Data processing device, data processing system, data processing method, and data processing program

Abstract

The present application provides a data processing apparatus, a data processing system, a data processing method, and a data processing program that suppress the lack of feature data and compress and visualize time-series data of multiple wavelengths. The data processing apparatus has a preprocessing section that generates normalized data by normalizing time-series data of multiple wavelengths using prescribed reference data; an extraction section that divides the normalized data into a plurality of regions for each prescribed time range and prescribed wavelength range, and extracts outliers in each region as representative values; and a generation section that converts the representative values of each region into color data, and generates image data.

Description

Technical Field

[0001] This disclosure relates to data processing apparatus, data processing system, data processing method, and data processing program. Background Technology

[0002] In semiconductor manufacturing processes, measurement data, such as that obtained by light emission spectrophotometers and quality analyzers (so-called multi-wavelength time series data), typically has high resolution and large data volume. Therefore, managing this data is very costly. Furthermore, with multi-wavelength time series data, it is difficult for users to directly determine the presence or absence of anomalies.

[0003] In contrast, for example, if multi-wavelength time series data is visualized and displayed, it is believed that users can easily determine whether anomalies have occurred. Furthermore, if the visualized data is compressed for management, it is believed that management costs will also be reduced.

[0004] However, in the case of multi-wavelength time series data, when displaying the visualized data, or when compressing the visualized data, there may be a lack of feature data indicating the occurrence of anomalies.

[0005] Patent Document 1: Japanese Patent Application Publication No. 2020-65013

[0006] Patent Document 2: Japanese Patent Application Publication No. 2005-217448

[0007] Patent Document 3: Japanese Patent No. 5315025 Summary of the Invention

[0008] This disclosure provides a data processing apparatus, data processing system, data processing method, and data processing program for compressing and visualizing multi-wavelength time series data to suppress the lack of feature data.

[0009] A data processing apparatus according to one aspect of this disclosure has, for example, the following structure: That is, it has:

[0010] The preprocessing unit uses specified reference data to normalize the multi-wavelength time series data to generate normalized data;

[0011] The extraction unit divides the normalized data into multiple regions according to each specified time range and wavelength range, and extracts outliers from each region as representative values; and

[0012] The generation unit converts the representative values ​​of the above regions into color data and generates image data.

[0013] According to this disclosure, there are data processing apparatuses, data processing systems, data processing methods, and data processing programs that can compress and visualize multi-wavelength time series data to suppress the lack of feature data. Attached Figure Description

[0014] Figure 1 The first diagram is an example of the system architecture of a data processing system.

[0015] Figure 2 This is a diagram illustrating an example of a semiconductor manufacturing process.

[0016] Figure 3 This is a diagram illustrating an example of the hardware structure of a data processing device.

[0017] Figure 4 This is a diagram representing an example of OES data.

[0018] Figure 5 This is a diagram illustrating a specific example of the preprocessing steps.

[0019] Figure 6 This is a diagram illustrating a specific example of the compression process.

[0020] Figure 7 It is a diagram representing a specific example of image data.

[0021] Figure 8 It is a flowchart representing the process of image processing.

[0022] Figure 9A The second figure is an example of the system architecture of a data processing system.

[0023] Figure 9B The third figure is an example of the system architecture of a data processing system.

[0024] Figure 10 This is a graph representing an example of data used in learning.

[0025] Figure 11 This is a diagram illustrating a specific example of the learning process handled by the learning department.

[0026] Figure 12 This is a diagram illustrating a specific example of the inference process of the inference unit.

[0027] Figure 13 This is a flowchart illustrating the process of anomaly detection and handling.

[0028] Figure 14 The fourth figure is an example of the system architecture of a data processing system. Detailed Implementation

[0029] Hereinafter, various embodiments will be described with reference to the accompanying drawings. Furthermore, in this specification and the accompanying drawings, redundant descriptions of constituent elements having substantially the same functional structure are omitted by using the same reference numerals.

[0030] [First Implementation Method]

[0031] <System Architecture of Data Processing System>

[0032] First, the system structure of the data processing system involved in the first embodiment will be described. Figure 1 This is the first diagram, representing an example of the system architecture of a data processing system. (See diagram for example.) Figure 1 As shown, the data processing system 100 includes a semiconductor manufacturing process, a light emission spectrophotometer 140, and a data processing device 150.

[0033] In a semiconductor manufacturing process, within a defined processing space 120, the object to be processed (pre-processed wafer 110) is processed to generate the result (post-processed wafer 130). Furthermore, the pre-processed wafer 110 refers to the wafer (substrate) before processing in the processing space 120, and the post-processed wafer 130 refers to the wafer (substrate) after processing in the processing space 120.

[0034] The light emission spectrophotometer 140 measures OES (Optical Emission Spectroscopy) data in the processing space 120, accompanying the processing of the pre-processing wafer 110. OES data is a "multi-wavelength time series data" containing light emission intensity data at various times, corresponding to the number of wavelengths measured.

[0035] The OES data measured by the luminescence spectrophotometer 140 is stored in the OES data storage unit 155 of the data processing unit 150.

[0036] A data processing program is installed in the data processing device 150. By executing the program, the data processing device 150 functions as a preprocessing unit 151, a compression unit 152, an image processing unit 153, and a display control unit 154.

[0037] The preprocessing unit 151 reads OES data from the OES data storage unit 155 and performs preprocessing (e.g., normalization) using predetermined reference data. Additionally, the preprocessing unit 151 notifies the compression unit 152 of the preprocessed OES data.

[0038] The compression unit 152 compresses the preprocessed OES data and notifies the imaging unit 153 of the compressed OES data. Specifically, the compression unit 152 compresses the preprocessed OES data by dividing it into multiple regions of a predetermined size (a predetermined time range and a predetermined wavelength range) and extracting outliers from each region as representative values ​​for that region. Outliers are values ​​that statistically deviate significantly from other values ​​in each region. By compressing the preprocessed OES data while retaining the outliers in each region, it is possible to suppress the loss of feature data (feature data indicating anomalies) contained in the OES data.

[0039] The image processing unit 153 generates image data by converting the compressed OES data received from the compression unit 152 into color data. Furthermore, the image processing unit 153 stores the generated image data in the image data storage unit 156 and notifies the display control unit 154. Thus, storing image data in the image data storage unit 156 significantly reduces the amount of data compared to OES data, thereby reducing management costs.

[0040] The display control unit 154 controls the display to show the image data generated by the image processing unit 153 on a display device (not shown). As described above, since the image data generated by the image processing unit 153 is compressed, it can be displayed on the display device without further processing, regardless of whether the number of pixels on the display device is limited. In other words, when displaying on the display device, for example, situations such as image data being divided or feature data being missing can be avoided.

[0041] Furthermore, since the lack of feature data in the image data generated by the image processing unit 153 is suppressed, the user can visually determine the presence or absence of anomalies by looking at the image data displayed on the display device.

[0042] <Processing Space in Semiconductor Manufacturing Processes>

[0043] Next, the specified processing space for semiconductor manufacturing processes will be explained. Figure 2 This is a diagram illustrating an example of a semiconductor manufacturing process. (For example...) Figure 2 As shown, the semiconductor manufacturing process 200 has multiple chambers as an example of a processing space. Figure 2 In the example, semiconductor manufacturing process 200 has three chambers, reference numeral 121 (name = "chamber A") to reference numeral 123 (name = "chamber C"), in which the pre-processed wafer 110 is processed.

[0044] Furthermore, in the semiconductor manufacturing process 200, the aforementioned light emission spectrophotometer 140 is provided in each chamber, and OES data is measured in each chamber. However, for the sake of simplicity, the following description will focus on the case where OES data measured in one chamber is displayed. Additionally, in the following description, it will be assumed that this one chamber is chamber A. Furthermore, in the following description, it will be assumed that chamber A is, for example, a chamber of an etching apparatus.

[0045] <Hardware Structure of Data Processing Device>

[0046] Next, the hardware structure of the data processing device 150 will be described. Figure 3 This is a diagram illustrating an example of the hardware structure of a data processing device. For example... Figure 3 As shown, the data processing device 150 includes a CPU (Central Processing Unit) 301, a ROM (Read Only Memory) 302, and a RAM (Random Access Memory) 303. Additionally, the data processing device 150 includes a GPU (Graphics Processing Unit) 304. Furthermore, the processors (processing circuits) such as the CPU 301 and GPU 304, and the memories such as the ROM 302 and RAM 303, form what is known as a computer.

[0047] Furthermore, the data processing device 150 includes: an auxiliary storage device 305, a display device 306, an operation device 307, an I / F (Interface) device 308, and a drive device 309. In addition, the various hardware components of the data processing device 150 are interconnected via a bus 310.

[0048] CPU 301 is a computing device that executes various programs (e.g., data processing programs) installed in auxiliary storage device 305.

[0049] ROM 302 is a non-volatile memory that functions as the main storage device. ROM 302 stores various programs and data required by CPU 301 to execute various programs installed on auxiliary storage device 305. Specifically, ROM 302 stores boot programs such as BIOS (Basic Input / Output System) and EFI (Extensible Firmware Interface).

[0050] RAM303 is a volatile memory such as DRAM (Dynamic Random Access Memory) or SRAM (Static Random Access Memory), which functions as the main storage device. RAM303 provides a working area for the CPU301 to execute various programs installed on the auxiliary storage device 305.

[0051] GPU 304 is a computing device for image processing. In this embodiment, when the CPU 301 executes the data processing program, it performs high-speed parallel processing on the OES data. Furthermore, GPU 304 is equipped with internal memory (GPU memory) to temporarily hold the information required for parallel processing of the OES data.

[0052] The auxiliary storage device 305 stores various programs and data used by the CPU 301 when executing these programs. For example, the auxiliary storage device 305 implements the OES data storage unit 155 and the image data storage unit 156.

[0053] Display device 306 is, for example, a display device that displays image data generated by image processing unit 153. Operation device 307 is an input device used by the user of data processing device 150 to input various instructions to data processing device 150. I / F device 308 is a connection device for connecting to a network (not shown) and for transmitting and receiving data with other devices (e.g., light emission spectrometer).

[0054] The drive unit 309 is a device for setting the recording medium 320. The recording medium 320 referred to here includes media that record information optically, electrically, or magnetically, such as CD-ROMs, floppy disks, and optical discs. Alternatively, the recording medium 320 may also include semiconductor memories that record information electrically, such as ROMs and flash memory.

[0055] Furthermore, various programs installed on the auxiliary storage device 305 can be set on the drive device 309 via the allocated recording medium 320, and the various programs recorded on the recording medium 320 can be read and installed by the drive device 309. Alternatively, various programs installed on the auxiliary storage device 305 can also be downloaded and installed via a network (not shown).

[0056] <Specific examples of OES data>

[0057] Next, a specific example of OES data measured by the luminescence spectrophotometer 140 and stored in the OES data storage unit 155 of the data processing unit 150 will be described. Figure 4 This is a diagram representing an example of OES data. (Example:) Figure 4As shown, OES data 410 consists of sets of luminescence intensity data for each time period within the wavelength range (200 nm to 800 nm) of visible light measured every 0.5 nm. Furthermore, in OES data 410, the horizontal axis represents time, and the vertical axis represents the luminescence intensity at each wavelength.

[0058] exist Figure 4 In the case of OES data 410, for example, the top chart shows the luminescence intensity data at various times for wavelength = 200 [nm], the second chart shows the luminescence intensity data at various times for wavelength = 200.5 [nm], and the third chart shows the luminescence intensity data at various times for wavelength = 201 [nm].

[0059] Furthermore, the duration of OES data 410 may be, for example, the processing time (one wafer) during which the wafer 110 is processed in chamber A before processing. Alternatively, the duration of OES data 410 may also be, for example, the processing time of a portion of the processing steps when the wafer 110 is processed in chamber A under multiple processing steps before processing. Or, the duration of OES data 410 may also be, for example, the processing time of a portion of the processing steps performed with a portion of the recipes when the wafer 110 is processed with a portion of the processing steps with multiple recipes in chamber A before processing.

[0060] On the other hand, Figure 4 In the OES data 420, the horizontal axis represents wavelength and the vertical axis represents time, with the luminous intensity data arranged at each point (each time, each wavelength). In the case of OES data 420, 1201 points of luminous intensity data are arranged along the horizontal axis. Furthermore, in the case of OES data 420, if the sampling period for the luminous intensity data is 0.1 seconds, and the processing time (for one wafer) of the wafer 110 in chamber A before processing is 300 seconds, then 3000 points of luminous intensity data are arranged along the vertical axis.

[0061] Therefore, if we want to convert the luminous intensity data of each point of OES data 420 into color data and display the generated image data on display device 306, we need a display device with 1201 pixels horizontally and 3000 pixels vertically.

[0062] On the other hand, when the number of pixels of the display device 306 is limited, the OES data 420 is displayed in an interleaved manner. In this case, there may be a lack of characteristic data indicating the occurrence of an abnormality. Therefore, in the data processing device 150 according to the present embodiment, as described above, after preprocessing the OES data, it is compressed, and the compressed OES data is imaged to generate image data. Thus, regardless of whether the number of pixels of the display device 306 is limited, it is possible to avoid the situation where the image data is displayed in an interleaved manner and lacks characteristic data when displayed on the display device 306.

[0063] <Specific Example of Processing of Preprocessing Unit>

[0064] Next, a specific example of the processing of the OES data 420 by the preprocessing unit 151 of the data processing device 150 will be described. Figure 5 It is a diagram showing a specific example of the processing of the preprocessing unit. As Figure 5 shown, the preprocessing unit 151 includes a normalization processing unit 510.

[0065] The normalization processing unit 510 normalizes the light emission intensity data of each point by reading out the OES data 420 stored in the OES data storage unit 155 and dividing the light emission intensity data of each point included in the OES data 420 by the reference data.

[0066] In addition, it is assumed that the reference data uses, for example, the average value of the light emission intensity data of each wavelength of the OES data measured when processing a wafer used as a reference (a wafer determined to be a qualified product). In this case, the difference from the wafer determined to be a qualified product can be made significant, so it is easy to determine the presence or absence of an abnormality when generating image data.

[0067] Alternatively, the reference data may also use the average value of the light emission intensity data of the wavelength used as a reference in the measured OES data. In this case, in the processing of the pre-processed wafer, it is possible to make the relative strengthening of the light emission intensity of which wavelength significant, so it is easy to determine the presence or absence of an abnormality when generating image data.

[0068] In Figure 5 , the preprocessed OES data 500 is an example of the preprocessed OES data obtained by the normalization processing unit 510 normalizing the light emission intensity data of each point. In addition, in Figure 5 , the normalized data 501 of each point represents the normalized data of each point included in a region of a specified size (3 points in the horizontal axis direction and 3 points in the vertical axis direction) within the preprocessed OES data 500.

[0069] As shown in the normalized data 501 for each point, by dividing the luminous intensity data of each point using the reference data, the value of the normalized data becomes approximately close to "1.0".

[0070] <Specific examples of compression section processing>

[0071] Next, a specific example of how the compression unit 152 of the data processing device 150 processes the preprocessed OES data 500 will be described. Figure 6 This is a diagram illustrating a specific example of the compression process. For example... Figure 6 As shown, the compression unit 152 has an average value calculation unit 610 and a representative value extraction unit 620.

[0072] The average calculation unit 610 divides the normalized data of each point contained in the preprocessed OES data 500 into multiple regions of each specified size (e.g., 3 points in the horizontal direction and 3 points in the vertical direction), and calculates the average value of the normalized data of each point in each region.

[0073] The representative value extraction unit 620 compares the difference between the average value calculated in each region and the normalized data (e.g., 9 points of normalized data) contained in each region, and extracts the value of the normalized data with the largest difference (i.e., the outlier) as the representative value of each region.

[0074] Figure 6 The example shows the average value calculated by the average value calculation unit 610, which calculates the average value of 9 points of normalized data 501 contained in a region of a specified size of "0.99". Additionally, Figure 6 The example illustrates the case where the representative value extraction unit 620 extracts the value of the normalized data (outlier = "1.5") that has the largest difference from the mean ("0.99") from the normalized data 501 of 9 points contained in a region of a specified size as the representative value 601.

[0075] In this way, by extracting representative values ​​for each region of a specified size (3 points on the horizontal axis and 3 points on the vertical axis), the number of points in the normalized data contained in the preprocessed OES data 500 is compressed to 1 / 3 on the horizontal axis and 1 / 3 on the vertical axis. The result is as follows: Figure 6 As shown,

[0076] • After preprocessing, the OES data 500, which consists of 1201 points along the horizontal axis and 3000 points along the vertical axis, is compressed to...

[0077] • The compressed OES data, consisting of 400 points on the horizontal axis and 1000 points on the vertical axis, can reduce management costs.

[0078] <Specific Examples of Image Data>

[0079] Next, a specific example of image data generated by the imaging unit 153 of the data processing device 150 will be described. Figure 7 It is a diagram representing a specific example of image data.

[0080] exist Figure 7 In the image data 600', the image data represents a specific example of two-dimensional image data generated by converting the compressed data of each point of the compressed OES data 600 notified from the compression unit 152 into color data. In the image data 600', the horizontal axis represents wavelength, and the vertical axis represents time. Furthermore, different colors at each point represent different values ​​of the compressed data. Figure 7 The example shows how to convert the point with the largest value in the compressed data to red, the point with the average value to green, and the point with the smallest value to blue. Alternatively, it shows how to convert points with values ​​between the largest and average values ​​to colors between red and green on the color wheel, and points with values ​​between the average and smallest values ​​to colors between green and blue on the color wheel. However, the assignment of colors relative to the values ​​in the compressed data is arbitrary, and other assignment methods can also be used.

[0081] In the case of image data 600', the lack of feature data is suppressed. Therefore, the user can visually determine the presence or absence of anomalies by looking at the image data 600' displayed on the display device 306.

[0082] In addition, such as Figure 7 As shown, in the image processing unit 153, when a specified area 701 of the specified image data 600' is designated, the image data of the specified area 701 can also be magnified to display magnified image data 702. Therefore, the user can both view the image data 600' from above and magnify to view a specific wavelength range and a specific time range.

[0083] In addition, although Figure 7 Although not shown in the diagram, the compressed OES data 600 can also be displayed as three-dimensional image data. For example, the compressed data can be displayed in three dimensions by taking wavelength on the horizontal axis, time on the depth axis, and the value of the compressed data on the height axis. In addition, in this case, different values ​​of the compressed data can also be represented by different colors.

[0084] In addition, although Figure 7 Although not shown in the image data, the excited species (excited-state molecules) corresponding to the wavelength on the horizontal axis can also be displayed in the image data 600'. This allows the user to infer the cause of the anomaly.

[0085] <Flowchart of Image Processing in Data Processing Device>

[0086] Next, the image processing of OES data from the data processing unit 150 will be explained. Figure 8 It is a flowchart representing the process of image processing.

[0087] In step S801, the data processing device 150 acquires OES data from the light emission spectrophotometer 140 and stores it in the OES data storage unit 155.

[0088] In step S802, the preprocessing unit 151 of the data processing device 150 reads OES data from the OES data storage unit 155 and divides the luminous intensity data of each point by the reference data to perform preprocessing.

[0089] In step S803, the compression unit 152 of the data processing apparatus 150 divides the normalized data of each point contained in the preprocessed OES data into multiple regions of each specified size. Furthermore, the compression unit 152 of the data processing apparatus 150 compresses the preprocessed OES data by extracting the value of the normalized data (outlier) that has the largest difference from the average value calculated for each region of the specified size, using this value as a representative value for each region of the specified size.

[0090] In step S804, the image processing unit 153 of the data processing device 150 generates image data by converting the compressed data of each point of the compressed OES data into color data.

[0091] In step S805, the image processing unit 153 of the data processing device 150 displays image data of the compressed OES data.

[0092] <Summary>

[0093] From the above explanation, it can be seen that...

[0094] The data processing apparatus 150 according to the first embodiment has a preprocessing unit that normalizes the luminescence intensity data of each point of the OES data by dividing it using predetermined reference data, thereby generating normalized data.

[0095] • It has a compression unit that divides the preprocessed OES data into multiple regions according to each specified time range and a specified wavelength range, and extracts the value of the normalized data (outlier) that has the largest difference from the average value of each region as the representative value of each region.

[0096] It has a generation unit that converts representative values ​​of each region into color data to generate image data.

[0097] In this way, by normalizing the OES data when it is visualized, and by using a method of extracting outliers for each region of a specified size for compression, the OES data can be compressed without losing the feature data contained in the OES data.

[0098] In other words, according to the first embodiment, a data processing apparatus, a data processing system, a data processing method, and a data processing program are provided that can compress and visualize OES data to suppress the lack of feature data.

[0099] [Second Implementation]

[0100] In the first embodiment described above, a structure was explained that compresses and visualizes OES data to suppress the lack of feature data, thereby reducing the management cost of OES data and enabling users to determine the presence or absence of anomalies. In contrast, in the second embodiment, a structure is explained that automatically determines the presence or absence of anomalies using visualized OES data. Furthermore, the structure for automatically determining the presence or absence of anomalies using visualized OES data includes, for example:

[0101] • Determine whether the corresponding wafer is a qualified product that has undergone normal processing or includes an abnormal wafer structure;

[0102] • Determine whether the corresponding wafer is a qualified product that has undergone normal processing or a structure that is not a qualified product that has undergone normal processing;

[0103] • It determines whether the corresponding wafer is a qualified product that has undergone normal processing or any structure, including a wafer with an abnormal structure among multiple modes. In the second embodiment, the structure for automatically determining whether the corresponding wafer is a qualified product that has undergone normal processing or a wafer with an abnormal structure among multiple modes will be described. Hereinafter, the second embodiment will be described focusing on the differences from the first embodiment described above.

[0104] <System Architecture of Data Processing System>

[0105] First, use Figure 9A as well as Figure 9B The system architecture of the data processing system involved in the second embodiment will be described. Figure 9A The second diagram is an example of the system architecture of a data processing system. Specifically, Figure 9A This is an example of the system architecture of a data processing system 900 in the "learning phase," which learns the correspondence between image data generated by visualizing OES data and the corresponding recipes and the processing results of the processed wafers.

[0106] and Figure 1 The difference between the data processing system 100 shown and the data processing system 900 is that the data processing device 920 has a learning unit 921.

[0107] In the data processing system 900, the data processing device 920 acquires processing result information about the processed wafer 130. From the processing result information, it can be assumed that:

[0108] • This indicates whether the processed wafer 130 is a normal, qualified product or includes defective wafers.

[0109] • This indicates whether the processed wafer 130 is a qualified product that has undergone normal processing or not (at least not a qualified product that has undergone normal processing).

[0110] The information can be varied, such as indicating whether the processed wafer 130 is a qualified product that has undergone normal processing or includes information indicating which of the multiple modes it belongs to as an abnormal wafer. However, in this embodiment, the case of using information indicating whether the processed wafer 130 is a qualified product that has undergone normal processing or includes information indicating which of the multiple modes it belongs to as an abnormal wafer will be described. Furthermore, the aforementioned information that can be included in the processing result information can also be generated, for example, based on the output of the inspection device when inspecting the processed wafer, i.e., information indicating whether it is a qualified product or a non-qualified product.

[0111] In the data processing device 920, the acquired processing result information, along with the corresponding recipe and the corresponding image data, is stored as learning data in the learning data storage unit 923.

[0112] The learning unit 921 of the data processing device 920 has an anomaly detection model that takes image data and recipe as input and outputs the processing result information of the processed wafer.

[0113] Furthermore, the data processing device 920 reads learning data from the learning data storage unit 923 and performs learning processing on the anomaly detection model. Specifically, the data processing device 920 inputs the recipe and image data into the anomaly detection model and updates the model parameters of the anomaly detection model so that the output of the anomaly detection model is close to the corresponding processing result information.

[0114] on the other hand, Figure 9B The third diagram is an example of a system architecture for a data processing system. Specifically, Figure 9B This illustrates an example of the system architecture of the data processing system 900' in the "inference stage," which infers the processing result information of the processed wafer 130 based on image data generated from the visualization of OES data and the corresponding recipe.

[0115] and Figure 1 The difference between the data processing system 100 shown is that, in the case of the data processing system 900', the data processing device 920 has an inference unit 922.

[0116] The inference unit 922 of the data processing apparatus 920 has a learned anomaly detection model generated by learning processing performed by the learning unit 921. The inference unit 922 inputs image data and recipe into the learned anomaly detection model to infer the corresponding processing result information of the processed wafer 130, and outputs it as the inference result. Furthermore, the inference result output from the inference unit 922 can include:

[0117] • This indicates whether the processed wafer 130 is a normal, qualified product or includes defective wafers.

[0118] • This indicates whether the processed wafer 130 is a qualified product that has undergone normal processing or a non-qualified product (at least not a qualified product that has undergone normal processing).

[0119] • Various variations may be used to indicate whether the processed wafer 130 is a qualified product that has undergone normal processing or includes information about which of the multiple modes the abnormal wafer belongs to. However, in this embodiment, the case where the output indicates whether the processed wafer 130 is a qualified product that has undergone normal processing or includes information about which of the multiple modes the abnormal wafer belongs to is explained.

[0120] <Specific Examples of Learning to Use Data>

[0121] Next, a specific example of the learning data stored in the learning data storage unit 923 will be explained. For example... Figure 10 As shown, the learning data 1000 includes items such as "device", "formula", "wafer", "image data", and "good / including abnormalities" as information.

[0122] The name of the chamber in which the pre-processed wafer 110 is processed is stored in the "Apparatus". The recipe identifier that identifies the recipe used when the pre-processed wafer 110 is processed in chamber A is stored in the "Recipe".

[0123] The wafer identifier that identifies the pre-processing wafer processed in chamber A is stored in the "Wafer" field. The image data identifier that identifies the image data generated by the imaging unit 153 by imaging the OES data obtained from the corresponding pre-processing wafer measured during processing in chamber A is stored in the "Image Data" field.

[0124] Furthermore, the image data used for learning is preprocessed based on the average value of the luminescence intensity data of each wavelength of the OES data measured when the wafer to be judged as qualified is processed. Additionally, the image data used for learning is image data of the processing time (one wafer) of the wafer 110 being processed in chamber A before processing.

[0125] The "Good / Including Abnormal" field stores the processing result information of the pre-processed wafer 130 generated from the processing of the pre-processed wafer in chamber A. Specifically, it stores information indicating whether the post-processed wafer 130 is a qualified product that has undergone normal processing or includes an abnormal wafer belonging to one of multiple modes.

[0126] Figure 10 The example shows how “Formula 1” is used in “Cavity A” to process “Wafer 1” to “Wafer 3” respectively to generate “Image Data 1” to “Image Data 3”, and output the processing result information (“Good”).

[0127] in addition, Figure 10 The example shows how “Recipe 1” is used in “Cavity A” to process “Wafer 4” to generate “Image Data 4” and output the processing result information (“Including anomalies (pattern a)”).

[0128] in addition, Figure 10 The example shows how “Recipe 2” is used in “Cavity A” to process “Wafer 5, 7, 8” to generate “Image Data 5, 7, 8” and output the processing result information (“Good”).

[0129] in addition, Figure 10 The example shows how “Recipe 2” is used in “Cavity A” to process “Wafer 6, 9” to generate “Image Data 6, 9” and output the processing result information (“Including anomalies (mode b)” or “Including anomalies (mode c)”).

[0130] Furthermore, it includes anomaly patterns (patterns a, b, c, ..., etc.) categorized, for example, by each anomaly cause in the excited state. Anomaly causes include information such as which molecule is in the excited state, or which component within the chamber is abnormal. In other words, based on the learned anomaly detection model trained using 1000 training data points, the anomaly causes can be inferred simultaneously by inferring anomaly patterns.

[0131] <Specific examples of how the Learning Department handles this>

[0132] Next, a specific example of the learning process of the learning unit 921 of the data processing device 920 will be explained. Figure 11 This is a diagram illustrating a specific example of the learning process handled by the learning department. For example... Figure 11As shown, the learning unit 921 has an anomaly detection model 1101 and a comparison / change unit 1102.

[0133] The learning unit 921 reads the recipe (e.g., the recipe determined by the recipe identifier = "recipe 1") and the image data (e.g., the image data determined by the image data identifier = "image data 1") from the learning data 1000.

[0134] In addition, the learning unit 921 executes the anomaly detection model 1101 by inputting the recipe and image data read from the learning data 1000 into the anomaly detection model 1101, and outputs the probability distribution of the processing result information.

[0135] The probability distribution of the processing result information output from the anomaly detection model 1101 is input to the comparison / modification unit 1102 and compared with the probability distribution of the processing result information read from the learning data 1000 as "Good / Including Anomalies". For example, if "Good" is read from the learning data 1000 as "Good / Including Anomalies", the probability distribution of the processing result information output from the anomaly detection model 1101 is compared with the probability distribution set to "Good = 100%, Other Processing Result Information = 0%".

[0136] The comparison / modification unit 1102 updates the model parameters of the anomaly detection model 1101 based on the comparison results. As a result, the learning unit 921 can update the model parameters of the anomaly detection model 1101 so that the output when the recipe and image data are used as inputs is close to the "good / including anomaly" processing result information stored in the learning data 1000.

[0137] <Specific examples of inference processing in the inference department>

[0138] Next, a specific example of the inference processing of the inference unit 922 of the data processing device 920 will be described. Figure 12 This is a diagram illustrating a specific example of the inference process in the inference unit. For example... Figure 12 As shown, the inference unit 922 has a learned anomaly detection model 1201 (a learned anomaly detection model generated by learning the anomaly detection model 1101) and an output unit 1202.

[0139] The inference unit 922 acquires the formula used when the pre-processed wafer 110 is processed in chamber A, and the image data generated based on the OES data obtained from the pre-processed wafer 110 during processing in chamber A, and inputs it into the learned anomaly detection model 1201.

[0140] If the formula and image data are input into the inference unit 922, the probability distribution of the processing result information output by the anomaly detection model 1201 after learning is completed.

[0141] If the probability distribution of the processing result information is output from the learned anomaly detection model 1201, the output unit 1202 outputs the processing result information corresponding to the largest probability distribution among the probability distributions above a predetermined threshold. For example, if the probability distribution of "good" is above the predetermined threshold and is the largest, the output unit 1202 outputs information indicating that the processed wafer 130 is a qualified product that has undergone normal processing as an inference result.

[0142] Furthermore, for example, if the probability distribution of "including anomalies (mode a)" is above a predetermined threshold and is at its maximum, the output unit 1202 outputs information indicating that the processed wafer 130 contains anomalies belonging to mode a, as well as information indicating the cause of the anomalies, as an inference result. Moreover, the output unit 1202 can also be configured to output an optimal formula in addition to the inference result.

[0143] <Anomaly Detection and Handling Process>

[0144] Next, the process of anomaly detection and handling in data processing systems 900 and 900' will be explained. Figure 13 This is a flowchart illustrating the anomaly detection and handling process. Furthermore, Figure 13 The processes shown in steps S801 to S804 of each process are similar to those shown in the diagram. Figure 8 The processes shown in steps S801 to S804 are the same, so the description is omitted here.

[0145] In step S1301, the data processing device 920 determines whether the current stage is a learning stage or an inference stage. If it is determined to be a learning stage in step S1301 (if it is in step S1301), the process proceeds to step S1302.

[0146] In step S1302, the data processing device 920 acquires the processing result information and the corresponding formula.

[0147] In step S1303, the data processing device 920 establishes a correspondence between the acquired processing result information, recipe and image data, generates learning data, and stores it in the learning data storage unit 923.

[0148] In step S1304, the learning unit 921 of the data processing device 920 uses learning data to learn the anomaly detection model, and after the anomaly detection model has been generated, the anomaly detection process ends.

[0149] On the other hand, in step S1301, if it is determined to be the inference stage (if not in step S1301), proceed to step S1305.

[0150] In step S1305, the inference unit 922 of the data processing device 920 inputs the image data and recipe into the learned anomaly detection model 1201 and outputs the probability distribution of the processing result information.

[0151] In step S1305, the inference unit 922 of the data processing device 920 sends the inference result output from the output unit 1202 to the semiconductor manufacturing process based on the probability distribution of the processing result information output from the learned anomaly detection model 1201.

[0152] <Summary>

[0153] From the above explanation, it can be seen that...

[0154] • The data processing apparatus 920 according to the second embodiment has an anomaly detection model that learns the correspondence between image data generated by visualizing OES data and the recipe and the processing result information of the processed wafer.

[0155] • It has a learned anomaly detection model generated by learning the anomaly detection model using learning data.

[0156] • By inputting the image data generated from the visualization of OES data and the recipe into the learned anomaly detection model, the processing result information of the processed wafer can be inferred.

[0157] In this way, by using image data to infer the processing result information of the processed wafer, the presence or absence of anomalies can be automatically determined according to the second embodiment.

[0158] [Third Implementation Method]

[0159] In the second embodiment described above, a case was explained in which a data processing device is set up for each processing space, and a learned anomaly detection model is generated using the learning data generated by each data processing device. In contrast, in the third embodiment, the server device collects the learning data generated by each data processing device and performs learning processing.

[0160] Therefore, according to the third embodiment, it is possible to provide a learning completion anomaly detection model generated by learning processing using more learning data.

[0161] Hereinafter, the third embodiment will be described focusing on its differences from the first and second embodiments described above.

[0162] <System Architecture of Data Processing System>

[0163] First, the system structure of the data processing system involved in the third embodiment will be described. Figure 14 The fourth figure is an example of the system architecture of a data processing system.

[0164] like Figure 14 As shown, the data processing system 1400 includes: multiple semiconductor manufacturing processes, a light emission spectrophotometer 140 corresponding to each semiconductor manufacturing process, a data processing unit 1401, a collaboration unit 1411, and a server unit 1420.

[0165] The description of the luminescence spectrophotometer 140 has already been completed, so it will be omitted here.

[0166] The data processing apparatus 1401 includes a preprocessing unit 151, a compression unit 152, an image processing unit 153, and an inference unit 922. Furthermore, the preprocessing unit 151, compression unit 152, image processing unit 153, and inference unit 922 of the data processing apparatus 1401 are used... Figure 9A , Figure 9B The explanation has already been completed, so it will be omitted here.

[0167] Furthermore, the data processing apparatus 1401 includes an OES data storage unit 155 that stores OES data measured by the luminescence spectrophotometer 140, and an image data storage unit 156 that stores image data generated by the image processing unit 153. The data processing apparatus 1401 also includes a learning data storage unit 923, which stores image data, processing result information, and recipe information as learning data. Since the OES data storage unit 155, image data storage unit 156, and learning data storage unit 923 have already been described, their descriptions are omitted here.

[0168] The collaboration unit 1411 reads the learning data (e.g., learning data 1) stored in the learning data storage unit 923 and sends it to the server device 1420. Furthermore, since the image data included in the learning data transmitted between the collaboration unit 1411 and the server device 1420 is compressed, the amount of communication can be reduced compared to the case of transmitting and receiving OES data itself.

[0169] Furthermore, the collaboration unit 1411 sends the learning data to the server device 1420, and correspondingly, the learning completion anomaly detection model obtained from the server device 1420 is applied to the inference unit 922. Thus, the inference unit 922 can use the learning completion anomaly detection model shared by various semiconductor manufacturing processes generated by the server device 1420 to make inferences.

[0170] When the server device 1420 sends learning data from each cooperating unit 1411 corresponding to each semiconductor manufacturing process, it stores the learning data in the learning data storage unit 1422. Additionally, the server device 1420 includes a learning unit 921.

[0171] The learning unit 921 uses the learning data stored in the learning data storage unit 1422 to learn the anomaly detection model it possesses. Thus, the learning unit 921 can generate a common, learned anomaly detection model applicable to various semiconductor manufacturing processes.

[0172] Furthermore, in the server device 1420, the learning completion anomaly detection model generated by the learning unit 921 is sent to each cooperation unit 1411 corresponding to each semiconductor manufacturing process.

[0173] <Summary>

[0174] As can be seen from the above description, in the third embodiment, by separating the learning unit 921 and the inference unit 922, the learning data obtained in each semiconductor manufacturing process is collected, thereby generating a common learning completion anomaly detection model.

[0175] Therefore, according to the third embodiment, a learned anomaly detection model that does not depend on individual differences in semiconductor manufacturing processes can be generated.

[0176] [Fourth Implementation Method]

[0177] In the first to third embodiments described above, OES data was cited as an example of multi-wavelength time-series data measured during the processing of the pre-processing wafer 110 in the processing space of the semiconductor manufacturing process. However, the multi-wavelength time-series data measured during the processing of the pre-processing wafer 110 in the processing space of the semiconductor manufacturing process is not limited to OES data. For example, it may also be quality analysis data measured by a gas quality analysis device in the analysis chamber. Alternatively, it may be reflected light data obtained by measuring the reflected light when light is projected onto the wafer surface from an external light source in the chamber.

[0178] Furthermore, in the first embodiment described above, an example of a region with 3 points in the horizontal direction and 3 points in the vertical direction is given as the region of a specified size, but the region of a specified size is not limited to this. Additionally, the shape of the region of a specified size is not limited to a square, and the number of points in the horizontal direction and the number of points in the vertical direction may be different.

[0179] Furthermore, in the first embodiment described above, the image processing unit 153 is described as image processing the compressed OES data compressed by the compression unit 152 and storing it in the image data storage unit 156. However, in the data processing apparatus 150, in practice, it is sometimes necessary to manage the OES data before compression (preprocessed OES data) or the OES data before preprocessing (raw OES data) for a certain period of time. Therefore, in addition to the functions described in the first embodiment, the image processing unit 153 of the data processing apparatus 150 may also have functions such as image processing the raw OES data and image processing the preprocessed OES data. Furthermore, the image processing unit 153 may also have the function of performing image compression processing such as JPEG compression on the image data and storing it in the image data storage unit 156.

[0180] Furthermore, while the second and third embodiments described above illustrate inputting recipe and image data into the anomaly detection model, it is also possible to input only image data into the anomaly detection model. Alternatively, in addition to inputting recipe and image data, other data besides recipe and image data may also be input into the anomaly detection model.

[0181] Furthermore, in the third embodiment described above, the case where the server device 1420 collects learning data generated by the data processing device 1401 corresponding to multiple semiconductor manufacturing processes was explained. However, the learning data can also be exchanged between data processing devices via the server device 1420 (or directly). In this case, the cooperation unit 1411 adds the learning data (e.g., learning data 2) received from the cooperation unit 1411 of other semiconductor manufacturing processes to the learning data storage unit 923. Thus, the learning unit 921 can perform anomaly detection model learning processing using learning data (e.g., learning data 1) obtained from the corresponding semiconductor manufacturing process, in addition to the learning data (e.g., learning data 1) obtained from other semiconductor manufacturing processes. That is, the learning unit 921 can also learn and process phenomena that do not occur in the corresponding semiconductor manufacturing process. In addition, the anomaly detection model generated in this way can be exchanged between data processing devices via the server device 1420 (or directly), and thus can also be applied to other semiconductor manufacturing processes. In other words, the learned anomaly detection model generated in any semiconductor manufacturing process can also be horizontally deployed in other semiconductor manufacturing processes via server device 1420 (or directly).

[0182] Furthermore, in the third embodiment described above, the server device 1420 is described as having a learning unit 921, which provides the generated learning completion anomaly detection model to the data processing device 1401. However, an inference unit applying the generated learning completion anomaly detection model can also be configured in the server device 1420. In this case, if image data and a recipe are received from a cooperation unit 1411 corresponding to any semiconductor manufacturing process in the server device 1420, an inference result based on the received image data and recipe is sent to the corresponding cooperation unit 1411. That is, the generated learning completion anomaly detection model can also be executablely configured in the server device 1420, thereby sharing the learning completion anomaly detection model among various semiconductor manufacturing processes.

[0183] Furthermore, the present invention is not limited to the structures listed in the above embodiments, or the combinations thereof with other elements, as shown herein. Modifications can be made in these aspects without departing from the spirit of the invention, and can be appropriately determined according to its application.

[0184] This application asserts priority based on Japanese Patent Application No. 2020-122172, filed on July 16, 2020, the contents of which are incorporated herein by reference.

[0185] Explanation of reference numerals in the attached figures

[0186] 100…Data processing system; 110…Wafer before processing; 120…Processing space; 130…Wafer after processing; 140…Light emission spectrophotometer; 150…Data processing unit; 151…Preprocessing unit; 152…Compression unit; 153…Image unit; 420…OES data; 500…Preprocessed OES data; 510…Normalization processing unit; 600…Compressed OES data; 610…Average value calculation unit; 620…Representative value extraction unit; 500'…Image data; 600'…Image data; 910…Inspection device; 920…Data processing unit; 921…Learning unit; 922…Inference unit; 1000…Learning data; 1101…Anomaly detection model; 1201…Anomaly detection model after learning; 1400…Data processing system; 1401…Data processing unit; 1411…Collaboration unit; 1420…Server unit.

Claims

1. A data processing apparatus, comprising: The preprocessing unit generates normalized data by normalizing multi-wavelength time series data using specified reference data. The extraction unit divides the normalized data into multiple regions according to each specified time range and wavelength range, and extracts outliers in each region as representative values. as well as The generation unit converts the representative values ​​of the above regions into color data to generate image data. The extraction unit calculates the average value of the normalized data contained in each region, and extracts the outlier value with the largest difference from each average value from the normalized data contained in each region as the representative value of each region.

2. The data processing apparatus according to claim 1, wherein, The preprocessing unit uses the multi-wavelength time series data, or the time series data of the reference wavelength in the multi-wavelength time series data, which is processed as a reference substrate in the semiconductor manufacturing process, as the specified reference data to normalize the multi-wavelength time series data.

3. The data processing apparatus according to claim 1, wherein, The aforementioned multi-wavelength time series data are data measured by a light emission spectrophotometer or a quality analysis device during substrate processing in semiconductor manufacturing processes.

4. The data processing apparatus according to claim 3, wherein, It also has an inference unit that infers the processing result of the processed substrate by inputting new image data into the model. The model is obtained by learning the correspondence between the image data and the processing result of the processed substrate. The image data is generated by the generation unit and is based on the multi-wavelength time series data measured during the processing of the substrate.

5. The data processing apparatus according to claim 4, wherein, The aforementioned inference unit infers that the processed substrate contains an abnormal substrate, and the cause of the abnormality of the processed substrate, as the processing result.

6. The data processing apparatus according to claim 4, wherein, It also has a learning unit that performs learning processing on the above model so that when image data generated by the above generation unit and image data generated based on the above multi-wavelength time series data measured in the processing of the above substrate are input into the above model, the output is close to the processing result of the above substrate after processing.

7. A data processing system, comprising: The plurality of data processing devices as claimed in claim 4; and The server device is connected to the plurality of data processing devices and receives from the plurality of data processing devices at least one of the following: image data generated based on the time series data of the multi-wavelengths measured in the processing of the substrate, or a model obtained by learning the correspondence between the image data and the processing result of the processed substrate.

8. A data processing system, comprising: The data processing apparatus of claim 4; and The server device has a learning unit that performs learning processing on the model so that when image data generated by the generation unit and image data generated based on the multi-wavelength time series data measured during the processing of the substrate are input into the model, the output is close to the processing result of the substrate after processing.

9. A data processing method, comprising: The preprocessing step involves normalizing multi-wavelength time series data using specified reference data to generate normalized data. The extraction process involves dividing the normalized data into multiple regions according to each specified time range and wavelength range, and extracting outliers from each region as representative values. as well as The generation process converts the representative values ​​of each of the above regions into color data, thereby generating image data. The above extraction process calculates the average value of the normalized data contained in each region, and extracts the outlier value with the largest difference from each average value from the normalized data contained in each region, which is used as the representative value of each region.

10. The data processing method according to claim 9, wherein, The aforementioned preprocessing step uses the aforementioned multi-wavelength time series data, or the time series data of the reference wavelength in the aforementioned multi-wavelength time series data, when processing the substrate as a reference in the semiconductor manufacturing process, as the aforementioned reference data, and normalizes the aforementioned multi-wavelength time series data.

11. The data processing method according to claim 9, wherein, The aforementioned multi-wavelength time series data are data measured by a light emission spectrophotometer or a quality analysis device during substrate processing in semiconductor manufacturing processes.

12. A data processing program product for causing a computer to perform: The preprocessing step involves normalizing the multi-wavelength time series data using specified reference data to generate normalized data. The extraction process involves dividing the normalized data into multiple regions according to each specified time range and wavelength range, and extracting outliers from each region as representative values. as well as The generation process converts the representative values ​​of each of the above regions into color data, thereby generating image data. The above extraction process calculates the average value of the normalized data contained in each region, and extracts the outlier value with the largest difference from each average value from the normalized data contained in each region as the representative value of each region.

13. The data processing program product according to claim 12, wherein, The aforementioned preprocessing step uses the aforementioned multi-wavelength time series data, or the time series data of the reference wavelength in the aforementioned multi-wavelength time series data, when processing the substrate as a reference in the semiconductor manufacturing process, as the aforementioned reference data, and normalizes the aforementioned multi-wavelength time series data.

14. The data processing program product according to claim 12, wherein, The aforementioned multi-wavelength time series data are data measured by a light emission spectrophotometer or a quality analysis device during substrate processing in semiconductor manufacturing processes.

Images