer (substrate) after being processed in the processing unit 120.

[0012] The time series data acquiring devices 140_1 to 140_n acquire time series data measured in association with the processing of the unprocessed wafer 110 in the processing unit 120. The time series data acquiring devices 140_1 to 140_n are assumed to perform measurements for different types of measurement items. The number of measurement items measured by each of the time series data acquiring devices 140_1 to 140_n may be one or more. The time series data measured in association with the processing of the unprocessed wafer 110 includes time series data measured during the processing of the unprocessed wafer 110, as well as time series data measured during pre-processing and post-processing performed before and after the processing of the unprocessed wafer 110. These processes may include pre-processing and post-processing performed in a state where there is no wafer (substrate).

[0013] The time-series data group acquired by the time-series data acquiring devices 140_1 to 140_n is stored in the learning data storage unit 163 of the anomaly detection device 160 as learning data (input data).

[0014] In addition, information indicating the degree of abnormality acquired during processing of the unprocessed wafer 110 in the processing unit 120 is associated with a time-series data group and stored in the learning data storage unit 163 of the abnormality detection device 160 as learning data (correct answer data).

[0015] An anomaly detection program is installed in the anomaly detection device 160, and the anomaly detection device 160 functions as a learning unit 161 and an inference unit 162 by executing the program.

[0016] The learning unit 161 performs machine learning using learning data (time-series data groups acquired by the time-series data acquiring devices 140_1 to 140_n and information indicating the degree of abnormality associated with the time-series data groups). Specifically, the learning unit 161 processes the time-series data groups, which are input data, using a plurality of network units, and performs machine learning on the plurality of network units so that a synthesis result of each output data outputted approaches information indicating the degree of abnormality, which is correct data.

[0017] The inference unit 162 inputs a group of time-series data acquired by the time-series data acquiring devices 140_1 to 140_n in association with the processing of a new object (unprocessed wafer) in the processing unit 120 to a plurality of machine-learned network units. As a result, the inference unit 162 infers information indicating the degree of abnormality based on the time-series data acquired in association with the processing of the new unprocessed wafer.

[0018] In this way, in a predetermined processing unit 120 of the semiconductor manufacturing process, by configuring the processing of a time series data group acquired in association with processing of an object using multiple network units, it becomes possible to analyze the time series data group of the predetermined processing unit from multiple angles. As a result, the anomaly detection device 160 makes it possible to generate a model (inference unit 162) that realizes anomaly detection processing with high accuracy, compared to a configuration in which processing is performed using a single network unit.

[0019] <Predetermined processing unit in semiconductor manufacturing process> Next, a predetermined processing unit 120 of a semiconductor manufacturing process will be described. Fig. 2 is a first diagram showing an example of a predetermined processing unit of a semiconductor manufacturing process. As shown in Fig. 2, a semiconductor manufacturing apparatus 200, which is an example of a substrate processing apparatus, has multiple chambers (examples of multiple processing spaces. In the example of Fig. 2, "Chamber A" to "Chamber C"), and wafers are processed in each chamber.

[0020] 2(a) shows a case where multiple chambers are defined as a processing unit 120. In this case, an unprocessed wafer 110 refers to a wafer before being processed in chamber A, and a processed wafer 130 refers to a wafer after being processed in chamber C.

[0021] In the processing unit 120 in FIG. 2(a), the time series data group measured in association with the processing of the unprocessed wafer 110 includes the following: A time series data group outputted in association with the processing of a wafer in chamber A (first processing space); A time series data group outputted in association with the processing of the wafer in chamber B (second processing space); A time series data group outputted in association with the processing of the wafer in chamber C (third processing space); Includes:

[0022] On the other hand, FIG. 2(b) shows a case where one chamber (in the example of FIG. 2(b), "chamber B") is defined as a processing unit 120. In this case, an unprocessed wafer 110 refers to a wafer before being processed in chamber B (a wafer after being processed in chamber A), and a processed wafer 130 refers to a wafer after being processed in chamber B (a wafer before being processed in chamber C).

[0023] 2B, the time series data group measured in association with the processing of the unprocessed wafer 110 includes a time series data group measured in chamber B in association with the processing of the unprocessed wafer 110.

[0024] 3 is a second diagram showing an example of a predetermined processing unit of a semiconductor manufacturing process. As in FIG. 2, the semiconductor manufacturing apparatus 200 has multiple chambers, and in each chamber, wafers are processed by multiple processing operations.

[0025] 3(a) shows a case where the processing content in chamber B excluding pre-processing and post-processing (called "wafer processing") is defined as a processing unit 120. In this case, an unprocessed wafer 110 refers to a wafer before wafer processing (a wafer after pre-processing) and a processed wafer 130 refers to a wafer after wafer processing (a wafer before post-processing).

[0026] In addition, in the processing unit 120 in FIG. 3(a), the time series data group measured in association with the processing of the unprocessed wafer 110 includes a time series data group measured in chamber B in association with the wafer processing of the unprocessed wafer 110.

[0027] 3(a) shows a case where pre-processing, wafer processing (main processing), and post-processing are performed in the same chamber (chamber B), with the wafer processing being treated as processing unit 120. However, when the processes are performed in different chambers, such as when pre-processing is performed in chamber A, wafer processing is performed in chamber B, and post-processing is performed in chamber C, each process in each chamber may be treated as processing unit 120.

[0028] On the other hand, FIG. 3(b) shows a case where, among the processing contents in chamber B, the processing of one process recipe ("process recipe III" in the example of FIG. 3(b)) included in the wafer processing is defined as a processing unit 120. In this case, the unprocessed wafer 110 refers to the wafer before being processed according to process recipe III (the wafer after being processed according to process recipe II). Also, the processed wafer 130 refers to the wafer after being processed according to process recipe III (the wafer before being processed according to process recipe IV (not shown)).

[0029] In addition, in the processing unit 120 in FIG. 3(b), the time series data group measured in association with the processing of the unprocessed wafer 110 includes a time series data group measured in association with the processing in chamber B according to process recipe III.

[0030] <Hardware configuration of anomaly detection device> Next, the hardware configuration of the anomaly detection device 160 will be described. Fig. 4 is a diagram showing an example of the hardware configuration of the anomaly detection device. As shown in Fig. 4, the anomaly detection device 160 has a CPU (Central Processing Unit) 401, a ROM (Read Only Memory) 402, and a RAM (Random Access Memory) 403. The anomaly detection device 160 also has a GPU (Graphics Processing Unit) 404. Note that the processors (processing circuits, processing circuits, processing circuitry) such as the CPU 401 and GPU 404 and the memories such as the ROM 402 and RAM 403 form a so-called computer.

[0031] Furthermore, 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 drive device 409. The hardware components of the abnormality detection device 160 are connected to each other via a bus 410.

[0032] The CPU 401 is a computing device that executes various programs (for example, an anomaly detection program, etc.) installed in the auxiliary storage device 405.

[0033] The ROM 402 is a non-volatile memory and functions as a main storage device. The ROM 402 stores various programs, data, etc. required for the CPU 401 to execute various programs installed in the auxiliary storage device 405. Specifically, the ROM 402 stores boot programs such as a Basic Input / Output System (BIOS) and an Extensible Firmware Interface (EFI).

[0034] The RAM 403 is a volatile memory such as a dynamic random access memory (DRAM) or a static random access memory (SRAM), and functions as a main storage device. The RAM 403 provides a working area in which various programs installed in the auxiliary storage device 405 are expanded when the CPU 401 executes them.

[0035] The GPU 404 is a computing device for image processing, and performs high-speed calculations by parallel processing on various image data (a time-series data group in this embodiment) when the anomaly detection program is executed by the CPU 401. The GPU 404 is equipped with an internal memory (GPU memory) and temporarily stores information required when performing parallel processing on various image data.

[0036] The auxiliary storage device 405 stores various programs and various data used when the various programs are executed by the CPU 401. For example, the learning data storage unit 163 is realized in the auxiliary storage device 405.

[0037] The display device 406 is a display device that displays the internal state of the anomaly detection device 160. The operation device 407 is an input device that is used when an administrator of the anomaly detection device 160 inputs various instructions to the anomaly detection device 160. The I / F device 408 is a connection device that connects to a network (not shown) and performs communication.

[0038] The drive device 409 is a device for setting the recording medium 420. The recording medium 420 here includes media that record information optically, electrically, or magnetically, such as a CD-ROM, a flexible disk, a magneto-optical disk, etc. The recording medium 420 may also include semiconductor memories that record information electrically, such as ROM, flash memory, etc.

[0039] The various programs to be installed in the auxiliary storage device 405 are installed, for example, by setting the distributed recording medium 420 in the drive device 409 and reading out the various programs recorded in the recording medium 420 by the drive device 409. Alternatively, the various programs to be installed in the auxiliary storage device 405 may be installed by being downloaded via a network (not shown).

[0040] <Specific examples of learning data> Next, the learning data read from the learning data storage unit 163 when the learning unit 161 performs machine learning will be described. Fig. 5 is a first diagram showing an example of the learning data. As shown in Fig. 5, the learning data 500 includes information items such as "device", "recipe type", "time series data group", and "degree of abnormality". Note that here, a case will be described in which the predetermined processing unit 120 is processing of one process recipe.

[0041] "Equipment" stores an identifier indicating a semiconductor manufacturing equipment (e.g., semiconductor manufacturing equipment 200) that is to be monitored for the occurrence of an abnormality. "Recipe type" stores an identifier (e.g., process recipe I) indicating the type of process recipe executed when the time series data group was measured, among the process recipes executed in the corresponding semiconductor manufacturing equipment (e.g., EqA).

[0042] The "time series data group" stores a group of time series data measured by the time series data acquisition devices 140_1 to 140_n when the processing of the process recipe identified by the "recipe type" is executed in the semiconductor manufacturing device identified by the "device".

[0043] In the "degree of abnormality", information indicating the degree of abnormality (information obtained by quantifying the degree of abnormality) acquired when the corresponding time series data group (for example, the time series data group 1) is measured by the time series data acquiring devices 140_1 to 140_n is stored. In the example of Fig. 5, the degrees of abnormality 0 to 3 are stored as information indicating the degree of abnormality.

[0044] <Examples of time series data> Next, a specific example of a time series data group measured by the time series data acquiring devices 140_1 to 140_n will be described. Fig. 6 is a diagram showing an example of a time series data group. Note that in the example of Fig. 6, for the sake of simplicity, each of the time series data acquiring devices 140_1 to 140_n is assumed to measure one-dimensional data, but one time series data acquiring device may measure two-dimensional data (a data set of multiple types of one-dimensional data).

[0045] Among these, 6a represents a time series data group when the processing unit 120 is defined as any one of Fig. 2(b), Fig. 3(a), and Fig. 3(b). In this case, the time series data acquiring devices 140_1 to 140_n each acquire time series data measured in association with the processing of the unprocessed wafer 110 in chamber B. Also, the time series data acquiring devices 140_1 to 140_n acquire time series data measured in the same time range as a time series data group.

[0046] 2(a). In this case, the time series data acquiring devices 140_1 to 140_3 acquire a time series data group 1 measured in association with the processing of an unprocessed wafer in, for example, chamber A. The time series data acquiring device 140_n-2 acquires a time series data group 2 measured in association with the processing of the wafer in, for example, chamber B. The time series data acquiring devices 140_n-1 to 140_n acquire a time series data group 3 measured in association with the processing of the wafer in, for example, chamber C.

[0047] In 6a, the time series data acquiring devices 140_1 to 140_n have been described as acquiring, as a time series data group, time series data in the same time range measured in association with the processing of unprocessed wafers in chamber B. However, the time series data acquiring devices 140_1 to 140_n may acquire, as a time series data group, time series data in different time ranges measured in association with the processing of unprocessed wafers in chamber B.

[0048] Specifically, the time series data acquiring devices 140_1 to 140_n may acquire a plurality of pieces of time series data measured during execution of pre-processing as a time series data group 1. Also, the time series data acquiring devices 140_1 to 140_n may acquire a plurality of pieces of time series data measured during execution of wafer processing as a time series data group 2. Furthermore, the time series data acquiring devices 140_1 to 140_n may acquire a plurality of pieces of time series data measured during execution of post-processing as a time series data group 3.

[0049] Similarly, the time series data acquiring devices 140_1 to 140_n may acquire a plurality of pieces of time series data measured during the execution of process recipe I as a time series data group 1. Also, the time series data acquiring devices 140_1 to 140_n may acquire a plurality of pieces of time series data measured during the execution of process recipe II as a time series data group 2. Furthermore, the time series data acquiring devices 140_1 to 140_n may acquire a plurality of pieces of time series data measured during the execution of process recipe III as a time series data group 3.

[0050] <Functional configuration of the learning section> Next, a functional configuration of the learning unit 161 will be described. Fig. 7 is a first diagram showing an example of the functional configuration of the learning unit. The learning unit 161 has a branching unit 710, a first network unit 720_1 to an M-th network unit 720_M, a coupling unit 730, and a comparison unit 740.

[0051] The branching unit 710 is an example of an acquisition unit, and reads out a time series data group from the learning data storage unit 163. In addition, the branching unit 710 processes the read time series data group so that the time series data group is processed by a plurality of network units from a first network unit 720_1 to an Mth network unit 720_M.

[0052] The first network unit 720_1 to the Mth network unit 720_M are configured based on a Convolutional Neural Network (CNN) and have a plurality of layers.

[0053] Specifically, the first network unit 720_1 has a first layer 720_11 to an N-th layer 720_1N. Similarly, the second network unit 720_2 has a first layer 720_21 to an N-th layer 720_2N. The remaining network units have a similar configuration, and the M-th network unit 720_M has a first layer 720_M1 to an N-th layer 720_MN.

[0054] 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, convolution, activation, pooling, etc. Also, in each of the layers of the second network unit 720_2 to the Mth network unit 720_M, similar various processes are performed.

[0055] The connection unit 730 combines the output data from the Nth layer 720_1N of the first network unit 720_1 to the output data from the Nth layer 720_MN of the Mth network unit 720_M, and outputs the combined result to the comparison unit 740.

[0056] The comparison unit 740 compares the synthesis result output from the connection unit 730 with information (correct data) indicating the degree of abnormality read from the learning data storage unit 163, and calculates an error. The learning unit 161 back-propagates the error so that the error calculated by the comparison unit 740 satisfies a predetermined condition, and performs machine learning on the first network unit 720_1 to the Mth network unit 720_M and the connection unit 730.

[0057] As a result, the model parameters of the first to Nth layers of the first network portion 720_1 to the M-th network portion 720_M and the model parameters of the connection portion 730 are optimized.

[0058] <Details of the processing in each part of the learning section> Next, the details of the processing of each part (particularly the branching part here) of learning unit 161 will be described with a concrete example.

[0059] (1) Details of branch processing 1 First, the details of the process of the branching unit 710 will be described. Fig. 8 is a first diagram showing a specific example of the process of the branching unit. In the case of Fig. 8, the branching unit 710 processes the time series data group measured by the time series data acquiring devices 140_1 to 140_n according to a first criterion to generate a time series data group 1 (first time series data group) and inputs it to the first network unit 720_1.

[0060] In addition, the branching unit 710 processes the time series data group measured by the time series data acquiring devices 140_1 to 140_n in accordance with a second criterion to generate a time series data group 2 (a second time series data group) and inputs it to the second network unit 720_2.

[0061] In this way, by performing machine learning after processing the time series data group according to different standards and dividing them into different network parts for processing, it becomes possible to analyze the processing unit 120 from multiple perspectives. As a result, it becomes possible to generate a model (inference unit 162) that achieves high inference accuracy compared to a configuration in which the time series data group is processed using one network part.

[0062] In the example of FIG. 8, two types of time series data groups are generated by processing the time series data groups according to two types of criteria. However, three or more types of time series data groups may be generated by processing the time series data groups according to three or more types of criteria.

[0063] (2) Details of processing by branching section 2 Next, details of other processing of the branching unit 710 will be described. Fig. 9 is a second diagram showing a specific example of processing of the branching unit. In the case of Fig. 9, the branching unit 710 generates a time series data group 1 (first time series data group) and a time series data group 2 (second time series data group) by grouping the time series data groups measured by the time series data acquiring devices 140_1 to 140_n according to the data type. In addition, the branching unit 710 inputs the generated time series data group 1 to the third network unit 720_3, and inputs the generated time series data group 2 to the fourth network unit 720_4.

[0064] In this way, by dividing a time series data group into a plurality of groups according to the data type, and configuring the processing to be performed using different network units, machine learning can be performed, thereby making it possible to analyze the processing unit 120 from multiple perspectives. As a result, it becomes possible to generate a model (inference unit 162) that achieves high inference accuracy, compared to the case where the time series data group is input to one network unit and machine learning is performed.

[0065] 9, the time series data groups are grouped according to the difference in data type based on the difference in the time series data acquiring devices 140_1 to 140_n, but the time series data groups may be grouped according to the time range in which the data was acquired. For example, when the time series data groups are time series data groups (time series data groups 1 to 3) measured in association with the processing according to a plurality of process recipes (process recipes I to III), the time series data groups may be grouped into three according to the time range of each process recipe.

[0066] (3) Details of processing by branching section 3 Next, details of other processing by the branching unit 710 will be described. Fig. 10 is a third diagram showing a specific example of processing by the branching unit. In the case of Fig. 10, the branching unit 710 inputs the time series data group acquired by the time series data acquiring devices 140_1 to 140_n to both the fifth network unit 720_5 and the sixth network unit 720_6. Then, the fifth network unit 720_5 and the sixth network unit 720_6 perform different processing (normalization processing) on ​​the same time series data group.

[0067] Fig. 11 is a diagram showing a specific example of the process of the normalization unit included in each network unit. As shown in Fig. 11, each layer of the fifth network unit 720_5 includes a normalization unit, a convolution unit, an activation function unit, and a pooling unit.

[0068] The example of Figure 11 shows that, among the layers included in the fifth network unit 720_5, the first layer 720_51 includes a normalization unit 1101, a convolution unit 1102, an activation function unit 1103, and a pooling unit 1104.

[0069] Of these, the normalization unit 1101 performs a first normalization process on the time series data group input by the branching unit 710 to generate a normalized time series data group 1 (a first time series data group).

[0070] Similarly, the example of Figure 11 shows that, among the layers included in the sixth network unit 720_6, the first layer 720_61 includes a normalization unit 1111, a convolution unit 1112, an activation function unit 1113, and a pooling unit 1114.

[0071] Of these, the normalization unit 1111 performs a second normalization process on the time series data group input by the branching unit 710 to generate a normalized time series data group 2 (a second time series data group).

[0072] In this way, by configuring a time series data group to be processed using multiple network units each including a normalization unit that performs normalization processing using a different method and then performing machine learning, it becomes possible to analyze the processing unit 120 from multiple perspectives. As a result, it becomes possible to generate a model (inference unit 162) that achieves high inference accuracy, compared to a configuration in which a single normalization process is performed on a time series data group using a single network unit.

[0073] (4) Details of processing by branching section 4 Next, details of other processing by the branching unit 710 will be described. Fig. 12 is a fourth diagram showing a specific example of processing by the branching unit. In the case of Fig. 12, the branching unit 710 inputs a time series data group 1 (a first time series data group) measured in association with the processing of a wafer in chamber A, among the time series data groups measured by the time series data acquiring devices 140_1 to 140_n, to a seventh network unit 720_7.

[0074] In addition, the branching unit 710 inputs, to an eighth network unit 720_8, a time series data group 2 (a second time series data group) measured in association with the processing of the wafer in chamber B, among the time series data groups measured by the time series data acquiring devices 140_1 to 140_n.

[0075] In this way, by configuring the time series data groups measured in association with the processes in different chambers (first process space, second process space) to be processed using different network units and then performing machine learning, it becomes possible to analyze the processing unit 120 from multiple perspectives. As a result, it becomes possible to generate a model (inference unit 162) that achieves high inference accuracy compared to the case where each time series data group is processed using a single network unit.

[0076] <Functional configuration of the inference unit> Next, a functional configuration of the inference unit 162 will be described. Fig. 13 is a first diagram showing an example of the functional configuration of the inference unit. As shown in Fig. 13, the inference unit 162 has a branching unit 1310, a first network unit 1320_1 to an M-th network unit 1320_M, and a connecting unit 1330.

[0077] The branching unit 1310 acquires a group of time series data newly measured by the time series data acquiring devices 140_1 to 140_N. Furthermore, the branching unit 1310 controls the acquired group of time series data to be processed in the first network unit 1320_1 to the M-th network unit 1320_M.

[0078] The first network unit 1320_1 to the Mth network unit 1320_M are formed by machine learning performed by the learning unit 161 and optimizing the model parameters of each layer of the first network unit 720_1 to the Mth network unit 720_M.

[0079] The connection unit 1330 is formed by the connection unit 730 in which machine learning is performed by the learning unit 161 and model parameters are optimized. The connection unit 1330 synthesizes each output data from the output data output from the Nth layer 1320_1N of the first network unit 1320_1 to the output data output from the Nth layer 1320_MN of the Mth network unit 1320_M. As a result, the connection unit 1330 outputs information indicating the degree of abnormality.

[0080] In this way, the inference unit 162 is generated by machine learning performed by the learning unit 161 that analyzes a time-series data group of a predetermined processing unit 120 from various angles. Therefore, the inference unit 162 can be applied to different process recipes, different chambers, and different devices. Alternatively, the inference unit 162 can be applied before and after maintenance of the same chamber. In other words, according to the inference unit 162 according to this embodiment, for example, it is not necessary to maintain or re-learn a model when chamber maintenance is performed, as in the conventional case.

[0081] <Anomaly detection process flow> Next, a description will be given of the overall flow of anomaly detection processing by the anomaly detection device 160. Fig. 14 is a first flowchart showing the flow of the anomaly detection processing.

[0082] In step S1401, the learning unit 161 acquires a time-series data group and information indicating an abnormality degree as learning data.

[0083] In step S1402, the learning unit 161 performs machine learning using the time-series data group from the acquired learning data as input data and the information indicating the degree of abnormality as correct answer data.

[0084] In step S1403, the learning unit 161 determines whether or not to continue machine learning. If further learning data is acquired and machine learning is to be continued (Yes in step S1403), the process returns to step S1401. On the other hand, if machine learning is to be ended (No in step S1403), the process proceeds to step S1404.

[0085] In step S1404, the inference unit 162 generates the first network unit 1320_1 to the M-th network unit 1320_2 by reflecting the model parameters optimized by machine learning.

[0086] In step S1405, the inference unit 162 infers information indicating the degree of abnormality by inputting a group of time-series data measured in association with the processing of a new unprocessed wafer.

[0087] In step S1206, the inference unit 162 outputs the inference result together with an identifier indicating the corresponding semiconductor manufacturing device, an identifier indicating the type of the corresponding process recipe, and the like.

[0088] <Summary> As is clear from the above description, the anomaly detection device according to the first embodiment has At a specific processing unit in the manufacturing process, a time series data set measured in conjunction with the processing of an object is acquired. The acquired time series data group, - processing the data in accordance with first and second criteria to generate a first set of time series data and a second set of time series data; or - Grouping according to data type or time range, The output data output by processing using a plurality of network units is synthesized, or The acquired time series data is input to multiple network sections that perform normalization using different methods, and the output data output by processing using the multiple network sections is then synthesized. Machine learning is performed on multiple network components so that the composite result obtained by combining each output data approaches the information indicating the degree of anomaly obtained when processing the target object at a specified processing unit in the manufacturing process. A group of time series data measured by a time series data acquisition device in association with the processing of a new object is processed using multiple machine learning network units, and information indicating the degree of anomaly is inferred as a composite result of each output data output from the multiple network units.

[0089] In this way, by configuring the time series data group to be input to multiple network units and machine learning to be performed, it becomes possible to analyze the time series data group of a predetermined processing unit of the semiconductor manufacturing process from multiple angles. As a result, it becomes possible to generate a model that realizes anomaly detection processing with high accuracy compared to the case where the time series data group is input to one network unit and machine learning is performed.

[0090] That is, according to the first embodiment, it is possible to provide an anomaly detection device capable of performing an anomaly detection process with high accuracy.

[0091] [Second embodiment] In the anomaly detection device 160 according to the first embodiment, four configurations have been shown as configurations for processing acquired time series data groups using multiple network units. In contrast, in the second embodiment, of these four configurations, a configuration for processing time series data groups using multiple network units each including a normalization unit that performs normalization processing using a different method will be described in more detail. In the description, The time series data acquisition device is an optical emission spectrometer. The time-series data group is OES (Optical Emission Spectroscopy) data (a data set that contains time-series data of emission intensity according to the number of wavelengths). The second embodiment will be described below, focusing on the differences from the first embodiment.

[0092] <Overall configuration of a system including a semiconductor manufacturing process and an anomaly detection device> First, the overall configuration of a system including a semiconductor manufacturing process and an anomaly detection device when the time-series data acquisition device is an optical emission spectrometer will be described. Fig. 15 is a second diagram showing an example of the overall configuration of a system including a semiconductor manufacturing process and an anomaly detection device. As shown in Fig. 15, the system 1500 has a semiconductor manufacturing process, an optical emission spectrometer 1501, and an anomaly detection device 160.

[0093] In the system 1500 shown in FIG. 15, the emission spectroscopic analyzer 1501 measures OES data, which is a time-series data group, in accordance with the processing of the pre-process wafer 110 in the processing unit 120 by emission spectroscopic analysis technology. A part of the OES data measured by the emission spectroscopic analyzer 1501 is stored in the learning data storage unit 163 of the anomaly detection device 160 as learning data (input data) when performing machine learning.

[0094] <Specific Example of Learning Data> Next, the learning data read from the learning data storage unit 163 when the learning unit 161 performs machine learning will be described. FIG. 16 is a second diagram showing an example of the learning data. As shown in FIG. 16, the learning data 1600 includes information items similar to those of the learning data 500 shown in FIG. 5 as information items. The difference from FIG. 5 is that "OES data" is included as an information item instead of "time-series data group", and the OES data measured by the emission spectroscopic analyzer 1501 is stored.

[0095] <Specific Example of OES Data> Next, a specific example of the OES data measured by the emission spectroscopic analyzer 1501 will be described. FIG. 17 is a diagram showing an example of the OES data.

[0096] In FIG. 17, the graph 1710 is a graph representing the characteristics of the OES data, which is a time-series data group measured by the emission spectroscopic analyzer 1501. The horizontal axis indicates the wafer identification number for identifying each pre-process wafer 110 processed in the processing unit 120. The vertical axis indicates the time length of the OES data measured by the emission spectroscopic analyzer 1501 in accordance with the processing of each pre-process wafer 110.

[0097] As shown in the graph 1710, the OES data measured by the emission spectroscopic analyzer 1501 has different time lengths for each processed wafer.

[0098] 17, for example, OES data 1720 indicates OES data measured in association with processing of an unprocessed wafer with wafer identification number="770". The vertical data size of OES data 1720 depends on the range of wavelengths measured by optical emission spectrometer 1501. In the second embodiment, optical emission spectrometer 1501 measures the emission intensity in a predetermined wavelength range, and therefore the vertical data size of OES data 1720 is, for example, the number of wavelengths "Nλ" included in the predetermined wavelength range.

[0099] On the other hand, the data size in the horizontal direction of the OES data 1720 depends on the time length when it is measured in the optical emission spectrometer 1501. In the example of Fig. 17, the data size in the horizontal direction of the OES data 1720 is "LT".

[0100] In this way, the OES data 1720 can be said to be a time-series data group in which one-dimensional time-series data having a predetermined time length for each wavelength is collected for a predetermined number of wavelengths.

[0101] In addition, when the branching section 710 inputs the OES data 1720 to the fifth network section 720_5 and the sixth network section 720_6, resizing processing is performed on a mini-batch basis so that the data size is the same as that of the OES data of other wafer identification numbers.

[0102] <Specific example of normalization process> Next, a specific example of the process of the normalization unit of the fifth network unit 720_5 and the sixth network unit 720_6 to which the OES data 1720 is input by the branch unit 710 will be described.

[0103] Fig. 18 is a diagram showing a specific example of processing by a normalization unit included in each network unit to which OES data is input. As shown in Fig. 18, among the layers included in the fifth network unit 720_5, the first layer 720_51 has a normalization unit 1101. In addition, the normalization unit 1101 normalizes the OES data 1720 by a first method (average and standard deviation of emission intensity over all wavelengths) to generate normalized data (normalized OES data 1810).

[0104] 18, among the layers included in the sixth network portion 720_6, the first layer 720_61 has a normalization unit 1111. The normalization unit 1111 normalizes the OES data 1720 by a second method (average and standard deviation of emission intensity for each wavelength) to generate normalized data (normalized OES data 1820).

[0105] 19A and 19B are diagrams showing specific examples of the processing of each normalization unit. As shown in 19A, the normalization unit 1101 normalizes the emission intensity for all wavelengths using the average and standard deviation of the emission intensity. On the other hand, the normalization unit 1111 normalizes the emission intensity for each wavelength using the average and standard deviation of the emission intensity.

[0106] In this way, depending on what is used as a criterion for viewing changes in emission intensity (i.e., depending on the method of analysis), the information that is seen will change even for the same OES data 1720. In the anomaly detection device 160 according to the second embodiment, the same OES data 1720 is processed using different network units for different normalization processes. By combining multiple normalization processes in this way, it becomes possible to analyze the OES data 1720 in the processing unit 120 from multiple perspectives. As a result, it becomes possible to generate a model (inference unit 162) that achieves high inference accuracy, compared to a configuration in which one normalization process is performed on the OES data 1720 using one network unit.

[0107] In the above specific example, the case where normalization is performed using the average and standard deviation of the emission intensity has been described, but the statistical value used for normalization is not limited to this. For example, normalization may be performed using the maximum value and standard deviation of the emission intensity, or other statistical value may be used. Also, a configuration may be made in which it is possible to select which statistical value is used for normalization.

[0108] <Specific example of pooling process> Next, a specific example of the process of the pooling unit included in the final layer of the fifth network unit 720_5 and the sixth network unit 720_6 will be described. Fig. 20 is a diagram showing a specific example of the process of the pooling unit.

[0109] Since the data size differs between mini-batches, the pooling units 1104, 1114 included in the final layers of the fifth network unit 720_5 and the sixth network unit 720_6 perform pooling processing so that fixed-length data is output between mini-batches.

[0110] 20 is a diagram showing a specific example of the process of the pooling unit. As shown in FIG. 20, the pooling units 1104 and 1114 perform GAP (Global Average Pooling) processing on the feature data output from the activation function units 1103 and 1113.

[0111] 20, feature data 2011_1 to 2011_m indicate feature data input to the pooling unit 1104 of the N-th layer 720_5N of the fifth network unit 720_5, and generated based on OES data belonging to mini-batch 1. Each of the feature data 2011_1 to 2011_m indicates feature data for one channel.

[0112] Moreover, feature data 2012_1 to 2012_m indicate feature data that is input to the pooling unit 1104 of the N-th layer 720_5N of the fifth network unit 720_5 and that is generated based on the OES data belonging to mini-batch 2. Each of the feature data 2012_1 to 2012_m indicates feature data for one channel.

[0113] Moreover, the feature data 2031_1 to 2031_m and 2032_1 to 2032_m are similar to the feature data 2011_1 to 2011_m and 2012_1 to 2012_m (however, each is feature data for Nλ channels).

[0114] Here, the pooling units 1104 and 1114 output fixed-length output data by calculating the average value for each channel for each feature value included in the input feature data. This allows the data output from the pooling units 1104 and 1114 to have the same data size between mini-batches.

[0115] <Functional configuration of the inference unit> Next, a functional configuration of the inference unit 162 will be described. Fig. 21 is a second diagram showing an example of the functional configuration of the inference unit. As shown in Fig. 21, the inference unit 162 has a branching unit 1310, a fifth network unit 1320_5 and a sixth network unit 1320_6, and a coupling unit 1330.

[0116] The branching unit 1310 acquires newly measured OES data from the optical emission spectrometer 1501. In addition, the branching unit 1310 branches the acquired OES data so that the acquired OES data is processed in a fifth network unit 1320_5 and a sixth network unit 1320_6.

[0117] The fifth network part 1320_5 and the sixth network part 1320_6 are formed by machine learning performed by the learning part 161 and optimizing the model parameters of each layer of the fifth network part 1320_5 and the sixth network part 1320_6.

[0118] The connection unit 1330 is formed by the connection unit 1330 in which machine learning is performed by the learning unit 161 and model parameters are optimized. The connection unit 1330 combines the output data output from the N-th layer 1320_5N of the fifth network unit 1320_5 and the output data output from the N-th layer 1320_6N of the sixth network unit 1320_6, and outputs information indicating the degree of abnormality.

[0119] In this way, the inference unit 162 is generated by machine learning performed by the learning unit 161 that analyzes the OES data of a predetermined processing unit 120 from various angles. Therefore, the inference unit 162 can be applied to different recipes, different chambers, and different devices. Alternatively, the inference unit 162 can be applied before and after maintenance of the same chamber. In other words, according to the inference unit 162 according to this embodiment, for example, it is not necessary to maintain or re-learn a model when chamber maintenance is performed, as in the conventional case.

[0120] <Anomaly detection process flow> Next, a description will be given of the overall flow of the anomaly detection process by the anomaly detection device 160. Fig. 22 is a second flowchart showing the flow of the anomaly detection process.

[0121] In step S2201, the learning unit 161 acquires OES data and information indicating the degree of abnormality as learning data.

[0122] In step S2202, the learning unit 161 performs machine learning using the OES data from the acquired learning data as input data and the information indicating the degree of abnormality as correct answer data.

[0123] In step S2203, the learning unit 161 determines whether to continue machine learning. If further learning data is acquired and machine learning is to be continued (Yes in step S2203), the process returns to step S2201. On the other hand, if machine learning is to be ended (No in step S2203), the process proceeds to step S2204.

[0124] In step S2204, the inference unit 162 generates a fifth network part 1320_5 and a sixth network part 1320_6 by reflecting the model parameters optimized by machine learning.

[0125] In step S2205, the inference unit 162 infers information indicating the degree of abnormality by inputting OES data measured by the optical emission spectrometer 1501 in association with the processing of a new unprocessed wafer.

[0126] In step S2206, the inference unit 162 outputs the inference result together with an identifier indicating the corresponding semiconductor manufacturing device, an identifier indicating the corresponding type of recipe, and the like.

[0127] <Summary> As is clear from the above description, the anomaly detection device according to the second embodiment has At a specific processing unit in the manufacturing process, OES data measured by an optical emission spectrometer is acquired as the target object is processed. The acquired OES data is input to two networks that perform normalization using different methods, and the output data from the two networks is then synthesized. Machine learning is performed on the two network components so that the composite result obtained by combining the output data approaches the information indicating the degree of anomaly obtained when processing the target object at a specified processing unit in the manufacturing process. The OES data measured by the optical emission spectrometer when processing a new object is processed using two networks that have undergone machine learning, and information indicating the degree of anomaly is inferred as a composite result of the output data from the two networks.

[0128] In this way, by configuring the OES data to be input to two network units and machine learning to be performed, it becomes possible to perform multifaceted analysis of the OES data for a given processing unit of the semiconductor manufacturing process. As a result, it becomes possible to generate a model that realizes anomaly detection processing with high accuracy, compared to the case where the OES data is input to one network unit and machine learning is performed.

[0129] That is, according to the second embodiment, it is possible to provide an anomaly detection device capable of performing an anomaly detection process with high accuracy.

[0130] [Third embodiment] In the above second embodiment, an optical emission spectrometer is given as an example of a time-series data acquiring device, but the time-series data acquiring device described in the first embodiment is not limited to an optical emission spectrometer.

[0131] For example, the time series data acquiring device described in the first embodiment may include a process data acquiring device that acquires various process data such as temperature data, pressure data, gas flow rate data, etc. as one-dimensional time series data. Alternatively, the time series data acquiring device described in the first embodiment may include a plasma high frequency power supply device that acquires various RF data such as RF power supply voltage data, etc. as one-dimensional time series data.

[0132] Although the types of abnormalities are not specifically mentioned in the above first and second embodiments, the types of abnormalities that occur in the processing unit 120 include any types of abnormalities. Note that the types of abnormalities that occur in the processing unit 120 include, for example, abnormalities that occur due to changes over time and abnormalities that occur suddenly.

[0133] Among these, abnormalities caused by changes over time include abnormalities occurring in the resultant product, such as the film thickness of the wafer and the etching rate. Furthermore, abnormalities caused by changes over time include the changes over time themselves of the parts in the apparatus and the devices connected to the apparatus (wear and tear of parts, wear of electrodes, deterioration of devices, deposition of films on parts in the apparatus, etc.). Furthermore, abnormalities caused by changes over time include abnormalities caused by the changes over time of the parts in the apparatus and the devices connected to the apparatus (fluctuations in gas flow rate, temperature abnormalities, etc.).

[0134] On the other hand, sudden abnormalities include, for example, abnormal discharge, droplets (an event in which a large number of micron-sized particles adhere), unstable discharge behavior, air leaks or helium leaks, and the like.

[0135] In the above first and second embodiments, information indicating the degree of abnormality is stored as learning data (correct answer data). However, instead of the information indicating the degree of abnormality, the information indicating the degree of abnormality may be compared with a predetermined threshold value, and the comparison result (presence or absence of abnormality) may be stored as learning data (correct answer data). In this case, the learning unit 161 performs machine learning so as to match the presence or absence of abnormality stored in the learning data, and the inference unit 162 infers the presence or absence of abnormality.

[0136] Alternatively, the learning unit 161 may be configured to perform machine learning similar to that of the first and second embodiments described above, and the inference unit 162 may convert the information indicating the degree of anomaly output from the connection unit 730 into information indicating the presence or absence of an anomaly, and then output this as an inference result.

[0137] In converting the information indicating the degree of abnormality into information indicating the presence or absence of an abnormality, for example, a sigmoid function may be used. Specifically, in the inference unit 162, if the output when the information indicating the degree of abnormality is input to the sigmoid function is 0.5 or more, the inference unit 162 outputs "abnormal" as information indicating the presence or absence of an abnormality. On the other hand, if the output when the information indicating the degree of abnormality is input to the sigmoid function is less than 0.5, the inference unit 162 outputs "no abnormality" as information indicating the presence or absence of an abnormality.

[0138] In the above first and second embodiments, the learning unit has been described as performing machine learning using the same learning data regardless of the type of anomaly. However, the method of machine learning by the learning unit is not limited to this, and the learning unit may be configured to perform machine learning using different learning data depending on the type of anomaly. Specifically, for example, A plurality of network units that perform machine learning so that, when an abnormal discharge occurs, information indicating the presence of an abnormality is output; A plurality of network units that perform machine learning so that, when a droplet is generated, information indicating an abnormality is output; Alternatively, the machine learning device 100 may be configured to perform machine learning using separate learning data.

[0139] Furthermore, in the anomaly detection device 160 according to the first and second embodiments, it has been described that machine learning is performed by the learning unit so that information indicating the degree of anomaly is output regardless of the type of anomaly that has occurred.

[0140] However, the method of machine learning by the learning unit is not limited to this, and the machine learning may be configured to output information indicating the type of anomaly in addition to information indicating the degree of anomaly. Specifically, the linking unit may be provided with output units in a number corresponding to the number of types of anomaly, and configured to output a confidence level that the degree of anomaly has occurred for each type.

[0141] [Other embodiments] In the above first to third embodiments, the machine learning algorithm of each network part of the learning unit 161 has been described as being configured based on a convolutional neural network. However, the machine learning algorithm of each network part of the learning unit 161 is not limited to a convolutional neural network, and may be configured based on another machine learning algorithm.

[0142] In the above first to third embodiments, the anomaly detection device 160 has been described as functioning as the learning unit 161 and the inference unit 162. However, the device functioning as the learning unit 161 and the device functioning as the inference unit 162 do not need to be integrated and may be configured separately. In other words, the anomaly detection device 160 may function as the learning unit 161 without the inference unit 162, or may function as the inference unit 162 without the learning unit 161.

[0143] The present invention is not limited to the configurations shown in the above embodiments, combinations with other elements, etc. These points can be changed without departing from the spirit of the present invention, and can be appropriately determined according to the application form. [Explanation of symbols]

[0144] 100: System 110: Unprocessed wafer 120: Processing unit 130: Processed wafer 140_1~140_n: Time series data acquisition device 160: Anomaly detection device 161: Learning Department 162: Reasoning part 200: Semiconductor manufacturing equipment 500: Training data 710: Branch 720_1 to 720_M: 1st network section to Mth network section 720_11~720_1N: 1st layer ~ Nth layer 720_21~720_2N: 1st layer ~ Nth layer 720_M1~720_MN: 1st layer ~ Nth layer 730:Connection part 740: Comparison section 1101, 1111: Normalization section 1104, 1114: Pooling section 1310: Branch 1320_1~1320_M: 1st network section~Mth network section 1320_11~1320_1N: 1st layer ~ Nth layer 1320_21~1320_2N: 1st layer ~ Nth layer 1320_M1~1320_MN: 1st layer ~ Nth layer 1330 :Connection part

Claims

1. an acquisition unit that acquires a group of time series data measured in association with processing of an object in a predetermined processing unit of a manufacturing process; the time-series data group acquired by the acquisition unit and resized according to the number of output data for each mini-batch, the time-series data group being branched according to processing performed in a corresponding network unit; and a linking unit that synthesizes each output data output by processing using the plurality of network units, and a learning unit that performs machine learning on the plurality of network units and the linking unit to generate the plurality of machine-learned network units and the linking unit so that the synthesis result output by the linking unit approaches information indicating a degree of abnormality acquired when the target object is processed in the predetermined processing unit of the manufacturing process; An anomaly detection device that outputs output data of the same data size to the connection unit so that the data size of the output data from the final layer of each of the multiple network units corresponds to the data size of each mini-batch.

2. 2. The anomaly detection device according to claim 1, further comprising an inference unit that infers, as information indicating a degree of anomaly when the new object is processed, a synthesis result synthesized by the machine-learned connection unit, the synthesis result being a time-series data group obtained for a new object and resized according to the number of output data for each mini-batch, the time-series data group being branched according to processing performed in a corresponding machine-learned network unit, and each output data being output by processing the time-series data group branched according to processing performed in a corresponding machine-learned network unit, using the plurality of machine-learned network units.

3. The anomaly detection device according to claim 2 , wherein the inference unit outputs information indicating the presence or absence of an anomaly based on the information indicating the inferred degree of anomaly.

4. an acquisition unit that acquires a group of time series data measured in association with processing of an object in a predetermined processing unit of a manufacturing process; the time-series data group acquired by the acquisition unit and resized according to the number of output data for each mini-batch, the time-series data group being acquired by the acquisition unit, the time-series data group being resized according to the number of output data for each mini-batch, the time-series data group being resized according to the processing performed in the corresponding network ... an inference unit that infers a synthesis result synthesized by the machine-learned connection unit as information indicating a degree of anomaly when the new object is processed, the synthesis result being output data that is output by processing each of the time-series data groups branched according to processing performed in a corresponding machine-learned network unit, the time-series data groups being acquired for a new object and resized according to the number of output data for each mini-batch, the synthesis result being output data that is output by processing each of the time-series data groups branched according to processing performed in a corresponding machine-learned network unit, the synthesis result being output data that is output by the machine-learned connection unit, An anomaly detection device that outputs output data of the same data size to the machine-learned connection part so that the data size of the output data from the final layer of each of the multiple machine-learned network parts corresponds to the data size of each mini-batch during machine learning.

5. The learning unit is 2. The anomaly detection device according to claim 1, further comprising: processing the time series data group acquired by the acquisition unit and resized according to a number of output data for each mini-batch according to a first criterion and a second criterion, respectively, to generate a first time series data group and a second time series data group; and performing machine learning on the different network units and the connection units so that a composite result of each of the output data output by processing the generated first time series data group and the generated second time series data group using a different network unit approaches information indicating a degree of anomaly acquired when the target object is processed in the predetermined processing unit of the manufacturing process.

6. 6. The anomaly detection device according to claim 5, further comprising an inference unit that generates a first time series data group and a second time series data group by processing a time series data group obtained for a new object and resized for each mini-batch according to the number of output data, according to the first criterion and the second criterion, respectively, and infers a synthesis result synthesized by the machine-learned connection unit, which is each of the output data output by processing using the different network units that have been machine-learned, as information indicating a degree of anomaly when the new object is processed.

7. The learning unit is 2. The anomaly detection device according to claim 1, further comprising: grouping the time series data group acquired by the acquisition unit and resized according to the number of output data for each mini-batch according to a data type or a time range; processing each group using a different network unit; and performing machine learning on the different network units and the connection units so that a composite result of each output data outputted by processing the group approaches information indicating a degree of anomaly acquired when the target object is processed in the predetermined processing unit of the manufacturing process.

8. 8. The anomaly detection device according to claim 7, further comprising an inference unit that infers, as information indicating a degree of anomaly when the new object is processed, a synthesis result obtained by the machine-learned connection unit, which is output data obtained by processing each group using the different network units that have been machine-learned, and which is obtained by grouping a time-series data group that has been obtained for a new object and resized according to the number of output data for each mini-batch according to the data type or the time range, and which outputs each output data by processing each group using the different network units that have been machine-learned.

9. The learning unit is 2. The anomaly detection device according to claim 1, wherein the time series data group acquired by the acquisition unit and resized according to the number of output data for each mini-batch is input to different network units each including a normalization unit that performs normalization by a different method, and machine learning is performed on the different network units and the connection unit so that a composite result of each output data output by processing using the different network units approaches information indicating a degree of anomaly acquired when the target object is processed in the manufacturing process.

10. The anomaly detection device according to claim 9, further comprising an inference unit that infers, as information indicating a degree of anomaly when the new object is processed, a synthesis result synthesized by the machine-learned connection unit, which is each output data output by processing a group of time-series data acquired for a new object and resized according to the number of output data for each mini-batch, inputted to the different network units that have been machine-learned, and the synthesis result synthesized by the machine-learned connection unit.

11. The learning unit is 2. The anomaly detection device according to claim 1, further comprising: a first time series data group measured in association with the processing of the object in a first processing space included in the predetermined processing unit and resized according to the number of output data for each mini-batch; and a second time series data group measured in association with the processing of the object in a second processing space and resized according to the number of output data for each mini-batch, the first time series data group being measured in association with the processing of the object in a first processing space included in the predetermined processing unit, the second time series data group being measured in association with the processing of the object in a second processing space, the second time series data group being resized according to the number of output data for each mini-batch. The anomaly detection device according to claim 1, further comprising:

12. 12. The anomaly detection device according to claim 11, further comprising an inference unit that infers, as information indicating a degree of anomaly when the new object is processed, each of output data output by processing, using the different network unit that has been machine-learned, a first time-series data group measured in connection with processing in the first processing space included in the specified processing unit and resized according to the number of output data for each mini-batch, and a second time-series data group measured in connection with processing in the second processing space and resized according to the number of output data for each mini-batch, the combined result being combined by the connection unit that has been machine-learned.

13. The anomaly detection device according to claim 1 , wherein the time series data group is data measured in association with processing in a substrate processing apparatus.

14. The learning unit is 2. The anomaly detection device according to claim 1, further comprising: a plurality of network units that processes the time series data group acquired by the acquisition unit and resized according to the number of output data for each mini-batch, and the plurality of network units processes the time series data group to obtain a composite result of each output data outputted from the processing of the time series data group. The plurality of network units and the connection unit perform machine learning so that the composite result of each output data outputted from the processing of the time series data group acquired by the acquisition unit and resized according to the number of output data for each mini-batch approaches information indicating a degree of anomaly and information indicating a type of anomaly acquired when the target object is processed in the predetermined processing unit of the manufacturing process.

15. 10. The anomaly detection device according to claim 9, wherein the time series data group is data measured by an optical emission spectrometer in association with processing in a substrate processing device, and is data indicating optical emission intensity at each wavelength measured at each time.

16. The anomaly detection device according to claim 15 , wherein the normalization unit included in a first network unit of the different networks performs normalization using a statistical value of emission intensity for all wavelengths.

17. The anomaly detection device according to claim 15 , wherein the normalization unit included in a second network unit of the different networks performs normalization using a statistical value of emission intensity for each wavelength.

18. an acquiring step of acquiring a group of time series data measured in association with processing of an object in a predetermined processing unit of the manufacturing process; a learning step of generating the plurality of machine-learned network parts and the machine-learned connection parts by performing machine learning on the plurality of network parts and the connection parts so that the synthesis result output by the connection parts approaches information indicating an anomaly level obtained when the target object is processed in the predetermined processing unit of the manufacturing process, the learning step being executed by a computer, the plurality of network parts including a plurality of network parts processing each of the time-series data groups branched according to processing performed in a corresponding network part, the time-series data groups being acquired in the acquisition step and resized according to the number of output data for each mini-batch, the plurality of network parts processing each of the time-series data groups branched according to processing performed in a corresponding network part, and a connection part synthesizing each of the output data output by processing using the plurality of network parts, the plurality of network parts and the connection parts being machine-learned so that the synthesis result output by the connection part approaches information indicating an anomaly level obtained when the target object is processed in the predetermined processing unit of the manufacturing process, An anomaly detection method in which output data of the same data size is output to the connection section so that the data size of the output data from the final layer of each of the multiple network sections corresponds to the data size of each mini-batch.

19. an acquiring step of acquiring a group of time series data measured in association with processing of an object in a predetermined processing unit of the manufacturing process; a learning step of generating the plurality of machine-learned network parts and the machine-learned connection parts by performing machine learning on the plurality of network parts and the connection parts so that the synthesis result output by the connection parts approaches information indicating a degree of anomaly obtained when the target object is processed in the predetermined processing unit of the manufacturing process, the learning step including a plurality of network parts processing each of the time-series data groups branched according to processing performed in a corresponding network part, the time-series data groups being acquired in the acquisition step and resized according to the number of output data for each mini-batch, the plurality of network parts processing each of the time-series data groups branched according to processing performed in a corresponding network part, and a connection part synthesizing each of the output data output by processing using the plurality of network parts, the plurality of network parts and the connection parts being machine-learned so that the synthesis result output by the connection part approaches information indicating a degree of anomaly obtained when the target object is processed in the predetermined processing unit of the manufacturing process, An anomaly detection program that outputs output data of the same data size to the connection unit so that the data size of the output data from the final layer of each of the multiple network units corresponds to the data size of each mini-batch.

Images